git-p4: introduce asciidoc documentation
[git/mjg.git] / sha1_file.c
blobf291f3f0f7d919c7d59b8b7dc3c3f2450a5a9f09
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"
23 #ifndef O_NOATIME
24 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
25 #define O_NOATIME 01000000
26 #else
27 #define O_NOATIME 0
28 #endif
29 #endif
31 #define SZ_FMT PRIuMAX
32 static inline uintmax_t sz_fmt(size_t s) { return s; }
34 const unsigned char null_sha1[20];
37 * This is meant to hold a *small* number of objects that you would
38 * want read_sha1_file() to be able to return, but yet you do not want
39 * to write them into the object store (e.g. a browse-only
40 * application).
42 static struct cached_object {
43 unsigned char sha1[20];
44 enum object_type type;
45 void *buf;
46 unsigned long size;
47 } *cached_objects;
48 static int cached_object_nr, cached_object_alloc;
50 static struct cached_object empty_tree = {
51 EMPTY_TREE_SHA1_BIN_LITERAL,
52 OBJ_TREE,
53 "",
57 static struct cached_object *find_cached_object(const unsigned char *sha1)
59 int i;
60 struct cached_object *co = cached_objects;
62 for (i = 0; i < cached_object_nr; i++, co++) {
63 if (!hashcmp(co->sha1, sha1))
64 return co;
66 if (!hashcmp(sha1, empty_tree.sha1))
67 return &empty_tree;
68 return NULL;
71 int mkdir_in_gitdir(const char *path)
73 if (mkdir(path, 0777)) {
74 int saved_errno = errno;
75 struct stat st;
76 struct strbuf sb = STRBUF_INIT;
78 if (errno != EEXIST)
79 return -1;
81 * Are we looking at a path in a symlinked worktree
82 * whose original repository does not yet have it?
83 * e.g. .git/rr-cache pointing at its original
84 * repository in which the user hasn't performed any
85 * conflict resolution yet?
87 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
88 strbuf_readlink(&sb, path, st.st_size) ||
89 !is_absolute_path(sb.buf) ||
90 mkdir(sb.buf, 0777)) {
91 strbuf_release(&sb);
92 errno = saved_errno;
93 return -1;
95 strbuf_release(&sb);
97 return adjust_shared_perm(path);
100 int safe_create_leading_directories(char *path)
102 char *pos = path + offset_1st_component(path);
103 struct stat st;
105 while (pos) {
106 pos = strchr(pos, '/');
107 if (!pos)
108 break;
109 while (*++pos == '/')
111 if (!*pos)
112 break;
113 *--pos = '\0';
114 if (!stat(path, &st)) {
115 /* path exists */
116 if (!S_ISDIR(st.st_mode)) {
117 *pos = '/';
118 return -3;
121 else if (mkdir(path, 0777)) {
122 *pos = '/';
123 return -1;
125 else if (adjust_shared_perm(path)) {
126 *pos = '/';
127 return -2;
129 *pos++ = '/';
131 return 0;
134 int safe_create_leading_directories_const(const char *path)
136 /* path points to cache entries, so xstrdup before messing with it */
137 char *buf = xstrdup(path);
138 int result = safe_create_leading_directories(buf);
139 free(buf);
140 return result;
143 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 int i;
146 for (i = 0; i < 20; i++) {
147 static char hex[] = "0123456789abcdef";
148 unsigned int val = sha1[i];
149 char *pos = pathbuf + i*2 + (i > 0);
150 *pos++ = hex[val >> 4];
151 *pos = hex[val & 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
158 * filename.
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1)
167 static char buf[PATH_MAX];
168 const char *objdir;
169 int len;
171 objdir = get_object_directory();
172 len = strlen(objdir);
174 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
175 if (len + 43 > PATH_MAX)
176 die("insanely long object directory %s", objdir);
177 memcpy(buf, objdir, len);
178 buf[len] = '/';
179 buf[len+3] = '/';
180 buf[len+42] = '\0';
181 fill_sha1_path(buf + len + 1, sha1);
182 return buf;
185 static char *sha1_get_pack_name(const unsigned char *sha1,
186 char **name, char **base, const char *which)
188 static const char hex[] = "0123456789abcdef";
189 char *buf;
190 int i;
192 if (!*base) {
193 const char *sha1_file_directory = get_object_directory();
194 int len = strlen(sha1_file_directory);
195 *base = xmalloc(len + 60);
196 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
197 sha1_file_directory, which);
198 *name = *base + len + 11;
201 buf = *name;
203 for (i = 0; i < 20; i++) {
204 unsigned int val = *sha1++;
205 *buf++ = hex[val >> 4];
206 *buf++ = hex[val & 0xf];
209 return *base;
212 char *sha1_pack_name(const unsigned char *sha1)
214 static char *name, *base;
216 return sha1_get_pack_name(sha1, &name, &base, "pack");
219 char *sha1_pack_index_name(const unsigned char *sha1)
221 static char *name, *base;
223 return sha1_get_pack_name(sha1, &name, &base, "idx");
226 struct alternate_object_database *alt_odb_list;
227 static struct alternate_object_database **alt_odb_tail;
229 static void read_info_alternates(const char * alternates, int depth);
230 static int git_open_noatime(const char *name);
233 * Prepare alternate object database registry.
235 * The variable alt_odb_list points at the list of struct
236 * alternate_object_database. The elements on this list come from
237 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
238 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
239 * whose contents is similar to that environment variable but can be
240 * LF separated. Its base points at a statically allocated buffer that
241 * contains "/the/directory/corresponding/to/.git/objects/...", while
242 * its name points just after the slash at the end of ".git/objects/"
243 * in the example above, and has enough space to hold 40-byte hex
244 * SHA1, an extra slash for the first level indirection, and the
245 * terminating NUL.
247 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
249 const char *objdir = get_object_directory();
250 struct alternate_object_database *ent;
251 struct alternate_object_database *alt;
252 int pfxlen, entlen;
253 struct strbuf pathbuf = STRBUF_INIT;
255 if (!is_absolute_path(entry) && relative_base) {
256 strbuf_addstr(&pathbuf, real_path(relative_base));
257 strbuf_addch(&pathbuf, '/');
259 strbuf_add(&pathbuf, entry, len);
261 normalize_path_copy(pathbuf.buf, pathbuf.buf);
263 pfxlen = strlen(pathbuf.buf);
266 * The trailing slash after the directory name is given by
267 * this function at the end. Remove duplicates.
269 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
270 pfxlen -= 1;
272 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
273 ent = xmalloc(sizeof(*ent) + entlen);
274 memcpy(ent->base, pathbuf.buf, pfxlen);
275 strbuf_release(&pathbuf);
277 ent->name = ent->base + pfxlen + 1;
278 ent->base[pfxlen + 3] = '/';
279 ent->base[pfxlen] = ent->base[entlen-1] = 0;
281 /* Detect cases where alternate disappeared */
282 if (!is_directory(ent->base)) {
283 error("object directory %s does not exist; "
284 "check .git/objects/info/alternates.",
285 ent->base);
286 free(ent);
287 return -1;
290 /* Prevent the common mistake of listing the same
291 * thing twice, or object directory itself.
