Allow parse_pack_index on temporary files
[git.git] / sha1_file.c
blob9f3f514e713cf7d1e2efdbe5de093671cc6b8fdf
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 "tag.h"
15 #include "tree.h"
16 #include "refs.h"
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
20 #ifndef O_NOATIME
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
23 #else
24 #define O_NOATIME 0
25 #endif
26 #endif
28 #ifdef NO_C99_FORMAT
29 #define SZ_FMT "lu"
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 #else
32 #define SZ_FMT "zu"
33 static size_t sz_fmt(size_t s) { return s; }
34 #endif
36 const unsigned char null_sha1[20];
38 static inline int offset_1st_component(const char *path)
40 if (has_dos_drive_prefix(path))
41 return 2 + (path[2] == '/');
42 return *path == '/';
45 int safe_create_leading_directories(char *path)
47 char *pos = path + offset_1st_component(path);
48 struct stat st;
50 while (pos) {
51 pos = strchr(pos, '/');
52 if (!pos)
53 break;
54 while (*++pos == '/')
56 if (!*pos)
57 break;
58 *--pos = '\0';
59 if (!stat(path, &st)) {
60 /* path exists */
61 if (!S_ISDIR(st.st_mode)) {
62 *pos = '/';
63 return -3;
66 else if (mkdir(path, 0777)) {
67 *pos = '/';
68 return -1;
70 else if (adjust_shared_perm(path)) {
71 *pos = '/';
72 return -2;
74 *pos++ = '/';
76 return 0;
79 int safe_create_leading_directories_const(const char *path)
81 /* path points to cache entries, so xstrdup before messing with it */
82 char *buf = xstrdup(path);
83 int result = safe_create_leading_directories(buf);
84 free(buf);
85 return result;
88 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
90 int i;
91 for (i = 0; i < 20; i++) {
92 static char hex[] = "0123456789abcdef";
93 unsigned int val = sha1[i];
94 char *pos = pathbuf + i*2 + (i > 0);
95 *pos++ = hex[val >> 4];
96 *pos = hex[val & 0xf];
101 * NOTE! This returns a statically allocated buffer, so you have to be
102 * careful about using it. Do an "xstrdup()" if you need to save the
103 * filename.
105 * Also note that this returns the location for creating. Reading
106 * SHA1 file can happen from any alternate directory listed in the
107 * DB_ENVIRONMENT environment variable if it is not found in
108 * the primary object database.
110 char *sha1_file_name(const unsigned char *sha1)
112 static char *name, *base;
114 if (!base) {
115 const char *sha1_file_directory = get_object_directory();
116 int len = strlen(sha1_file_directory);
117 base = xmalloc(len + 60);
118 memcpy(base, sha1_file_directory, len);
119 memset(base+len, 0, 60);
120 base[len] = '/';
121 base[len+3] = '/';
122 name = base + len + 1;
124 fill_sha1_path(name, sha1);
125 return base;
128 static char *sha1_get_pack_name(const unsigned char *sha1,
129 char **name, char **base, const char *which)
131 static const char hex[] = "0123456789abcdef";
132 char *buf;
133 int i;
135 if (!*base) {
136 const char *sha1_file_directory = get_object_directory();
137 int len = strlen(sha1_file_directory);
138 *base = xmalloc(len + 60);
139 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
140 sha1_file_directory, which);
141 *name = *base + len + 11;
144 buf = *name;
146 for (i = 0; i < 20; i++) {
147 unsigned int val = *sha1++;
148 *buf++ = hex[val >> 4];
149 *buf++ = hex[val & 0xf];
152 return *base;
155 char *sha1_pack_name(const unsigned char *sha1)
157 static char *name, *base;
159 return sha1_get_pack_name(sha1, &name, &base, "pack");
162 char *sha1_pack_index_name(const unsigned char *sha1)
164 static char *name, *base;
166 return sha1_get_pack_name(sha1, &name, &base, "idx");
169 struct alternate_object_database *alt_odb_list;
170 static struct alternate_object_database **alt_odb_tail;
172 static void read_info_alternates(const char * alternates, int depth);
175 * Prepare alternate object database registry.
177 * The variable alt_odb_list points at the list of struct
178 * alternate_object_database. The elements on this list come from
179 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
180 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
181 * whose contents is similar to that environment variable but can be
182 * LF separated. Its base points at a statically allocated buffer that
183 * contains "/the/directory/corresponding/to/.git/objects/...", while
184 * its name points just after the slash at the end of ".git/objects/"
185 * in the example above, and has enough space to hold 40-byte hex
186 * SHA1, an extra slash for the first level indirection, and the
187 * terminating NUL.
189 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
191 const char *objdir = get_object_directory();
192 struct alternate_object_database *ent;
193 struct alternate_object_database *alt;
194 /* 43 = 40-byte + 2 '/' + terminating NUL */
195 int pfxlen = len;
196 int entlen = pfxlen + 43;
197 int base_len = -1;
199 if (!is_absolute_path(entry) && relative_base) {
200 /* Relative alt-odb */
201 if (base_len < 0)
202 base_len = strlen(relative_base) + 1;
203 entlen += base_len;
204 pfxlen += base_len;
206 ent = xmalloc(sizeof(*ent) + entlen);
208 if (!is_absolute_path(entry) && relative_base) {
209 memcpy(ent->base, relative_base, base_len - 1);
210 ent->base[base_len - 1] = '/';
211 memcpy(ent->base + base_len, entry, len);
213 else
214 memcpy(ent->base, entry, pfxlen);
216 ent->name = ent->base + pfxlen + 1;
217 ent->base[pfxlen + 3] = '/';
218 ent->base[pfxlen] = ent->base[entlen-1] = 0;
220 /* Detect cases where alternate disappeared */
221 if (!is_directory(ent->base)) {
222 error("object directory %s does not exist; "
223 "check .git/objects/info/alternates.",
224 ent->base);
225 free(ent);
226 return -1;
229 /* Prevent the common mistake of listing the same
230 * thing twice, or object directory itself.