293 for (alt = alt_odb_list; alt; alt = alt->next) {
294 if (!memcmp(ent->base, alt->base, pfxlen)) {
295 free(ent);
296 return -1;
299 if (!memcmp(ent->base, objdir, pfxlen)) {
300 free(ent);
301 return -1;
304 /* add the alternate entry */
305 *alt_odb_tail = ent;
306 alt_odb_tail = &(ent->next);
307 ent->next = NULL;
309 /* recursively add alternates */
310 read_info_alternates(ent->base, depth + 1);
312 ent->base[pfxlen] = '/';
314 return 0;
317 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
318 const char *relative_base, int depth)
320 const char *cp, *last;
322 if (depth > 5) {
323 error("%s: ignoring alternate object stores, nesting too deep.",
324 relative_base);
325 return;
328 last = alt;
329 while (last < ep) {
330 cp = last;
331 if (cp < ep && *cp == '#') {
332 while (cp < ep && *cp != sep)
333 cp++;
334 last = cp + 1;
335 continue;
337 while (cp < ep && *cp != sep)
338 cp++;
339 if (last != cp) {
340 if (!is_absolute_path(last) && depth) {
341 error("%s: ignoring relative alternate object store %s",
342 relative_base, last);
343 } else {
344 link_alt_odb_entry(last, cp - last,
345 relative_base, depth);
348 while (cp < ep && *cp == sep)
349 cp++;
350 last = cp;
354 static void read_info_alternates(const char * relative_base, int depth)
356 char *map;
357 size_t mapsz;
358 struct stat st;
359 const char alt_file_name[] = "info/alternates";
360 /* Given that relative_base is no longer than PATH_MAX,
361 ensure that "path" has enough space to append "/", the
362 file name, "info/alternates", and a trailing NUL. */
363 char path[PATH_MAX + 1 + sizeof alt_file_name];
364 int fd;
366 sprintf(path, "%s/%s", relative_base, alt_file_name);
367 fd = git_open_noatime(path);
368 if (fd < 0)
369 return;
370 if (fstat(fd, &st) || (st.st_size == 0)) {
371 close(fd);
372 return;
374 mapsz = xsize_t(st.st_size);
375 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
376 close(fd);
378 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
380 munmap(map, mapsz);
383 void add_to_alternates_file(const char *reference)
385 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
386 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
387 char *alt = mkpath("%s\n", reference);
388 write_or_die(fd, alt, strlen(alt));
389 if (commit_lock_file(lock))
390 die("could not close alternates file");
391 if (alt_odb_tail)
392 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
395 void foreach_alt_odb(alt_odb_fn fn, void *cb)
397 struct alternate_object_database *ent;
399 prepare_alt_odb();
400 for (ent = alt_odb_list; ent; ent = ent->next)
401 if (fn(ent, cb))
402 return;
405 void prepare_alt_odb(void)
407 const char *alt;
409 if (alt_odb_tail)
410 return;
412 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
413 if (!alt) alt = "";
415 alt_odb_tail = &alt_odb_list;
416 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
418 read_info_alternates(get_object_directory(), 0);
421 static int has_loose_object_local(const unsigned char *sha1)
423 char *name = sha1_file_name(sha1);
424 return !access(name, F_OK);
427 int has_loose_object_nonlocal(const unsigned char *sha1)
429 struct alternate_object_database *alt;
430 prepare_alt_odb();
431 for (alt = alt_odb_list; alt; alt = alt->next) {
432 fill_sha1_path(alt->name, sha1);
433 if (!access(alt->base, F_OK))
434 return 1;
436 return 0;
439 static int has_loose_object(const unsigned char *sha1)
441 return has_loose_object_local(sha1) ||
442 has_loose_object_nonlocal(sha1);
445 static unsigned int pack_used_ctr;
446 static unsigned int pack_mmap_calls;
447 static unsigned int peak_pack_open_windows;
448 static unsigned int pack_open_windows;
449 static unsigned int pack_open_fds;
450 static unsigned int pack_max_fds;
451 static size_t peak_pack_mapped;
452 static size_t pack_mapped;
453 struct packed_git *packed_git;
455 void pack_report(void)
457 fprintf(stderr,
458 "pack_report: getpagesize() = %10" SZ_FMT "\n"
459 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
460 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
461 sz_fmt(getpagesize()),
462 sz_fmt(packed_git_window_size),
463 sz_fmt(packed_git_limit));
464 fprintf(stderr,
465 "pack_report: pack_used_ctr = %10u\n"
466 "pack_report: pack_mmap_calls = %10u\n"
467 "pack_report: pack_open_windows = %10u / %10u\n"
468 "pack_report: pack_mapped = "
469 "%10" SZ_FMT " / %10" SZ_FMT "\n",
470 pack_used_ctr,
471 pack_mmap_calls,
472 pack_open_windows, peak_pack_open_windows,
473 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
476 static int check_packed_git_idx(const char *path, struct packed_git *p)
478 void *idx_map;
479 struct pack_idx_header *hdr;
480 size_t idx_size;
481 uint32_t version, nr, i, *index;
482 int fd = git_open_noatime(path);
483 struct stat st;
485 if (fd < 0)
486 return -1;
487 if (fstat(fd, &st)) {
488 close(fd);
489 return -1;
491 idx_size = xsize_t(st.st_size);
492 if (idx_size < 4 * 256 + 20 + 20) {
493 close(fd);
494 return error("index file %s is too small", path);
496 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
497 close(fd);
499 hdr = idx_map;
500 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
501 version = ntohl(hdr->idx_version);
502 if (version < 2 || version > 2) {
503 munmap(idx_map, idx_size);
504 return error("index file %s is version %"PRIu32
505 " and is not supported by this binary"
506 " (try upgrading GIT to a newer version)",
507 path, version);
509 } else
510 version = 1;
512 nr = 0;
513 index = idx_map;
514 if (version > 1)
515 index += 2; /* skip index header */
516 for (i = 0; i < 256; i++) {
517 uint32_t n = ntohl(index[i]);
518 if (n < nr) {
519 munmap(idx_map, idx_size);
520 return error("non-monotonic index %s", path);
522 nr = n;
525 if (version == 1) {
527 * Total size:
528 * - 256 index entries 4 bytes each
529 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
530 * - 20-byte SHA1 of the packfile
531 * - 20-byte SHA1 file checksum
533 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
534 munmap(idx_map, idx_size);
535 return error("wrong index v1 file size in %s", path);
537 } else if (version == 2) {
539 * Minimum size:
540 * - 8 bytes of header
541 * - 256 index entries 4 bytes each
542 * - 20-byte sha1 entry * nr
543 * - 4-byte crc entry * nr
544 * - 4-byte offset entry * nr
545 * - 20-byte SHA1 of the packfile
546 * - 20-byte SHA1 file checksum
547 * And after the 4-byte offset table might be a
548 * variable sized table containing 8-byte entries
549 * for offsets larger than 2^31.
551 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
552 unsigned long max_size = min_size;
553 if (nr)
554 max_size += (nr - 1)*8;
555 if (idx_size < min_size || idx_size > max_size) {
556 munmap(idx_map, idx_size);
557 return error("wrong index v2 file size in %s", path);
559 if (idx_size != min_size &&
561 * make sure we can deal with large pack offsets.
562 * 31-bit signed offset won't be enough, neither
563 * 32-bit unsigned one will be.
565 (sizeof(off_t) <= 4)) {
566 munmap(idx_map, idx_size);
567 return error("pack too large for current definition of off_t in %s", path);
571 p->index_version = version;
572 p->index_data = idx_map;
573 p->index_size = idx_size;
574 p->num_objects = nr;
575 return 0;
578 int open_pack_index(struct packed_git *p)
580 char *idx_name;
581 int ret;
583 if (p->index_data)
584 return 0;
586 idx_name = xstrdup(p->pack_name);
587 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
588 ret = check_packed_git_idx(idx_name, p);
589 free(idx_name);
590 return ret;
593 static void scan_windows(struct packed_git *p,
594 struct packed_git **lru_p,
595 struct pack_window **lru_w,
596 struct pack_window **lru_l)
598 struct pack_window *w, *w_l;
600 for (w_l = NULL, w = p->windows; w; w = w->next) {
601 if (!w->inuse_cnt) {
602 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
603 *lru_p = p;
604 *lru_w = w;
605 *lru_l = w_l;
608 w_l = w;
612 static int unuse_one_window(struct packed_git *current, int keep_fd)
614 struct packed_git *p, *lru_p = NULL;
615 struct pack_window *lru_w = NULL, *lru_l = NULL;
617 if (current)
618 scan_windows(current, &lru_p, &lru_w, &lru_l);
619 for (p = packed_git; p; p = p->next)
620 scan_windows(p, &lru_p, &lru_w, &lru_l);
621 if (lru_p) {
622 munmap(lru_w->base, lru_w->len);
623 pack_mapped -= lru_w->len;
624 if (lru_l)
625 lru_l->next = lru_w->next;
626 else {
627 lru_p->windows = lru_w->next;
628 if (!lru_p->windows && lru_p->pack_fd != -1
629 && lru_p->pack_fd != keep_fd) {
630 close(lru_p->pack_fd);
631 pack_open_fds--;
632 lru_p->pack_fd = -1;
635 free(lru_w);
636 pack_open_windows--;
637 return 1;
639 return 0;
642 void release_pack_memory(size_t need, int fd)
644 size_t cur = pack_mapped;
645 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
646 ; /* nothing */
649 void *xmmap(void *start, size_t length,
650 int prot, int flags, int fd, off_t offset)
652 void *ret = mmap(start, length, prot, flags, fd, offset);
653 if (ret == MAP_FAILED) {
654 if (!length)
655 return NULL;
656 release_pack_memory(length, fd);
657 ret = mmap(start, length, prot, flags, fd, offset);
658 if (ret == MAP_FAILED)
659 die_errno("Out of memory? mmap failed");
661 return ret;
664 void close_pack_windows(struct packed_git *p)
666 while (p->windows) {
667 struct pack_window *w = p->windows;
669 if (w->inuse_cnt)
670 die("pack '%s' still has open windows to it",
671 p->pack_name);
672 munmap(w->base, w->len);
673 pack_mapped -= w->len;
674 pack_open_windows--;
675 p->windows = w->next;
676 free(w);
680 void unuse_pack(struct pack_window **w_cursor)
682 struct pack_window *w = *w_cursor;
683 if (w) {
684 w->inuse_cnt--;
685 *w_cursor = NULL;
689 void close_pack_index(struct packed_git *p)
691 if (p->index_data) {
692 munmap((void *)p->index_data, p->index_size);
693 p->index_data = NULL;
698 * This is used by git-repack in case a newly created pack happens to
699 * contain the same set of objects as an existing one. In that case
700 * the resulting file might be different even if its name would be the
701 * same. It is best to close any reference to the old pack before it is
702 * replaced on disk. Of course no index pointers nor windows for given pack
703 * must subsist at this point. If ever objects from this pack are requested
704 * again, the new version of the pack will be reinitialized through
705 * reprepare_packed_git().
707 void free_pack_by_name(const char *pack_name)
709 struct packed_git *p, **pp = &packed_git;
711 while (*pp) {
712 p = *pp;
713 if (strcmp(pack_name, p->pack_name) == 0) {
714 clear_delta_base_cache();
715 close_pack_windows(p);
716 if (p->pack_fd != -1) {
717 close(p->pack_fd);
718 pack_open_fds--;
720 close_pack_index(p);
721 free(p->bad_object_sha1);
722 *pp = p->next;
723 free(p);
724 return;
726 pp = &p->next;
731 * Do not call this directly as this leaks p->pack_fd on error return;
732 * call open_packed_git() instead.