232 for (alt = alt_odb_list; alt; alt = alt->next) {
233 if (!memcmp(ent->base, alt->base, pfxlen)) {
234 free(ent);
235 return -1;
238 if (!memcmp(ent->base, objdir, pfxlen)) {
239 free(ent);
240 return -1;
243 /* add the alternate entry */
244 *alt_odb_tail = ent;
245 alt_odb_tail = &(ent->next);
246 ent->next = NULL;
248 /* recursively add alternates */
249 read_info_alternates(ent->base, depth + 1);
251 ent->base[pfxlen] = '/';
253 return 0;
256 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
257 const char *relative_base, int depth)
259 const char *cp, *last;
261 if (depth > 5) {
262 error("%s: ignoring alternate object stores, nesting too deep.",
263 relative_base);
264 return;
267 last = alt;
268 while (last < ep) {
269 cp = last;
270 if (cp < ep && *cp == '#') {
271 while (cp < ep && *cp != sep)
272 cp++;
273 last = cp + 1;
274 continue;
276 while (cp < ep && *cp != sep)
277 cp++;
278 if (last != cp) {
279 if (!is_absolute_path(last) && depth) {
280 error("%s: ignoring relative alternate object store %s",
281 relative_base, last);
282 } else {
283 link_alt_odb_entry(last, cp - last,
284 relative_base, depth);
287 while (cp < ep && *cp == sep)
288 cp++;
289 last = cp;
293 static void read_info_alternates(const char * relative_base, int depth)
295 char *map;
296 size_t mapsz;
297 struct stat st;
298 const char alt_file_name[] = "info/alternates";
299 /* Given that relative_base is no longer than PATH_MAX,
300 ensure that "path" has enough space to append "/", the
301 file name, "info/alternates", and a trailing NUL. */
302 char path[PATH_MAX + 1 + sizeof alt_file_name];
303 int fd;
305 sprintf(path, "%s/%s", relative_base, alt_file_name);
306 fd = open(path, O_RDONLY);
307 if (fd < 0)
308 return;
309 if (fstat(fd, &st) || (st.st_size == 0)) {
310 close(fd);
311 return;
313 mapsz = xsize_t(st.st_size);
314 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
315 close(fd);
317 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
319 munmap(map, mapsz);
322 void add_to_alternates_file(const char *reference)
324 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
325 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
326 char *alt = mkpath("%s/objects\n", reference);
327 write_or_die(fd, alt, strlen(alt));
328 if (commit_lock_file(lock))
329 die("could not close alternates file");
330 if (alt_odb_tail)
331 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
334 void foreach_alt_odb(alt_odb_fn fn, void *cb)
336 struct alternate_object_database *ent;
338 prepare_alt_odb();
339 for (ent = alt_odb_list; ent; ent = ent->next)
340 if (fn(ent, cb))
341 return;
344 void prepare_alt_odb(void)
346 const char *alt;
348 if (alt_odb_tail)
349 return;
351 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
352 if (!alt) alt = "";
354 alt_odb_tail = &alt_odb_list;
355 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
357 read_info_alternates(get_object_directory(), 0);
360 static int has_loose_object_local(const unsigned char *sha1)
362 char *name = sha1_file_name(sha1);
363 return !access(name, F_OK);
366 int has_loose_object_nonlocal(const unsigned char *sha1)
368 struct alternate_object_database *alt;
369 prepare_alt_odb();
370 for (alt = alt_odb_list; alt; alt = alt->next) {
371 fill_sha1_path(alt->name, sha1);
372 if (!access(alt->base, F_OK))
373 return 1;
375 return 0;
378 static int has_loose_object(const unsigned char *sha1)
380 return has_loose_object_local(sha1) ||
381 has_loose_object_nonlocal(sha1);
384 static unsigned int pack_used_ctr;
385 static unsigned int pack_mmap_calls;
386 static unsigned int peak_pack_open_windows;
387 static unsigned int pack_open_windows;
388 static size_t peak_pack_mapped;
389 static size_t pack_mapped;
390 struct packed_git *packed_git;
392 void pack_report(void)
394 fprintf(stderr,
395 "pack_report: getpagesize() = %10" SZ_FMT "\n"
396 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
397 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
398 sz_fmt(getpagesize()),
399 sz_fmt(packed_git_window_size),
400 sz_fmt(packed_git_limit));
401 fprintf(stderr,
402 "pack_report: pack_used_ctr = %10u\n"
403 "pack_report: pack_mmap_calls = %10u\n"
404 "pack_report: pack_open_windows = %10u / %10u\n"
405 "pack_report: pack_mapped = "
406 "%10" SZ_FMT " / %10" SZ_FMT "\n",
407 pack_used_ctr,
408 pack_mmap_calls,
409 pack_open_windows, peak_pack_open_windows,
410 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
413 static int check_packed_git_idx(const char *path, struct packed_git *p)
415 void *idx_map;
416 struct pack_idx_header *hdr;
417 size_t idx_size;
418 uint32_t version, nr, i, *index;
419 int fd = open(path, O_RDONLY);
420 struct stat st;
422 if (fd < 0)
423 return -1;
424 if (fstat(fd, &st)) {
425 close(fd);
426 return -1;
428 idx_size = xsize_t(st.st_size);
429 if (idx_size < 4 * 256 + 20 + 20) {
430 close(fd);
431 return error("index file %s is too small", path);