734 static int open_packed_git_1(struct packed_git *p)
736 struct stat st;
737 struct pack_header hdr;
738 unsigned char sha1[20];
739 unsigned char *idx_sha1;
740 long fd_flag;
742 if (!p->index_data && open_pack_index(p))
743 return error("packfile %s index unavailable", p->pack_name);
745 if (!pack_max_fds) {
746 struct rlimit lim;
747 unsigned int max_fds;
749 if (getrlimit(RLIMIT_NOFILE, &lim))
750 die_errno("cannot get RLIMIT_NOFILE");
752 max_fds = lim.rlim_cur;
754 /* Save 3 for stdin/stdout/stderr, 22 for work */
755 if (25 < max_fds)
756 pack_max_fds = max_fds - 25;
757 else
758 pack_max_fds = 1;
761 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
762 ; /* nothing */
764 p->pack_fd = git_open_noatime(p->pack_name);
765 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
766 return -1;
767 pack_open_fds++;
769 /* If we created the struct before we had the pack we lack size. */
770 if (!p->pack_size) {
771 if (!S_ISREG(st.st_mode))
772 return error("packfile %s not a regular file", p->pack_name);
773 p->pack_size = st.st_size;
774 } else if (p->pack_size != st.st_size)
775 return error("packfile %s size changed", p->pack_name);
777 /* We leave these file descriptors open with sliding mmap;
778 * there is no point keeping them open across exec(), though.
780 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
781 if (fd_flag < 0)
782 return error("cannot determine file descriptor flags");
783 fd_flag |= FD_CLOEXEC;
784 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
785 return error("cannot set FD_CLOEXEC");
787 /* Verify we recognize this pack file format. */
788 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
789 return error("file %s is far too short to be a packfile", p->pack_name);
790 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
791 return error("file %s is not a GIT packfile", p->pack_name);
792 if (!pack_version_ok(hdr.hdr_version))
793 return error("packfile %s is version %"PRIu32" and not"
794 " supported (try upgrading GIT to a newer version)",
795 p->pack_name, ntohl(hdr.hdr_version));
797 /* Verify the pack matches its index. */
798 if (p->num_objects != ntohl(hdr.hdr_entries))
799 return error("packfile %s claims to have %"PRIu32" objects"
800 " while index indicates %"PRIu32" objects",
801 p->pack_name, ntohl(hdr.hdr_entries),
802 p->num_objects);
803 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
804 return error("end of packfile %s is unavailable", p->pack_name);
805 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
806 return error("packfile %s signature is unavailable", p->pack_name);
807 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
808 if (hashcmp(sha1, idx_sha1))
809 return error("packfile %s does not match index", p->pack_name);
810 return 0;
813 static int open_packed_git(struct packed_git *p)
815 if (!open_packed_git_1(p))
816 return 0;
817 if (p->pack_fd != -1) {
818 close(p->pack_fd);
819 pack_open_fds--;
820 p->pack_fd = -1;
822 return -1;
825 static int in_window(struct pack_window *win, off_t offset)
827 /* We must promise at least 20 bytes (one hash) after the
828 * offset is available from this window, otherwise the offset
829 * is not actually in this window and a different window (which
830 * has that one hash excess) must be used. This is to support
831 * the object header and delta base parsing routines below.
833 off_t win_off = win->offset;
834 return win_off <= offset
835 && (offset + 20) <= (win_off + win->len);
838 unsigned char *use_pack(struct packed_git *p,
839 struct pack_window **w_cursor,
840 off_t offset,
841 unsigned long *left)
843 struct pack_window *win = *w_cursor;
845 /* Since packfiles end in a hash of their content and it's
846 * pointless to ask for an offset into the middle of that
847 * hash, and the in_window function above wouldn't match
848 * don't allow an offset too close to the end of the file.
850 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
851 die("packfile %s cannot be accessed", p->pack_name);
852 if (offset > (p->pack_size - 20))
853 die("offset beyond end of packfile (truncated pack?)");
855 if (!win || !in_window(win, offset)) {
856 if (win)
857 win->inuse_cnt--;
858 for (win = p->windows; win; win = win->next) {
859 if (in_window(win, offset))
860 break;
862 if (!win) {
863 size_t window_align = packed_git_window_size / 2;
864 off_t len;
866 if (p->pack_fd == -1 && open_packed_git(p))
867 die("packfile %s cannot be accessed", p->pack_name);
869 win = xcalloc(1, sizeof(*win));
870 win->offset = (offset / window_align) * window_align;
871 len = p->pack_size - win->offset;
872 if (len > packed_git_window_size)
873 len = packed_git_window_size;
874 win->len = (size_t)len;
875 pack_mapped += win->len;
876 while (packed_git_limit < pack_mapped
877 && unuse_one_window(p, p->pack_fd))
878 ; /* nothing */
879 win->base = xmmap(NULL, win->len,
880 PROT_READ, MAP_PRIVATE,
881 p->pack_fd, win->offset);
882 if (win->base == MAP_FAILED)
883 die("packfile %s cannot be mapped: %s",
884 p->pack_name,
885 strerror(errno));
886 if (!win->offset && win->len == p->pack_size
887 && !p->do_not_close) {
888 close(p->pack_fd);
889 pack_open_fds--;
890 p->pack_fd = -1;
892 pack_mmap_calls++;
893 pack_open_windows++;
894 if (pack_mapped > peak_pack_mapped)
895 peak_pack_mapped = pack_mapped;
896 if (pack_open_windows > peak_pack_open_windows)
897 peak_pack_open_windows = pack_open_windows;
898 win->next = p->windows;
899 p->windows = win;
902 if (win != *w_cursor) {
903 win->last_used = pack_used_ctr++;
904 win->inuse_cnt++;
905 *w_cursor = win;
907 offset -= win->offset;
908 if (left)
909 *left = win->len - xsize_t(offset);
910 return win->base + offset;
913 static struct packed_git *alloc_packed_git(int extra)
915 struct packed_git *p = xmalloc(sizeof(*p) + extra);
916 memset(p, 0, sizeof(*p));
917 p->pack_fd = -1;
918 return p;
921 static void try_to_free_pack_memory(size_t size)
923 release_pack_memory(size, -1);
926 struct packed_git *add_packed_git(const char *path, int path_len, int local)
928 static int have_set_try_to_free_routine;
929 struct stat st;
930 struct packed_git *p = alloc_packed_git(path_len + 2);
932 if (!have_set_try_to_free_routine) {
933 have_set_try_to_free_routine = 1;
934 set_try_to_free_routine(try_to_free_pack_memory);
938 * Make sure a corresponding .pack file exists and that
939 * the index looks sane.
941 path_len -= strlen(".idx");
942 if (path_len < 1) {
943 free(p);
944 return NULL;
946 memcpy(p->pack_name, path, path_len);
948 strcpy(p->pack_name + path_len, ".keep");
949 if (!access(p->pack_name, F_OK))
950 p->pack_keep = 1;
952 strcpy(p->pack_name + path_len, ".pack");
953 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
954 free(p);
955 return NULL;
958 /* ok, it looks sane as far as we can check without
959 * actually mapping the pack file.
961 p->pack_size = st.st_size;
962 p->pack_local = local;
963 p->mtime = st.st_mtime;
964 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
965 hashclr(p->sha1);
966 return p;
969 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
971 const char *path = sha1_pack_name(sha1);
972 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
974 strcpy(p->pack_name, path);
975 hashcpy(p->sha1, sha1);
976 if (check_packed_git_idx(idx_path, p)) {
977 free(p);
978 return NULL;
981 return p;
984 void install_packed_git(struct packed_git *pack)
986 if (pack->pack_fd != -1)
987 pack_open_fds++;
989 pack->next = packed_git;
990 packed_git = pack;
993 static void prepare_packed_git_one(char *objdir, int local)
995 /* Ensure that this buffer is large enough so that we can
996 append "/pack/" without clobbering the stack even if
997 strlen(objdir) were PATH_MAX. */
998 char path[PATH_MAX + 1 + 4 + 1 + 1];
999 int len;
1000 DIR *dir;
1001 struct dirent *de;
1003 sprintf(path, "%s/pack", objdir);
1004 len = strlen(path);
1005 dir = opendir(path);
1006 if (!dir) {
1007 if (errno != ENOENT)
1008 error("unable to open object pack directory: %s: %s",
1009 path, strerror(errno));
1010 return;
1012 path[len++] = '/';
1013 while ((de = readdir(dir)) != NULL) {
1014 int namelen = strlen(de->d_name);
1015 struct packed_git *p;
1017 if (!has_extension(de->d_name, ".idx"))
1018 continue;
1020 if (len + namelen + 1 > sizeof(path))
1021 continue;
1023 /* Don't reopen a pack we already have. */
1024 strcpy(path + len, de->d_name);
1025 for (p = packed_git; p; p = p->next) {
1026 if (!memcmp(path, p->pack_name, len + namelen - 4))
1027 break;
1029 if (p)
1030 continue;
1031 /* See if it really is a valid .idx file with corresponding
1032 * .pack file that we can map.
1034 p = add_packed_git(path, len + namelen, local);
1035 if (!p)
1036 continue;
1037 install_packed_git(p);
1039 closedir(dir);
1042 static int sort_pack(const void *a_, const void *b_)
1044 struct packed_git *a = *((struct packed_git **)a_);
1045 struct packed_git *b = *((struct packed_git **)b_);
1046 int st;
1049 * Local packs tend to contain objects specific to our
1050 * variant of the project than remote ones. In addition,
1051 * remote ones could be on a network mounted filesystem.
1052 * Favor local ones for these reasons.
1054 st = a->pack_local - b->pack_local;
1055 if (st)
1056 return -st;
1059 * Younger packs tend to contain more recent objects,
1060 * and more recent objects tend to get accessed more
1061 * often.
1063 if (a->mtime < b->mtime)
1064 return 1;
1065 else if (a->mtime == b->mtime)
1066 return 0;
1067 return -1;
1070 static void rearrange_packed_git(void)
1072 struct packed_git **ary, *p;
1073 int i, n;
1075 for (n = 0, p = packed_git; p; p = p->next)
1076 n++;
1077 if (n < 2)
1078 return;
1080 /* prepare an array of packed_git for easier sorting */
1081 ary = xcalloc(n, sizeof(struct packed_git *));
1082 for (n = 0, p = packed_git; p; p = p->next)
1083 ary[n++] = p;
1085 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1087 /* link them back again */
1088 for (i = 0; i < n - 1; i++)
1089 ary[i]->next = ary[i + 1];
1090 ary[n - 1]->next = NULL;
1091 packed_git = ary[0];
1093 free(ary);
1096 static int prepare_packed_git_run_once = 0;
1097 void prepare_packed_git(void)
1099 struct alternate_object_database *alt;
1101 if (prepare_packed_git_run_once)
1102 return;
1103 prepare_packed_git_one(get_object_directory(), 1);
1104 prepare_alt_odb();
1105 for (alt = alt_odb_list; alt; alt = alt->next) {
1106 alt->name[-1] = 0;
1107 prepare_packed_git_one(alt->base, 0);
1108 alt->name[-1] = '/';
1110 rearrange_packed_git();
1111 prepare_packed_git_run_once = 1;
1114 void reprepare_packed_git(void)
1116 discard_revindex();
1117 prepare_packed_git_run_once = 0;
1118 prepare_packed_git();
1121 static void mark_bad_packed_object(struct packed_git *p,
1122 const unsigned char *sha1)
1124 unsigned i;
1125 for (i = 0; i < p->num_bad_objects; i++)
1126 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1127 return;
1128 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1129 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1130 p->num_bad_objects++;
1133 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1135 struct packed_git *p;
1136 unsigned i;
1138 for (p = packed_git; p; p = p->next)
1139 for (i = 0; i < p->num_bad_objects; i++)
1140 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1141 return p;
1142 return NULL;
1145 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1147 unsigned char real_sha1[20];
1148 hash_sha1_file(map, size, type, real_sha1);
1149 return hashcmp(sha1, real_sha1) ? -1 : 0;
1152 static int git_open_noatime(const char *name)
1154 static int sha1_file_open_flag = O_NOATIME;
1156 for (;;) {
1157 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1158 if (fd >= 0)
1159 return fd;
1161 /* Might the failure be due to O_NOATIME? */
1162 if (errno != ENOENT && sha1_file_open_flag) {
1163 sha1_file_open_flag = 0;
1164 continue;
1167 return -1;
1171 static int open_sha1_file(const unsigned char *sha1)
1173 int fd;
1174 char *name = sha1_file_name(sha1);
1175 struct alternate_object_database *alt;
1177 fd = git_open_noatime(name);
1178 if (fd >= 0)
1179 return fd;
1181 prepare_alt_odb();
1182 errno = ENOENT;
1183 for (alt = alt_odb_list; alt; alt = alt->next) {
1184 name = alt->name;
1185 fill_sha1_path(name, sha1);
1186 fd = git_open_noatime(alt->base);
1187 if (fd >= 0)
1188 return fd;
1190 return -1;
1193 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1195 void *map;
1196 int fd;
1198 fd = open_sha1_file(sha1);
1199 map = NULL;
1200 if (fd >= 0) {
1201 struct stat st;
1203 if (!fstat(fd, &st)) {
1204 *size = xsize_t(st.st_size);
1205 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1207 close(fd);
1209 return map;
1213 * There used to be a second loose object header format which
1214 * was meant to mimic the in-pack format, allowing for direct
1215 * copy of the object data. This format turned up not to be
1216 * really worth it and we no longer write loose objects in that
1217 * format.
1219 static int experimental_loose_object(unsigned char *map)
1221 unsigned int word;
1224 * We must determine if the buffer contains the standard
1225 * zlib-deflated stream or the experimental format based
1226 * on the in-pack object format. Compare the header byte
1227 * for each format:
1229 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1230 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1232 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1233 * in standard loose-object format, UNLESS it is a Git-pack
1234 * format object *exactly* 8 bytes in size when inflated.
1236 * However, RFC1950 also specifies that the 1st 16-bit word
1237 * must be divisible by 31 - this checksum tells us our buffer
1238 * is in the standard format, giving a false positive only if
1239 * the 1st word of the Git-pack format object happens to be
1240 * divisible by 31, ie:
1241 * ((byte0 * 256) + byte1) % 31 = 0
1242 * => 0ttt10000www1000 % 31 = 0
1244 * As it happens, this case can only arise for www=3 & ttt=1
1245 * - ie, a Commit object, which would have to be 8 bytes in
1246 * size. As no Commit can be that small, we find that the
1247 * combination of these two criteria (bitmask & checksum)
1248 * can always correctly determine the buffer format.
1250 word = (map[0] << 8) + map[1];
1251 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1252 return 0;
1253 else
1254 return 1;
1257 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1258 unsigned long len, enum object_type *type, unsigned long *sizep)
1260 unsigned shift;
1261 unsigned long size, c;
1262 unsigned long used = 0;
1264 c = buf[used++];
1265 *type = (c >> 4) & 7;
1266 size = c & 15;
1267 shift = 4;
1268 while (c & 0x80) {
1269 if (len <= used || bitsizeof(long) <= shift) {
1270 error("bad object header");
1271 size = used = 0;
1272 break;
1274 c = buf[used++];
1275 size += (c & 0x7f) << shift;
1276 shift += 7;
1278 *sizep = size;
1279 return used;
1282 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1284 unsigned long size, used;
1285 static const char valid_loose_object_type[8] = {
1286 0, /* OBJ_EXT */
1287 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1288 0, /* "delta" and others are invalid in a loose object */
1290 enum object_type type;
1292 /* Get the data stream */
1293 memset(stream, 0, sizeof(*stream));
1294 stream->next_in = map;
1295 stream->avail_in = mapsize;
1296 stream->next_out = buffer;
1297 stream->avail_out = bufsiz;
1299 if (experimental_loose_object(map)) {
1301 * The old experimental format we no longer produce;
1302 * we can still read it.
1304 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1305 if (!used || !valid_loose_object_type[type])
1306 return -1;
1307 map += used;
1308 mapsize -= used;
1310 /* Set up the stream for the rest.. */
1311 stream->next_in = map;
1312 stream->avail_in = mapsize;
1313 git_inflate_init(stream);
1315 /* And generate the fake traditional header */
1316 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1317 typename(type), size);
1318 return 0;
1320 git_inflate_init(stream);
1321 return git_inflate(stream, 0);
1324 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1326 int bytes = strlen(buffer) + 1;
1327 unsigned char *buf = xmallocz(size);
1328 unsigned long n;
1329 int status = Z_OK;
1331 n = stream->total_out - bytes;
1332 if (n > size)
1333 n = size;
1334 memcpy(buf, (char *) buffer + bytes, n);
1335 bytes = n;
1336 if (bytes <= size) {
1338 * The above condition must be (bytes <= size), not
1339 * (bytes < size). In other words, even though we
1340 * expect no more output and set avail_out to zero,
1341 * the input zlib stream may have bytes that express
1342 * "this concludes the stream", and we *do* want to
1343 * eat that input.
1345 * Otherwise we would not be able to test that we
1346 * consumed all the input to reach the expected size;
1347 * we also want to check that zlib tells us that all
1348 * went well with status == Z_STREAM_END at the end.
1350 stream->next_out = buf + bytes;
1351 stream->avail_out = size - bytes;
1352 while (status == Z_OK)
1353 status = git_inflate(stream, Z_FINISH);
1355 if (status == Z_STREAM_END && !stream->avail_in) {
1356 git_inflate_end(stream);
1357 return buf;
1360 if (status < 0)
1361 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1362 else if (stream->avail_in)
1363 error("garbage at end of loose object '%s'",
1364 sha1_to_hex(sha1));
1365 free(buf);
1366 return NULL;
1370 * We used to just use "sscanf()", but that's actually way
1371 * too permissive for what we want to check. So do an anal
1372 * object header parse by hand.
1374 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1376 char type[10];
1377 int i;
1378 unsigned long size;
1381 * The type can be at most ten bytes (including the
1382 * terminating '\0' that we add), and is followed by
1383 * a space.
1385 i = 0;
1386 for (;;) {
1387 char c = *hdr++;
1388 if (c == ' ')
1389 break;
1390 type[i++] = c;
1391 if (i >= sizeof(type))
1392 return -1;
1394 type[i] = 0;
1397 * The length must follow immediately, and be in canonical
1398 * decimal format (ie "010" is not valid).