433 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
434 close(fd);
436 hdr = idx_map;
437 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
438 version = ntohl(hdr->idx_version);
439 if (version < 2 || version > 2) {
440 munmap(idx_map, idx_size);
441 return error("index file %s is version %"PRIu32
442 " and is not supported by this binary"
443 " (try upgrading GIT to a newer version)",
444 path, version);
446 } else
447 version = 1;
449 nr = 0;
450 index = idx_map;
451 if (version > 1)
452 index += 2; /* skip index header */
453 for (i = 0; i < 256; i++) {
454 uint32_t n = ntohl(index[i]);
455 if (n < nr) {
456 munmap(idx_map, idx_size);
457 return error("non-monotonic index %s", path);
459 nr = n;
462 if (version == 1) {
464 * Total size:
465 * - 256 index entries 4 bytes each
466 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
467 * - 20-byte SHA1 of the packfile
468 * - 20-byte SHA1 file checksum
470 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
471 munmap(idx_map, idx_size);
472 return error("wrong index v1 file size in %s", path);
474 } else if (version == 2) {
476 * Minimum size:
477 * - 8 bytes of header
478 * - 256 index entries 4 bytes each
479 * - 20-byte sha1 entry * nr
480 * - 4-byte crc entry * nr
481 * - 4-byte offset entry * nr
482 * - 20-byte SHA1 of the packfile
483 * - 20-byte SHA1 file checksum
484 * And after the 4-byte offset table might be a
485 * variable sized table containing 8-byte entries
486 * for offsets larger than 2^31.
488 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
489 unsigned long max_size = min_size;
490 if (nr)
491 max_size += (nr - 1)*8;
492 if (idx_size < min_size || idx_size > max_size) {
493 munmap(idx_map, idx_size);
494 return error("wrong index v2 file size in %s", path);
496 if (idx_size != min_size &&
498 * make sure we can deal with large pack offsets.
499 * 31-bit signed offset won't be enough, neither
500 * 32-bit unsigned one will be.
502 (sizeof(off_t) <= 4)) {
503 munmap(idx_map, idx_size);
504 return error("pack too large for current definition of off_t in %s", path);
508 p->index_version = version;
509 p->index_data = idx_map;
510 p->index_size = idx_size;
511 p->num_objects = nr;
512 return 0;
515 int open_pack_index(struct packed_git *p)
517 char *idx_name;
518 int ret;
520 if (p->index_data)
521 return 0;
523 idx_name = xstrdup(p->pack_name);
524 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
525 ret = check_packed_git_idx(idx_name, p);
526 free(idx_name);
527 return ret;
530 static void scan_windows(struct packed_git *p,
531 struct packed_git **lru_p,
532 struct pack_window **lru_w,
533 struct pack_window **lru_l)
535 struct pack_window *w, *w_l;
537 for (w_l = NULL, w = p->windows; w; w = w->next) {
538 if (!w->inuse_cnt) {
539 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
540 *lru_p = p;
541 *lru_w = w;
542 *lru_l = w_l;
545 w_l = w;
549 static int unuse_one_window(struct packed_git *current, int keep_fd)
551 struct packed_git *p, *lru_p = NULL;
552 struct pack_window *lru_w = NULL, *lru_l = NULL;
554 if (current)
555 scan_windows(current, &lru_p, &lru_w, &lru_l);
556 for (p = packed_git; p; p = p->next)
557 scan_windows(p, &lru_p, &lru_w, &lru_l);
558 if (lru_p) {
559 munmap(lru_w->base, lru_w->len);
560 pack_mapped -= lru_w->len;
561 if (lru_l)
562 lru_l->next = lru_w->next;
563 else {
564 lru_p->windows = lru_w->next;
565 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
566 close(lru_p->pack_fd);
567 lru_p->pack_fd = -1;
570 free(lru_w);
571 pack_open_windows--;
572 return 1;
574 return 0;
577 void release_pack_memory(size_t need, int fd)
579 size_t cur = pack_mapped;
580 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
581 ; /* nothing */
584 void close_pack_windows(struct packed_git *p)
586 while (p->windows) {
587 struct pack_window *w = p->windows;
589 if (w->inuse_cnt)
590 die("pack '%s' still has open windows to it",
591 p->pack_name);
592 munmap(w->base, w->len);
593 pack_mapped -= w->len;
594 pack_open_windows--;
595 p->windows = w->next;
596 free(w);
600 void unuse_pack(struct pack_window **w_cursor)
602 struct pack_window *w = *w_cursor;
603 if (w) {
604 w->inuse_cnt--;
605 *w_cursor = NULL;
609 void close_pack_index(struct packed_git *p)
611 if (p->index_data) {
612 munmap((void *)p->index_data, p->index_size);
613 p->index_data = NULL;
618 * This is used by git-repack in case a newly created pack happens to
619 * contain the same set of objects as an existing one. In that case
620 * the resulting file might be different even if its name would be the
621 * same. It is best to close any reference to the old pack before it is
622 * replaced on disk. Of course no index pointers nor windows for given pack
623 * must subsist at this point. If ever objects from this pack are requested
624 * again, the new version of the pack will be reinitialized through
625 * reprepare_packed_git().