1400 size = *hdr++ - '0';
1401 if (size > 9)
1402 return -1;
1403 if (size) {
1404 for (;;) {
1405 unsigned long c = *hdr - '0';
1406 if (c > 9)
1407 break;
1408 hdr++;
1409 size = size * 10 + c;
1412 *sizep = size;
1415 * The length must be followed by a zero byte
1417 return *hdr ? -1 : type_from_string(type);
1420 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1422 int ret;
1423 git_zstream stream;
1424 char hdr[8192];
1426 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1427 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1428 return NULL;
1430 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1433 unsigned long get_size_from_delta(struct packed_git *p,
1434 struct pack_window **w_curs,
1435 off_t curpos)
1437 const unsigned char *data;
1438 unsigned char delta_head[20], *in;
1439 git_zstream stream;
1440 int st;
1442 memset(&stream, 0, sizeof(stream));
1443 stream.next_out = delta_head;
1444 stream.avail_out = sizeof(delta_head);
1446 git_inflate_init(&stream);
1447 do {
1448 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1449 stream.next_in = in;
1450 st = git_inflate(&stream, Z_FINISH);
1451 curpos += stream.next_in - in;
1452 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1453 stream.total_out < sizeof(delta_head));
1454 git_inflate_end(&stream);
1455 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1456 error("delta data unpack-initial failed");
1457 return 0;
1460 /* Examine the initial part of the delta to figure out
1461 * the result size.
1463 data = delta_head;
1465 /* ignore base size */
1466 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1468 /* Read the result size */
1469 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1472 static off_t get_delta_base(struct packed_git *p,
1473 struct pack_window **w_curs,
1474 off_t *curpos,
1475 enum object_type type,
1476 off_t delta_obj_offset)
1478 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1479 off_t base_offset;
1481 /* use_pack() assured us we have [base_info, base_info + 20)
1482 * as a range that we can look at without walking off the
1483 * end of the mapped window. Its actually the hash size
1484 * that is assured. An OFS_DELTA longer than the hash size
1485 * is stupid, as then a REF_DELTA would be smaller to store.
1487 if (type == OBJ_OFS_DELTA) {
1488 unsigned used = 0;
1489 unsigned char c = base_info[used++];
1490 base_offset = c & 127;
1491 while (c & 128) {
1492 base_offset += 1;
1493 if (!base_offset || MSB(base_offset, 7))
1494 return 0; /* overflow */
1495 c = base_info[used++];
1496 base_offset = (base_offset << 7) + (c & 127);
1498 base_offset = delta_obj_offset - base_offset;
1499 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1500 return 0; /* out of bound */
1501 *curpos += used;
1502 } else if (type == OBJ_REF_DELTA) {
1503 /* The base entry _must_ be in the same pack */
1504 base_offset = find_pack_entry_one(base_info, p);
1505 *curpos += 20;
1506 } else
1507 die("I am totally screwed");
1508 return base_offset;
1511 /* forward declaration for a mutually recursive function */
1512 static int packed_object_info(struct packed_git *p, off_t offset,
1513 unsigned long *sizep, int *rtype);
1515 static int packed_delta_info(struct packed_git *p,
1516 struct pack_window **w_curs,
1517 off_t curpos,
1518 enum object_type type,
1519 off_t obj_offset,
1520 unsigned long *sizep)
1522 off_t base_offset;
1524 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1525 if (!base_offset)
1526 return OBJ_BAD;
1527 type = packed_object_info(p, base_offset, NULL, NULL);
1528 if (type <= OBJ_NONE) {
1529 struct revindex_entry *revidx;
1530 const unsigned char *base_sha1;
1531 revidx = find_pack_revindex(p, base_offset);
1532 if (!revidx)
1533 return OBJ_BAD;
1534 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1535 mark_bad_packed_object(p, base_sha1);
1536 type = sha1_object_info(base_sha1, NULL);
1537 if (type <= OBJ_NONE)
1538 return OBJ_BAD;
1541 /* We choose to only get the type of the base object and
1542 * ignore potentially corrupt pack file that expects the delta
1543 * based on a base with a wrong size. This saves tons of
1544 * inflate() calls.
1546 if (sizep) {
1547 *sizep = get_size_from_delta(p, w_curs, curpos);
1548 if (*sizep == 0)
1549 type = OBJ_BAD;
1552 return type;
1555 int unpack_object_header(struct packed_git *p,
1556 struct pack_window **w_curs,
1557 off_t *curpos,
1558 unsigned long *sizep)
1560 unsigned char *base;
1561 unsigned long left;
1562 unsigned long used;
1563 enum object_type type;
1565 /* use_pack() assures us we have [base, base + 20) available
1566 * as a range that we can look at. (Its actually the hash
1567 * size that is assured.) With our object header encoding
1568 * the maximum deflated object size is 2^137, which is just
1569 * insane, so we know won't exceed what we have been given.
1571 base = use_pack(p, w_curs, *curpos, &left);
1572 used = unpack_object_header_buffer(base, left, &type, sizep);
1573 if (!used) {
1574 type = OBJ_BAD;
1575 } else
1576 *curpos += used;
1578 return type;
1581 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1582 unsigned long *sizep, int *rtype)
1584 struct pack_window *w_curs = NULL;
1585 unsigned long size;
1586 off_t curpos = obj_offset;
1587 enum object_type type;
1589 type = unpack_object_header(p, &w_curs, &curpos, &size);
1590 if (rtype)
1591 *rtype = type; /* representation type */
1593 switch (type) {
1594 case OBJ_OFS_DELTA:
1595 case OBJ_REF_DELTA:
1596 type = packed_delta_info(p, &w_curs, curpos,
1597 type, obj_offset, sizep);
1598 break;
1599 case OBJ_COMMIT:
1600 case OBJ_TREE:
1601 case OBJ_BLOB:
1602 case OBJ_TAG:
1603 if (sizep)
1604 *sizep = size;
1605 break;
1606 default:
1607 error("unknown object type %i at offset %"PRIuMAX" in %s",
1608 type, (uintmax_t)obj_offset, p->pack_name);
1609 type = OBJ_BAD;
1611 unuse_pack(&w_curs);
1612 return type;
1615 static void *unpack_compressed_entry(struct packed_git *p,
1616 struct pack_window **w_curs,
1617 off_t curpos,
1618 unsigned long size)
1620 int st;
1621 git_zstream stream;
1622 unsigned char *buffer, *in;
1624 buffer = xmallocz(size);
1625 memset(&stream, 0, sizeof(stream));
1626 stream.next_out = buffer;
1627 stream.avail_out = size + 1;
1629 git_inflate_init(&stream);
1630 do {
1631 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1632 stream.next_in = in;
1633 st = git_inflate(&stream, Z_FINISH);
1634 if (!stream.avail_out)
1635 break; /* the payload is larger than it should be */
1636 curpos += stream.next_in - in;
1637 } while (st == Z_OK || st == Z_BUF_ERROR);
1638 git_inflate_end(&stream);
1639 if ((st != Z_STREAM_END) || stream.total_out != size) {
1640 free(buffer);
1641 return NULL;
1644 return buffer;
1647 #define MAX_DELTA_CACHE (256)
1649 static size_t delta_base_cached;
1651 static struct delta_base_cache_lru_list {
1652 struct delta_base_cache_lru_list *prev;
1653 struct delta_base_cache_lru_list *next;
1654 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1656 static struct delta_base_cache_entry {
1657 struct delta_base_cache_lru_list lru;
1658 void *data;
1659 struct packed_git *p;
1660 off_t base_offset;
1661 unsigned long size;
1662 enum object_type type;
1663 } delta_base_cache[MAX_DELTA_CACHE];
1665 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1667 unsigned long hash;
1669 hash = (unsigned long)p + (unsigned long)base_offset;
1670 hash += (hash >> 8) + (hash >> 16);
1671 return hash % MAX_DELTA_CACHE;
1674 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1676 unsigned long hash = pack_entry_hash(p, base_offset);
1677 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1678 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1681 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1682 unsigned long *base_size, enum object_type *type, int keep_cache)
1684 void *ret;
1685 unsigned long hash = pack_entry_hash(p, base_offset);
1686 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1688 ret = ent->data;
1689 if (!ret || ent->p != p || ent->base_offset != base_offset)
1690 return unpack_entry(p, base_offset, type, base_size);
1692 if (!keep_cache) {
1693 ent->data = NULL;
1694 ent->lru.next->prev = ent->lru.prev;
1695 ent->lru.prev->next = ent->lru.next;
1696 delta_base_cached -= ent->size;
1697 } else {
1698 ret = xmemdupz(ent->data, ent->size);
1700 *type = ent->type;
1701 *base_size = ent->size;
1702 return ret;
1705 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1707 if (ent->data) {
1708 free(ent->data);
1709 ent->data = NULL;
1710 ent->lru.next->prev = ent->lru.prev;
1711 ent->lru.prev->next = ent->lru.next;
1712 delta_base_cached -= ent->size;
1716 void clear_delta_base_cache(void)
1718 unsigned long p;
1719 for (p = 0; p < MAX_DELTA_CACHE; p++)
1720 release_delta_base_cache(&delta_base_cache[p]);
1723 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1724 void *base, unsigned long base_size, enum object_type type)
1726 unsigned long hash = pack_entry_hash(p, base_offset);
1727 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1728 struct delta_base_cache_lru_list *lru;
1730 release_delta_base_cache(ent);
1731 delta_base_cached += base_size;
1733 for (lru = delta_base_cache_lru.next;
1734 delta_base_cached > delta_base_cache_limit
1735 && lru != &delta_base_cache_lru;
1736 lru = lru->next) {
1737 struct delta_base_cache_entry *f = (void *)lru;
1738 if (f->type == OBJ_BLOB)
1739 release_delta_base_cache(f);
1741 for (lru = delta_base_cache_lru.next;
1742 delta_base_cached > delta_base_cache_limit
1743 && lru != &delta_base_cache_lru;
1744 lru = lru->next) {
1745 struct delta_base_cache_entry *f = (void *)lru;
1746 release_delta_base_cache(f);
1749 ent->p = p;
1750 ent->base_offset = base_offset;
1751 ent->type = type;
1752 ent->data = base;
1753 ent->size = base_size;
1754 ent->lru.next = &delta_base_cache_lru;
1755 ent->lru.prev = delta_base_cache_lru.prev;
1756 delta_base_cache_lru.prev->next = &ent->lru;
1757 delta_base_cache_lru.prev = &ent->lru;
1760 static void *read_object(const unsigned char *sha1, enum object_type *type,
1761 unsigned long *size);
1763 static void *unpack_delta_entry(struct packed_git *p,
1764 struct pack_window **w_curs,
1765 off_t curpos,
1766 unsigned long delta_size,
1767 off_t obj_offset,
1768 enum object_type *type,
1769 unsigned long *sizep)
1771 void *delta_data, *result, *base;
1772 unsigned long base_size;
1773 off_t base_offset;
1775 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1776 if (!base_offset) {
1777 error("failed to validate delta base reference "
1778 "at offset %"PRIuMAX" from %s",
1779 (uintmax_t)curpos, p->pack_name);
1780 return NULL;
1782 unuse_pack(w_curs);
1783 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1784 if (!base) {
1786 * We're probably in deep shit, but let's try to fetch
1787 * the required base anyway from another pack or loose.