627 void free_pack_by_name(const char *pack_name)
629 struct packed_git *p, **pp = &packed_git;
631 while (*pp) {
632 p = *pp;
633 if (strcmp(pack_name, p->pack_name) == 0) {
634 clear_delta_base_cache();
635 close_pack_windows(p);
636 if (p->pack_fd != -1)
637 close(p->pack_fd);
638 close_pack_index(p);
639 free(p->bad_object_sha1);
640 *pp = p->next;
641 free(p);
642 return;
644 pp = &p->next;
649 * Do not call this directly as this leaks p->pack_fd on error return;
650 * call open_packed_git() instead.
652 static int open_packed_git_1(struct packed_git *p)
654 struct stat st;
655 struct pack_header hdr;
656 unsigned char sha1[20];
657 unsigned char *idx_sha1;
658 long fd_flag;
660 if (!p->index_data && open_pack_index(p))
661 return error("packfile %s index unavailable", p->pack_name);
663 p->pack_fd = open(p->pack_name, O_RDONLY);
664 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
665 p->pack_fd = open(p->pack_name, O_RDONLY);
666 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
667 return -1;
669 /* If we created the struct before we had the pack we lack size. */
670 if (!p->pack_size) {
671 if (!S_ISREG(st.st_mode))
672 return error("packfile %s not a regular file", p->pack_name);
673 p->pack_size = st.st_size;
674 } else if (p->pack_size != st.st_size)
675 return error("packfile %s size changed", p->pack_name);
677 /* We leave these file descriptors open with sliding mmap;
678 * there is no point keeping them open across exec(), though.
680 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
681 if (fd_flag < 0)
682 return error("cannot determine file descriptor flags");
683 fd_flag |= FD_CLOEXEC;
684 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
685 return error("cannot set FD_CLOEXEC");
687 /* Verify we recognize this pack file format. */
688 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
689 return error("file %s is far too short to be a packfile", p->pack_name);
690 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
691 return error("file %s is not a GIT packfile", p->pack_name);
692 if (!pack_version_ok(hdr.hdr_version))
693 return error("packfile %s is version %"PRIu32" and not"
694 " supported (try upgrading GIT to a newer version)",
695 p->pack_name, ntohl(hdr.hdr_version));
697 /* Verify the pack matches its index. */
698 if (p->num_objects != ntohl(hdr.hdr_entries))
699 return error("packfile %s claims to have %"PRIu32" objects"
700 " while index indicates %"PRIu32" objects",
701 p->pack_name, ntohl(hdr.hdr_entries),
702 p->num_objects);
703 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
704 return error("end of packfile %s is unavailable", p->pack_name);
705 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
706 return error("packfile %s signature is unavailable", p->pack_name);
707 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
708 if (hashcmp(sha1, idx_sha1))
709 return error("packfile %s does not match index", p->pack_name);
710 return 0;
713 static int open_packed_git(struct packed_git *p)
715 if (!open_packed_git_1(p))
716 return 0;
717 if (p->pack_fd != -1) {
718 close(p->pack_fd);
719 p->pack_fd = -1;
721 return -1;
724 static int in_window(struct pack_window *win, off_t offset)
726 /* We must promise at least 20 bytes (one hash) after the
727 * offset is available from this window, otherwise the offset
728 * is not actually in this window and a different window (which
729 * has that one hash excess) must be used. This is to support
730 * the object header and delta base parsing routines below.
732 off_t win_off = win->offset;
733 return win_off <= offset
734 && (offset + 20) <= (win_off + win->len);
737 unsigned char *use_pack(struct packed_git *p,
738 struct pack_window **w_cursor,
739 off_t offset,
740 unsigned int *left)
742 struct pack_window *win = *w_cursor;
744 if (p->pack_fd == -1 && open_packed_git(p))
745 die("packfile %s cannot be accessed", p->pack_name);
747 /* Since packfiles end in a hash of their content and it's
748 * pointless to ask for an offset into the middle of that
749 * hash, and the in_window function above wouldn't match
750 * don't allow an offset too close to the end of the file.