1788 * This is costly but should happen only in the presence
1789 * of a corrupted pack, and is better than failing outright.
1791 struct revindex_entry *revidx;
1792 const unsigned char *base_sha1;
1793 revidx = find_pack_revindex(p, base_offset);
1794 if (!revidx)
1795 return NULL;
1796 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1797 error("failed to read delta base object %s"
1798 " at offset %"PRIuMAX" from %s",
1799 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1800 p->pack_name);
1801 mark_bad_packed_object(p, base_sha1);
1802 base = read_object(base_sha1, type, &base_size);
1803 if (!base)
1804 return NULL;
1807 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1808 if (!delta_data) {
1809 error("failed to unpack compressed delta "
1810 "at offset %"PRIuMAX" from %s",
1811 (uintmax_t)curpos, p->pack_name);
1812 free(base);
1813 return NULL;
1815 result = patch_delta(base, base_size,
1816 delta_data, delta_size,
1817 sizep);
1818 if (!result)
1819 die("failed to apply delta");
1820 free(delta_data);
1821 add_delta_base_cache(p, base_offset, base, base_size, *type);
1822 return result;
1825 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1827 static FILE *log_file;
1829 if (!log_file) {
1830 log_file = fopen(log_pack_access, "w");
1831 if (!log_file) {
1832 error("cannot open pack access log '%s' for writing: %s",
1833 log_pack_access, strerror(errno));
1834 log_pack_access = NULL;
1835 return;
1838 fprintf(log_file, "%s %"PRIuMAX"\n",
1839 p->pack_name, (uintmax_t)obj_offset);
1840 fflush(log_file);
1843 int do_check_packed_object_crc;
1845 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1846 enum object_type *type, unsigned long *sizep)
1848 struct pack_window *w_curs = NULL;
1849 off_t curpos = obj_offset;
1850 void *data;
1852 if (log_pack_access)
1853 write_pack_access_log(p, obj_offset);
1855 if (do_check_packed_object_crc && p->index_version > 1) {
1856 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1857 unsigned long len = revidx[1].offset - obj_offset;
1858 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1859 const unsigned char *sha1 =
1860 nth_packed_object_sha1(p, revidx->nr);
1861 error("bad packed object CRC for %s",
1862 sha1_to_hex(sha1));
1863 mark_bad_packed_object(p, sha1);
1864 unuse_pack(&w_curs);
1865 return NULL;
1869 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1870 switch (*type) {
1871 case OBJ_OFS_DELTA:
1872 case OBJ_REF_DELTA:
1873 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1874 obj_offset, type, sizep);
1875 break;
1876 case OBJ_COMMIT:
1877 case OBJ_TREE:
1878 case OBJ_BLOB:
1879 case OBJ_TAG:
1880 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1881 break;
1882 default:
1883 data = NULL;
1884 error("unknown object type %i at offset %"PRIuMAX" in %s",
1885 *type, (uintmax_t)obj_offset, p->pack_name);
1887 unuse_pack(&w_curs);
1888 return data;
1891 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1892 uint32_t n)
1894 const unsigned char *index = p->index_data;
1895 if (!index) {
1896 if (open_pack_index(p))
1897 return NULL;
1898 index = p->index_data;
1900 if (n >= p->num_objects)
1901 return NULL;
1902 index += 4 * 256;
1903 if (p->index_version == 1) {
1904 return index + 24 * n + 4;
1905 } else {
1906 index += 8;
1907 return index + 20 * n;
1911 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1913 const unsigned char *index = p->index_data;
1914 index += 4 * 256;
1915 if (p->index_version == 1) {
1916 return ntohl(*((uint32_t *)(index + 24 * n)));
1917 } else {
1918 uint32_t off;
1919 index += 8 + p->num_objects * (20 + 4);
1920 off = ntohl(*((uint32_t *)(index + 4 * n)));
1921 if (!(off & 0x80000000))
1922 return off;
1923 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1924 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1925 ntohl(*((uint32_t *)(index + 4)));
1929 off_t find_pack_entry_one(const unsigned char *sha1,
1930 struct packed_git *p)
1932 const uint32_t *level1_ofs = p->index_data;
1933 const unsigned char *index = p->index_data;
1934 unsigned hi, lo, stride;
1935 static int use_lookup = -1;
1936 static int debug_lookup = -1;
1938 if (debug_lookup < 0)
1939 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1941 if (!index) {
1942 if (open_pack_index(p))
1943 return 0;
1944 level1_ofs = p->index_data;
1945 index = p->index_data;
1947 if (p->index_version > 1) {
1948 level1_ofs += 2;
1949 index += 8;
1951 index += 4 * 256;
1952 hi = ntohl(level1_ofs[*sha1]);
1953 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1954 if (p->index_version > 1) {
1955 stride = 20;
1956 } else {
1957 stride = 24;
1958 index += 4;
1961 if (debug_lookup)
1962 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1963 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1965 if (use_lookup < 0)
1966 use_lookup = !!getenv("GIT_USE_LOOKUP");
1967 if (use_lookup) {
1968 int pos = sha1_entry_pos(index, stride, 0,
1969 lo, hi, p->num_objects, sha1);
1970 if (pos < 0)
1971 return 0;
1972 return nth_packed_object_offset(p, pos);
1975 do {
1976 unsigned mi = (lo + hi) / 2;
1977 int cmp = hashcmp(index + mi * stride, sha1);
1979 if (debug_lookup)
1980 printf("lo %u hi %u rg %u mi %u\n",
1981 lo, hi, hi - lo, mi);
1982 if (!cmp)
1983 return nth_packed_object_offset(p, mi);
1984 if (cmp > 0)
1985 hi = mi;
1986 else
1987 lo = mi+1;
1988 } while (lo < hi);
1989 return 0;
1992 int is_pack_valid(struct packed_git *p)
1994 /* An already open pack is known to be valid. */
1995 if (p->pack_fd != -1)
1996 return 1;
1998 /* If the pack has one window completely covering the
1999 * file size, the pack is known to be valid even if
2000 * the descriptor is not currently open.
2002 if (p->windows) {
2003 struct pack_window *w = p->windows;
2005 if (!w->offset && w->len == p->pack_size)
2006 return 1;
2009 /* Force the pack to open to prove its valid. */
2010 return !open_packed_git(p);
2013 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2015 static struct packed_git *last_found = (void *)1;
2016 struct packed_git *p;
2017 off_t offset;
2019 prepare_packed_git();
2020 if (!packed_git)
2021 return 0;
2022 p = (last_found == (void *)1) ? packed_git : last_found;
2024 do {
2025 if (p->num_bad_objects) {
2026 unsigned i;
2027 for (i = 0; i < p->num_bad_objects; i++)
2028 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2029 goto next;
2032 offset = find_pack_entry_one(sha1, p);
2033 if (offset) {
2035 * We are about to tell the caller where they can
2036 * locate the requested object. We better make
2037 * sure the packfile is still here and can be
2038 * accessed before supplying that answer, as
2039 * it may have been deleted since the index
2040 * was loaded!