752 if (offset > (p->pack_size - 20))
753 die("offset beyond end of packfile (truncated pack?)");
755 if (!win || !in_window(win, offset)) {
756 if (win)
757 win->inuse_cnt--;
758 for (win = p->windows; win; win = win->next) {
759 if (in_window(win, offset))
760 break;
762 if (!win) {
763 size_t window_align = packed_git_window_size / 2;
764 off_t len;
765 win = xcalloc(1, sizeof(*win));
766 win->offset = (offset / window_align) * window_align;
767 len = p->pack_size - win->offset;
768 if (len > packed_git_window_size)
769 len = packed_git_window_size;
770 win->len = (size_t)len;
771 pack_mapped += win->len;
772 while (packed_git_limit < pack_mapped
773 && unuse_one_window(p, p->pack_fd))
774 ; /* nothing */
775 win->base = xmmap(NULL, win->len,
776 PROT_READ, MAP_PRIVATE,
777 p->pack_fd, win->offset);
778 if (win->base == MAP_FAILED)
779 die("packfile %s cannot be mapped: %s",
780 p->pack_name,
781 strerror(errno));
782 pack_mmap_calls++;
783 pack_open_windows++;
784 if (pack_mapped > peak_pack_mapped)
785 peak_pack_mapped = pack_mapped;
786 if (pack_open_windows > peak_pack_open_windows)
787 peak_pack_open_windows = pack_open_windows;
788 win->next = p->windows;
789 p->windows = win;
792 if (win != *w_cursor) {
793 win->last_used = pack_used_ctr++;
794 win->inuse_cnt++;
795 *w_cursor = win;
797 offset -= win->offset;
798 if (left)
799 *left = win->len - xsize_t(offset);
800 return win->base + offset;
803 static struct packed_git *alloc_packed_git(int extra)
805 struct packed_git *p = xmalloc(sizeof(*p) + extra);
806 memset(p, 0, sizeof(*p));
807 p->pack_fd = -1;
808 return p;
811 struct packed_git *add_packed_git(const char *path, int path_len, int local)
813 struct stat st;
814 struct packed_git *p = alloc_packed_git(path_len + 2);
817 * Make sure a corresponding .pack file exists and that
818 * the index looks sane.
820 path_len -= strlen(".idx");
821 if (path_len < 1) {
822 free(p);
823 return NULL;
825 memcpy(p->pack_name, path, path_len);
827 strcpy(p->pack_name + path_len, ".keep");
828 if (!access(p->pack_name, F_OK))
829 p->pack_keep = 1;
831 strcpy(p->pack_name + path_len, ".pack");
832 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
833 free(p);
834 return NULL;
837 /* ok, it looks sane as far as we can check without
838 * actually mapping the pack file.
840 p->pack_size = st.st_size;
841 p->pack_local = local;
842 p->mtime = st.st_mtime;
843 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
844 hashclr(p->sha1);
845 return p;
848 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
850 const char *path = sha1_pack_name(sha1);
851 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
853 strcpy(p->pack_name, path);
854 hashcpy(p->sha1, sha1);
855 if (check_packed_git_idx(idx_path, p)) {
856 free(p);
857 return NULL;
860 return p;
863 void install_packed_git(struct packed_git *pack)
865 pack->next = packed_git;
866 packed_git = pack;
869 static void prepare_packed_git_one(char *objdir, int local)
871 /* Ensure that this buffer is large enough so that we can
872 append "/pack/" without clobbering the stack even if
873 strlen(objdir) were PATH_MAX. */
874 char path[PATH_MAX + 1 + 4 + 1 + 1];
875 int len;
876 DIR *dir;
877 struct dirent *de;
879 sprintf(path, "%s/pack", objdir);
880 len = strlen(path);
881 dir = opendir(path);
882 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
883 dir = opendir(path);
884 if (!dir) {
885 if (errno != ENOENT)
886 error("unable to open object pack directory: %s: %s",
887 path, strerror(errno));
888 return;
890 path[len++] = '/';
891 while ((de = readdir(dir)) != NULL) {
892 int namelen = strlen(de->d_name);
893 struct packed_git *p;
895 if (!has_extension(de->d_name, ".idx"))
896 continue;
898 if (len + namelen + 1 > sizeof(path))
899 continue;
901 /* Don't reopen a pack we already have. */
902 strcpy(path + len, de->d_name);
903 for (p = packed_git; p; p = p->next) {
904 if (!memcmp(path, p->pack_name, len + namelen - 4))
905 break;
907 if (p)
908 continue;
909 /* See if it really is a valid .idx file with corresponding
910 * .pack file that we can map.
912 p = add_packed_git(path, len + namelen, local);
913 if (!p)
914 continue;
915 install_packed_git(p);
917 closedir(dir);
920 static int sort_pack(const void *a_, const void *b_)
922 struct packed_git *a = *((struct packed_git **)a_);
923 struct packed_git *b = *((struct packed_git **)b_);
924 int st;
927 * Local packs tend to contain objects specific to our
928 * variant of the project than remote ones. In addition,
929 * remote ones could be on a network mounted filesystem.
930 * Favor local ones for these reasons.
932 st = a->pack_local - b->pack_local;
933 if (st)
934 return -st;
937 * Younger packs tend to contain more recent objects,
938 * and more recent objects tend to get accessed more
939 * often.