2042 if (!is_pack_valid(p)) {
2043 warning("packfile %s cannot be accessed", p->pack_name);
2044 goto next;
2046 e->offset = offset;
2047 e->p = p;
2048 hashcpy(e->sha1, sha1);
2049 last_found = p;
2050 return 1;
2053 next:
2054 if (p == last_found)
2055 p = packed_git;
2056 else
2057 p = p->next;
2058 if (p == last_found)
2059 p = p->next;
2060 } while (p);
2061 return 0;
2064 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2065 struct packed_git *packs)
2067 struct packed_git *p;
2069 for (p = packs; p; p = p->next) {
2070 if (find_pack_entry_one(sha1, p))
2071 return p;
2073 return NULL;
2077 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2079 int status;
2080 unsigned long mapsize, size;
2081 void *map;
2082 git_zstream stream;
2083 char hdr[32];
2085 map = map_sha1_file(sha1, &mapsize);
2086 if (!map)
2087 return error("unable to find %s", sha1_to_hex(sha1));
2088 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2089 status = error("unable to unpack %s header",
2090 sha1_to_hex(sha1));
2091 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2092 status = error("unable to parse %s header", sha1_to_hex(sha1));
2093 else if (sizep)
2094 *sizep = size;
2095 git_inflate_end(&stream);
2096 munmap(map, mapsize);
2097 return status;
2100 /* returns enum object_type or negative */
2101 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2103 struct cached_object *co;
2104 struct pack_entry e;
2105 int status, rtype;
2107 co = find_cached_object(sha1);
2108 if (co) {
2109 if (oi->sizep)
2110 *(oi->sizep) = co->size;
2111 oi->whence = OI_CACHED;
2112 return co->type;
2115 if (!find_pack_entry(sha1, &e)) {
2116 /* Most likely it's a loose object. */
2117 status = sha1_loose_object_info(sha1, oi->sizep);
2118 if (status >= 0) {
2119 oi->whence = OI_LOOSE;
2120 return status;
2123 /* Not a loose object; someone else may have just packed it. */
2124 reprepare_packed_git();
2125 if (!find_pack_entry(sha1, &e))
2126 return status;
2129 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2130 if (status < 0) {
2131 mark_bad_packed_object(e.p, sha1);
2132 status = sha1_object_info_extended(sha1, oi);
2133 } else if (in_delta_base_cache(e.p, e.offset)) {
2134 oi->whence = OI_DBCACHED;
2135 } else {
2136 oi->whence = OI_PACKED;
2137 oi->u.packed.offset = e.offset;
2138 oi->u.packed.pack = e.p;
2139 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2140 rtype == OBJ_OFS_DELTA);
2143 return status;
2146 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2148 struct object_info oi;
2150 oi.sizep = sizep;
2151 return sha1_object_info_extended(sha1, &oi);
2154 static void *read_packed_sha1(const unsigned char *sha1,
2155 enum object_type *type, unsigned long *size)
2157 struct pack_entry e;
2158 void *data;
2160 if (!find_pack_entry(sha1, &e))
2161 return NULL;
2162 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2163 if (!data) {
2165 * We're probably in deep shit, but let's try to fetch
2166 * the required object anyway from another pack or loose.
2167 * This should happen only in the presence of a corrupted
2168 * pack, and is better than failing outright.
2170 error("failed to read object %s at offset %"PRIuMAX" from %s",
2171 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2172 mark_bad_packed_object(e.p, sha1);
2173 data = read_object(sha1, type, size);
2175 return data;
2178 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2179 unsigned char *sha1)
2181 struct cached_object *co;
2183 hash_sha1_file(buf, len, typename(type), sha1);
2184 if (has_sha1_file(sha1) || find_cached_object(sha1))
2185 return 0;
2186 if (cached_object_alloc <= cached_object_nr) {
2187 cached_object_alloc = alloc_nr(cached_object_alloc);
2188 cached_objects = xrealloc(cached_objects,
2189 sizeof(*cached_objects) *
2190 cached_object_alloc);
2192 co = &cached_objects[cached_object_nr++];
2193 co->size = len;
2194 co->type = type;
2195 co->buf = xmalloc(len);
2196 memcpy(co->buf, buf, len);
2197 hashcpy(co->sha1, sha1);
2198 return 0;
2201 static void *read_object(const unsigned char *sha1, enum object_type *type,
2202 unsigned long *size)
2204 unsigned long mapsize;
2205 void *map, *buf;
2206 struct cached_object *co;
2208 co = find_cached_object(sha1);
2209 if (co) {
2210 *type = co->type;
2211 *size = co->size;
2212 return xmemdupz(co->buf, co->size);
2215 buf = read_packed_sha1(sha1, type, size);
2216 if (buf)
2217 return buf;
2218 map = map_sha1_file(sha1, &mapsize);
2219 if (map) {
2220 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2221 munmap(map, mapsize);
2222 return buf;
2224 reprepare_packed_git();
2225 return read_packed_sha1(sha1, type, size);
2229 * This function dies on corrupt objects; the callers who want to
2230 * deal with them should arrange to call read_object() and give error
2231 * messages themselves.
2233 void *read_sha1_file_extended(const unsigned char *sha1,
2234 enum object_type *type,
2235 unsigned long *size,
2236 unsigned flag)
2238 void *data;
2239 char *path;
2240 const struct packed_git *p;
2241 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2242 ? lookup_replace_object(sha1) : sha1;
2244 errno = 0;
2245 data = read_object(repl, type, size);
2246 if (data)
2247 return data;
2249 if (errno && errno != ENOENT)
2250 die_errno("failed to read object %s", sha1_to_hex(sha1));
2252 /* die if we replaced an object with one that does not exist */
2253 if (repl != sha1)
2254 die("replacement %s not found for %s",
2255 sha1_to_hex(repl), sha1_to_hex(sha1));
2257 if (has_loose_object(repl)) {
2258 path = sha1_file_name(sha1);
2259 die("loose object %s (stored in %s) is corrupt",
2260 sha1_to_hex(repl), path);
2263 if ((p = has_packed_and_bad(repl)) != NULL)
2264 die("packed object %s (stored in %s) is corrupt",
2265 sha1_to_hex(repl), p->pack_name);
2267 return NULL;
2270 void *read_object_with_reference(const unsigned char *sha1,
2271 const char *required_type_name,
2272 unsigned long *size,
2273 unsigned char *actual_sha1_return)
2275 enum object_type type, required_type;
2276 void *buffer;
2277 unsigned long isize;
2278 unsigned char actual_sha1[20];
2280 required_type = type_from_string(required_type_name);
2281 hashcpy(actual_sha1, sha1);
2282 while (1) {
2283 int ref_length = -1;
2284 const char *ref_type = NULL;
2286 buffer = read_sha1_file(actual_sha1, &type, &isize);
2287 if (!buffer)
2288 return NULL;
2289 if (type == required_type) {
2290 *size = isize;
2291 if (actual_sha1_return)
2292 hashcpy(actual_sha1_return, actual_sha1);
2293 return buffer;
2295 /* Handle references */
2296 else if (type == OBJ_COMMIT)
2297 ref_type = "tree ";
2298 else if (type == OBJ_TAG)
2299 ref_type = "object ";
2300 else {
2301 free(buffer);
2302 return NULL;
2304 ref_length = strlen(ref_type);
2306 if (ref_length + 40 > isize ||
2307 memcmp(buffer, ref_type, ref_length) ||
2308 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2309 free(buffer);
2310 return NULL;
2312 free(buffer);
2313 /* Now we have the ID of the referred-to object in
2314 * actual_sha1. Check again. */
2318 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2319 const char *type, unsigned char *sha1,
2320 char *hdr, int *hdrlen)
2322 git_SHA_CTX c;
2324 /* Generate the header */
2325 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2327 /* Sha1.. */
2328 git_SHA1_Init(&c);
2329 git_SHA1_Update(&c, hdr, *hdrlen);
2330 git_SHA1_Update(&c, buf, len);
2331 git_SHA1_Final(sha1, &c);
2335 * Move the just written object into its final resting place.
2336 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2337 * "moving" is only a part of what it does, when no patch between
2338 * master to pu changes the call sites of this function.
2340 int move_temp_to_file(const char *tmpfile, const char *filename)
2342 int ret = 0;
2344 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2345 goto try_rename;
2346 else if (link(tmpfile, filename))
2347 ret = errno;
2350 * Coda hack - coda doesn't like cross-directory links,
2351 * so we fall back to a rename, which will mean that it
2352 * won't be able to check collisions, but that's not a
2353 * big deal.
2355 * The same holds for FAT formatted media.
2357 * When this succeeds, we just return. We have nothing
2358 * left to unlink.
2360 if (ret && ret != EEXIST) {
2361 try_rename:
2362 if (!rename(tmpfile, filename))
2363 goto out;
2364 ret = errno;
2366 unlink_or_warn(tmpfile);
2367 if (ret) {
2368 if (ret != EEXIST) {
2369 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2371 /* FIXME!!! Collision check here ? */
2374 out:
2375 if (adjust_shared_perm(filename))
2376 return error("unable to set permission to '%s'", filename);
2377 return 0;
2380 static int write_buffer(int fd, const void *buf, size_t len)
2382 if (write_in_full(fd, buf, len) < 0)
2383 return error("file write error (%s)", strerror(errno));
2384 return 0;
2387 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2388 unsigned char *sha1)
2390 char hdr[32];
2391 int hdrlen;
2392 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2393 return 0;
2396 /* Finalize a file on disk, and close it. */
2397 static void close_sha1_file(int fd)
2399 if (fsync_object_files)
2400 fsync_or_die(fd, "sha1 file");
2401 if (close(fd) != 0)
2402 die_errno("error when closing sha1 file");
2405 /* Size of directory component, including the ending '/' */
2406 static inline int directory_size(const char *filename)
2408 const char *s = strrchr(filename, '/');
2409 if (!s)
2410 return 0;
2411 return s - filename + 1;
2415 * This creates a temporary file in the same directory as the final
2416 * 'filename'
2418 * We want to avoid cross-directory filename renames, because those
2419 * can have problems on various filesystems (FAT, NFS, Coda).