941 if (a->mtime < b->mtime)
942 return 1;
943 else if (a->mtime == b->mtime)
944 return 0;
945 return -1;
948 static void rearrange_packed_git(void)
950 struct packed_git **ary, *p;
951 int i, n;
953 for (n = 0, p = packed_git; p; p = p->next)
954 n++;
955 if (n < 2)
956 return;
958 /* prepare an array of packed_git for easier sorting */
959 ary = xcalloc(n, sizeof(struct packed_git *));
960 for (n = 0, p = packed_git; p; p = p->next)
961 ary[n++] = p;
963 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
965 /* link them back again */
966 for (i = 0; i < n - 1; i++)
967 ary[i]->next = ary[i + 1];
968 ary[n - 1]->next = NULL;
969 packed_git = ary[0];
971 free(ary);
974 static int prepare_packed_git_run_once = 0;
975 void prepare_packed_git(void)
977 struct alternate_object_database *alt;
979 if (prepare_packed_git_run_once)
980 return;
981 prepare_packed_git_one(get_object_directory(), 1);
982 prepare_alt_odb();
983 for (alt = alt_odb_list; alt; alt = alt->next) {
984 alt->name[-1] = 0;
985 prepare_packed_git_one(alt->base, 0);
986 alt->name[-1] = '/';
988 rearrange_packed_git();
989 prepare_packed_git_run_once = 1;
992 void reprepare_packed_git(void)
994 discard_revindex();
995 prepare_packed_git_run_once = 0;
996 prepare_packed_git();
999 static void mark_bad_packed_object(struct packed_git *p,
1000 const unsigned char *sha1)
1002 unsigned i;
1003 for (i = 0; i < p->num_bad_objects; i++)
1004 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1005 return;
1006 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1007 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1008 p->num_bad_objects++;
1011 static int has_packed_and_bad(const unsigned char *sha1)
1013 struct packed_git *p;
1014 unsigned i;
1016 for (p = packed_git; p; p = p->next)
1017 for (i = 0; i < p->num_bad_objects; i++)
1018 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1019 return 1;
1020 return 0;
1023 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1025 unsigned char real_sha1[20];
1026 hash_sha1_file(map, size, type, real_sha1);
1027 return hashcmp(sha1, real_sha1) ? -1 : 0;
1030 static int git_open_noatime(const char *name)
1032 static int sha1_file_open_flag = O_NOATIME;
1033 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1035 /* Might the failure be due to O_NOATIME? */
1036 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1037 fd = open(name, O_RDONLY);
1038 if (fd >= 0)
1039 sha1_file_open_flag = 0;
1041 return fd;
1044 static int open_sha1_file(const unsigned char *sha1)
1046 int fd;
1047 char *name = sha1_file_name(sha1);
1048 struct alternate_object_database *alt;
1050 fd = git_open_noatime(name);
1051 if (fd >= 0)
1052 return fd;
1054 prepare_alt_odb();
1055 errno = ENOENT;
1056 for (alt = alt_odb_list; alt; alt = alt->next) {
1057 name = alt->name;
1058 fill_sha1_path(name, sha1);
1059 fd = git_open_noatime(alt->base);
1060 if (fd >= 0)
1061 return fd;
1063 return -1;
1066 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1068 void *map;
1069 int fd;
1071 fd = open_sha1_file(sha1);
1072 map = NULL;
1073 if (fd >= 0) {
1074 struct stat st;
1076 if (!fstat(fd, &st)) {
1077 *size = xsize_t(st.st_size);
1078 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1080 close(fd);
1082 return map;
1085 static int legacy_loose_object(unsigned char *map)
1087 unsigned int word;
1090 * Is it a zlib-compressed buffer? If so, the first byte
1091 * must be 0x78 (15-bit window size, deflated), and the
1092 * first 16-bit word is evenly divisible by 31
1094 word = (map[0] << 8) + map[1];
1095 if (map[0] == 0x78 && !(word % 31))
1096 return 1;
1097 else
1098 return 0;
1101 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1102 unsigned long len, enum object_type *type, unsigned long *sizep)
1104 unsigned shift;
1105 unsigned long size, c;
1106 unsigned long used = 0;
1108 c = buf[used++];
1109 *type = (c >> 4) & 7;
1110 size = c & 15;
1111 shift = 4;
1112 while (c & 0x80) {
1113 if (len <= used || bitsizeof(long) <= shift) {
1114 error("bad object header");
1115 return 0;
1117 c = buf[used++];
1118 size += (c & 0x7f) << shift;
1119 shift += 7;
1121 *sizep = size;
1122 return used;
1125 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1127 unsigned long size, used;
1128 static const char valid_loose_object_type[8] = {
1129 0, /* OBJ_EXT */
1130 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1131 0, /* "delta" and others are invalid in a loose object */
1133 enum object_type type;
1135 /* Get the data stream */
1136 memset(stream, 0, sizeof(*stream));
1137 stream->next_in = map;
1138 stream->avail_in = mapsize;
1139 stream->next_out = buffer;
1140 stream->avail_out = bufsiz;
1142 if (legacy_loose_object(map)) {
1143 git_inflate_init(stream);
1144 return git_inflate(stream, 0);
1149 * There used to be a second loose object header format which
1150 * was meant to mimic the in-pack format, allowing for direct
1151 * copy of the object data. This format turned up not to be
1152 * really worth it and we don't write it any longer. But we
1153 * can still read it.
1155 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1156 if (!used || !valid_loose_object_type[type])
1157 return -1;
1158 map += used;
1159 mapsize -= used;
1161 /* Set up the stream for the rest.. */
1162 stream->next_in = map;
1163 stream->avail_in = mapsize;
1164 git_inflate_init(stream);
1166 /* And generate the fake traditional header */
1167 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1168 typename(type), size);
1169 return 0;
1172 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1174 int bytes = strlen(buffer) + 1;
1175 unsigned char *buf = xmallocz(size);
1176 unsigned long n;
1177 int status = Z_OK;
1179 n = stream->total_out - bytes;
1180 if (n > size)
1181 n = size;
1182 memcpy(buf, (char *) buffer + bytes, n);
1183 bytes = n;
1184 if (bytes <= size) {
1186 * The above condition must be (bytes <= size), not
1187 * (bytes < size). In other words, even though we
1188 * expect no more output and set avail_out to zer0,
1189 * the input zlib stream may have bytes that express
1190 * "this concludes the stream", and we *do* want to
1191 * eat that input.
1193 * Otherwise we would not be able to test that we
1194 * consumed all the input to reach the expected size;
1195 * we also want to check that zlib tells us that all
1196 * went well with status == Z_STREAM_END at the end.
1198 stream->next_out = buf + bytes;
1199 stream->avail_out = size - bytes;
1200 while (status == Z_OK)
1201 status = git_inflate(stream, Z_FINISH);
1203 if (status == Z_STREAM_END && !stream->avail_in) {
1204 git_inflate_end(stream);
1205 return buf;
1208 if (status < 0)
1209 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1210 else if (stream->avail_in)
1211 error("garbage at end of loose object '%s'",
1212 sha1_to_hex(sha1));
1213 free(buf);
1214 return NULL;
1218 * We used to just use "sscanf()", but that's actually way
1219 * too permissive for what we want to check. So do an anal
1220 * object header parse by hand.
1222 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1224 char type[10];
1225 int i;
1226 unsigned long size;
1229 * The type can be at most ten bytes (including the
1230 * terminating '\0' that we add), and is followed by
1231 * a space.
1233 i = 0;
1234 for (;;) {
1235 char c = *hdr++;
1236 if (c == ' ')
1237 break;
1238 type[i++] = c;
1239 if (i >= sizeof(type))
1240 return -1;
1242 type[i] = 0;
1245 * The length must follow immediately, and be in canonical
1246 * decimal format (ie "010" is not valid).
1248 size = *hdr++ - '0';
1249 if (size > 9)
1250 return -1;
1251 if (size) {
1252 for (;;) {
1253 unsigned long c = *hdr - '0';
1254 if (c > 9)
1255 break;
1256 hdr++;
1257 size = size * 10 + c;
1260 *sizep = size;
1263 * The length must be followed by a zero byte
1265 return *hdr ? -1 : type_from_string(type);
1268 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1270 int ret;
1271 z_stream stream;
1272 char hdr[8192];
1274 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1275 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1276 return NULL;
1278 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1281 unsigned long get_size_from_delta(struct packed_git *p,
1282 struct pack_window **w_curs,
1283 off_t curpos)
1285 const unsigned char *data;
1286 unsigned char delta_head[20], *in;
1287 z_stream stream;
1288 int st;
1290 memset(&stream, 0, sizeof(stream));
1291 stream.next_out = delta_head;
1292 stream.avail_out = sizeof(delta_head);
1294 git_inflate_init(&stream);
1295 do {
1296 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1297 stream.next_in = in;
1298 st = git_inflate(&stream, Z_FINISH);
1299 curpos += stream.next_in - in;
1300 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1301 stream.total_out < sizeof(delta_head));
1302 git_inflate_end(&stream);
1303 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1304 error("delta data unpack-initial failed");
1305 return 0;
1308 /* Examine the initial part of the delta to figure out
1309 * the result size.
1311 data = delta_head;
1313 /* ignore base size */
1314 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1316 /* Read the result size */
1317 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1320 static off_t get_delta_base(struct packed_git *p,
1321 struct pack_window **w_curs,
1322 off_t *curpos,
1323 enum object_type type,
1324 off_t delta_obj_offset)
1326 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1327 off_t base_offset;
1329 /* use_pack() assured us we have [base_info, base_info + 20)
1330 * as a range that we can look at without walking off the
1331 * end of the mapped window. Its actually the hash size
1332 * that is assured. An OFS_DELTA longer than the hash size
1333 * is stupid, as then a REF_DELTA would be smaller to store.
1335 if (type == OBJ_OFS_DELTA) {
1336 unsigned used = 0;
1337 unsigned char c = base_info[used++];
1338 base_offset = c & 127;
1339 while (c & 128) {
1340 base_offset += 1;
1341 if (!base_offset || MSB(base_offset, 7))
1342 return 0; /* overflow */
1343 c = base_info[used++];
1344 base_offset = (base_offset << 7) + (c & 127);
1346 base_offset = delta_obj_offset - base_offset;
1347 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1348 return 0; /* out of bound */
1349 *curpos += used;
1350 } else if (type == OBJ_REF_DELTA) {
1351 /* The base entry _must_ be in the same pack */
1352 base_offset = find_pack_entry_one(base_info, p);
1353 *curpos += 20;
1354 } else
1355 die("I am totally screwed");
1356 return base_offset;
1359 /* forward declaration for a mutually recursive function */
1360 static int packed_object_info(struct packed_git *p, off_t offset,
1361 unsigned long *sizep);
1363 static int packed_delta_info(struct packed_git *p,
1364 struct pack_window **w_curs,
1365 off_t curpos,
1366 enum object_type type,
1367 off_t obj_offset,
1368 unsigned long *sizep)
1370 off_t base_offset;
1372 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1373 if (!base_offset)
1374 return OBJ_BAD;
1375 type = packed_object_info(p, base_offset, NULL);
1376 if (type <= OBJ_NONE) {
1377 struct revindex_entry *revidx;
1378 const unsigned char *base_sha1;
1379 revidx = find_pack_revindex(p, base_offset);
1380 if (!revidx)
1381 return OBJ_BAD;
1382 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1383 mark_bad_packed_object(p, base_sha1);
1384 type = sha1_object_info(base_sha1, NULL);
1385 if (type <= OBJ_NONE)
1386 return OBJ_BAD;
1389 /* We choose to only get the type of the base object and
1390 * ignore potentially corrupt pack file that expects the delta
1391 * based on a base with a wrong size. This saves tons of
1392 * inflate() calls.