2421 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2423 int fd, dirlen = directory_size(filename);
2425 if (dirlen + 20 > bufsiz) {
2426 errno = ENAMETOOLONG;
2427 return -1;
2429 memcpy(buffer, filename, dirlen);
2430 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2431 fd = git_mkstemp_mode(buffer, 0444);
2432 if (fd < 0 && dirlen && errno == ENOENT) {
2433 /* Make sure the directory exists */
2434 memcpy(buffer, filename, dirlen);
2435 buffer[dirlen-1] = 0;
2436 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2437 return -1;
2439 /* Try again */
2440 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2441 fd = git_mkstemp_mode(buffer, 0444);
2443 return fd;
2446 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2447 const void *buf, unsigned long len, time_t mtime)
2449 int fd, ret;
2450 unsigned char compressed[4096];
2451 git_zstream stream;
2452 git_SHA_CTX c;
2453 unsigned char parano_sha1[20];
2454 char *filename;
2455 static char tmpfile[PATH_MAX];
2457 filename = sha1_file_name(sha1);
2458 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2459 if (fd < 0) {
2460 if (errno == EACCES)
2461 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2462 else
2463 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2466 /* Set it up */
2467 memset(&stream, 0, sizeof(stream));
2468 git_deflate_init(&stream, zlib_compression_level);
2469 stream.next_out = compressed;
2470 stream.avail_out = sizeof(compressed);
2471 git_SHA1_Init(&c);
2473 /* First header.. */
2474 stream.next_in = (unsigned char *)hdr;
2475 stream.avail_in = hdrlen;
2476 while (git_deflate(&stream, 0) == Z_OK)
2477 ; /* nothing */
2478 git_SHA1_Update(&c, hdr, hdrlen);
2480 /* Then the data itself.. */
2481 stream.next_in = (void *)buf;
2482 stream.avail_in = len;
2483 do {
2484 unsigned char *in0 = stream.next_in;
2485 ret = git_deflate(&stream, Z_FINISH);
2486 git_SHA1_Update(&c, in0, stream.next_in - in0);
2487 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2488 die("unable to write sha1 file");
2489 stream.next_out = compressed;
2490 stream.avail_out = sizeof(compressed);
2491 } while (ret == Z_OK);
2493 if (ret != Z_STREAM_END)
2494 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2495 ret = git_deflate_end_gently(&stream);
2496 if (ret != Z_OK)
2497 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2498 git_SHA1_Final(parano_sha1, &c);
2499 if (hashcmp(sha1, parano_sha1) != 0)
2500 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2502 close_sha1_file(fd);
2504 if (mtime) {
2505 struct utimbuf utb;
2506 utb.actime = mtime;
2507 utb.modtime = mtime;
2508 if (utime(tmpfile, &utb) < 0)
2509 warning("failed utime() on %s: %s",
2510 tmpfile, strerror(errno));
2513 return move_temp_to_file(tmpfile, filename);
2516 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2518 unsigned char sha1[20];
2519 char hdr[32];
2520 int hdrlen;
2522 /* Normally if we have it in the pack then we do not bother writing
2523 * it out into .git/objects/??/?{38} file.
2525 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2526 if (returnsha1)
2527 hashcpy(returnsha1, sha1);
2528 if (has_sha1_file(sha1))
2529 return 0;
2530 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2533 int force_object_loose(const unsigned char *sha1, time_t mtime)
2535 void *buf;
2536 unsigned long len;
2537 enum object_type type;
2538 char hdr[32];
2539 int hdrlen;
2540 int ret;
2542 if (has_loose_object(sha1))
2543 return 0;
2544 buf = read_packed_sha1(sha1, &type, &len);
2545 if (!buf)
2546 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2547 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2548 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2549 free(buf);
2551 return ret;
2554 int has_pack_index(const unsigned char *sha1)
2556 struct stat st;
2557 if (stat(sha1_pack_index_name(sha1), &st))
2558 return 0;
2559 return 1;
2562 int has_sha1_pack(const unsigned char *sha1)
2564 struct pack_entry e;
2565 return find_pack_entry(sha1, &e);
2568 int has_sha1_file(const unsigned char *sha1)
2570 struct pack_entry e;
2572 if (find_pack_entry(sha1, &e))
2573 return 1;
2574 return has_loose_object(sha1);
2577 static void check_tree(const void *buf, size_t size)
2579 struct tree_desc desc;
2580 struct name_entry entry;
2582 init_tree_desc(&desc, buf, size);
2583 while (tree_entry(&desc, &entry))
2584 /* do nothing
2585 * tree_entry() will die() on malformed entries */
2589 static void check_commit(const void *buf, size_t size)
2591 struct commit c;
2592 memset(&c, 0, sizeof(c));
2593 if (parse_commit_buffer(&c, buf, size))
2594 die("corrupt commit");
2597 static void check_tag(const void *buf, size_t size)
2599 struct tag t;
2600 memset(&t, 0, sizeof(t));
2601 if (parse_tag_buffer(&t, buf, size))
2602 die("corrupt tag");
2605 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2606 enum object_type type,
2607 const char *path, unsigned flags)
2609 int ret, re_allocated = 0;
2610 int write_object = flags & HASH_WRITE_OBJECT;
2612 if (!type)
2613 type = OBJ_BLOB;
2616 * Convert blobs to git internal format
2618 if ((type == OBJ_BLOB) && path) {
2619 struct strbuf nbuf = STRBUF_INIT;
2620 if (convert_to_git(path, buf, size, &nbuf,
2621 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2622 buf = strbuf_detach(&nbuf, &size);
2623 re_allocated = 1;
2626 if (flags & HASH_FORMAT_CHECK) {
2627 if (type == OBJ_TREE)
2628 check_tree(buf, size);
2629 if (type == OBJ_COMMIT)
2630 check_commit(buf, size);
2631 if (type == OBJ_TAG)
2632 check_tag(buf, size);
2635 if (write_object)
2636 ret = write_sha1_file(buf, size, typename(type), sha1);
2637 else
2638 ret = hash_sha1_file(buf, size, typename(type), sha1);
2639 if (re_allocated)
2640 free(buf);
2641 return ret;
2644 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2645 const char *path, unsigned flags)
2647 struct strbuf sbuf = STRBUF_INIT;
2648 int ret;
2650 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2651 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2652 else
2653 ret = -1;
2654 strbuf_release(&sbuf);
2655 return ret;
2658 #define SMALL_FILE_SIZE (32*1024)
2660 static int index_core(unsigned char *sha1, int fd, size_t size,
2661 enum object_type type, const char *path,
2662 unsigned flags)
2664 int ret;
2666 if (!size) {
2667 ret = index_mem(sha1, NULL, size, type, path, flags);
2668 } else if (size <= SMALL_FILE_SIZE) {
2669 char *buf = xmalloc(size);
2670 if (size == read_in_full(fd, buf, size))
2671 ret = index_mem(sha1, buf, size, type, path, flags);
2672 else
2673 ret = error("short read %s", strerror(errno));
2674 free(buf);
2675 } else {
2676 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2677 ret = index_mem(sha1, buf, size, type, path, flags);
2678 munmap(buf, size);
2680 return ret;
2684 * This creates one packfile per large blob unless bulk-checkin
2685 * machinery is "plugged".
2687 * This also bypasses the usual "convert-to-git" dance, and that is on
2688 * purpose. We could write a streaming version of the converting
2689 * functions and insert that before feeding the data to fast-import
2690 * (or equivalent in-core API described above), but the primary
2691 * motivation for trying to stream from the working tree file and to
2692 * avoid mmaping it in core is to deal with large binary blobs, and
2693 * by definition they do _not_ want to get any conversion.
2695 static int index_stream(unsigned char *sha1, int fd, size_t size,
2696 enum object_type type, const char *path,
2697 unsigned flags)
2699 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2702 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2703 enum object_type type, const char *path, unsigned flags)
2705 int ret;
2706 size_t size = xsize_t(st->st_size);
2708 if (!S_ISREG(st->st_mode))
2709 ret = index_pipe(sha1, fd, type, path, flags);
2710 else if (size <= big_file_threshold || type != OBJ_BLOB)
2711 ret = index_core(sha1, fd, size, type, path, flags);
2712 else
2713 ret = index_stream(sha1, fd, size, type, path, flags);
2714 close(fd);
2715 return ret;
2718 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2720 int fd;
2721 struct strbuf sb = STRBUF_INIT;
2723 switch (st->st_mode & S_IFMT) {
2724 case S_IFREG:
2725 fd = open(path, O_RDONLY);
2726 if (fd < 0)
2727 return error("open(\"%s\"): %s", path,
2728 strerror(errno));
2729 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2730 return error("%s: failed to insert into database",
2731 path);
2732 break;
2733 case S_IFLNK:
2734 if (strbuf_readlink(&sb, path, st->st_size)) {
2735 char *errstr = strerror(errno);
2736 return error("readlink(\"%s\"): %s", path,
2737 errstr);
2739 if (!(flags & HASH_WRITE_OBJECT))
2740 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2741 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2742 return error("%s: failed to insert into database",
2743 path);
2744 strbuf_release(&sb);
2745 break;
2746 case S_IFDIR:
2747 return resolve_gitlink_ref(path, "HEAD", sha1);
2748 default:
2749 return error("%s: unsupported file type", path);
2751 return 0;
2754 int read_pack_header(int fd, struct pack_header *header)
2756 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2757 /* "eof before pack header was fully read" */
2758 return PH_ERROR_EOF;
2760 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2761 /* "protocol error (pack signature mismatch detected)" */
2762 return PH_ERROR_PACK_SIGNATURE;
2763 if (!pack_version_ok(header->hdr_version))
2764 /* "protocol error (pack version unsupported)" */
2765 return PH_ERROR_PROTOCOL;
2766 return 0;
2769 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2771 enum object_type type = sha1_object_info(sha1, NULL);
2772 if (type < 0)
2773 die("%s is not a valid object", sha1_to_hex(sha1));
2774 if (type != expect)
2775 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2776 typename(expect));