1394 if (sizep) {
1395 *sizep = get_size_from_delta(p, w_curs, curpos);
1396 if (*sizep == 0)
1397 type = OBJ_BAD;
1400 return type;
1403 static int unpack_object_header(struct packed_git *p,
1404 struct pack_window **w_curs,
1405 off_t *curpos,
1406 unsigned long *sizep)
1408 unsigned char *base;
1409 unsigned int left;
1410 unsigned long used;
1411 enum object_type type;
1413 /* use_pack() assures us we have [base, base + 20) available
1414 * as a range that we can look at at. (Its actually the hash
1415 * size that is assured.) With our object header encoding
1416 * the maximum deflated object size is 2^137, which is just
1417 * insane, so we know won't exceed what we have been given.
1419 base = use_pack(p, w_curs, *curpos, &left);
1420 used = unpack_object_header_buffer(base, left, &type, sizep);
1421 if (!used) {
1422 type = OBJ_BAD;
1423 } else
1424 *curpos += used;
1426 return type;
1429 const char *packed_object_info_detail(struct packed_git *p,
1430 off_t obj_offset,
1431 unsigned long *size,
1432 unsigned long *store_size,
1433 unsigned int *delta_chain_length,
1434 unsigned char *base_sha1)
1436 struct pack_window *w_curs = NULL;
1437 off_t curpos;
1438 unsigned long dummy;
1439 unsigned char *next_sha1;
1440 enum object_type type;
1441 struct revindex_entry *revidx;
1443 *delta_chain_length = 0;
1444 curpos = obj_offset;
1445 type = unpack_object_header(p, &w_curs, &curpos, size);
1447 revidx = find_pack_revindex(p, obj_offset);
1448 *store_size = revidx[1].offset - obj_offset;
1450 for (;;) {
1451 switch (type) {
1452 default:
1453 die("pack %s contains unknown object type %d",
1454 p->pack_name, type);
1455 case OBJ_COMMIT:
1456 case OBJ_TREE:
1457 case OBJ_BLOB:
1458 case OBJ_TAG:
1459 unuse_pack(&w_curs);
1460 return typename(type);
1461 case OBJ_OFS_DELTA:
1462 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1463 if (!obj_offset)
1464 die("pack %s contains bad delta base reference of type %s",
1465 p->pack_name, typename(type));
1466 if (*delta_chain_length == 0) {
1467 revidx = find_pack_revindex(p, obj_offset);
1468 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1470 break;
1471 case OBJ_REF_DELTA:
1472 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1473 if (*delta_chain_length == 0)
1474 hashcpy(base_sha1, next_sha1);
1475 obj_offset = find_pack_entry_one(next_sha1, p);
1476 break;
1478 (*delta_chain_length)++;
1479 curpos = obj_offset;
1480 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1484 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1485 unsigned long *sizep)
1487 struct pack_window *w_curs = NULL;
1488 unsigned long size;
1489 off_t curpos = obj_offset;
1490 enum object_type type;
1492 type = unpack_object_header(p, &w_curs, &curpos, &size);
1494 switch (type) {
1495 case OBJ_OFS_DELTA:
1496 case OBJ_REF_DELTA:
1497 type = packed_delta_info(p, &w_curs, curpos,
1498 type, obj_offset, sizep);
1499 break;
1500 case OBJ_COMMIT:
1501 case OBJ_TREE:
1502 case OBJ_BLOB:
1503 case OBJ_TAG:
1504 if (sizep)
1505 *sizep = size;
1506 break;
1507 default:
1508 error("unknown object type %i at offset %"PRIuMAX" in %s",
1509 type, (uintmax_t)obj_offset, p->pack_name);
1510 type = OBJ_BAD;
1512 unuse_pack(&w_curs);
1513 return type;
1516 static void *unpack_compressed_entry(struct packed_git *p,
1517 struct pack_window **w_curs,
1518 off_t curpos,
1519 unsigned long size)
1521 int st;
1522 z_stream stream;
1523 unsigned char *buffer, *in;
1525 buffer = xmallocz(size);
1526 memset(&stream, 0, sizeof(stream));
1527 stream.next_out = buffer;
1528 stream.avail_out = size + 1;
1530 git_inflate_init(&stream);
1531 do {
1532 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1533 stream.next_in = in;
1534 st = git_inflate(&stream, Z_FINISH);
1535 if (!stream.avail_out)
1536 break; /* the payload is larger than it should be */
1537 curpos += stream.next_in - in;
1538 } while (st == Z_OK || st == Z_BUF_ERROR);
1539 git_inflate_end(&stream);
1540 if ((st != Z_STREAM_END) || stream.total_out != size) {
1541 free(buffer);
1542 return NULL;
1545 return buffer;
1548 #define MAX_DELTA_CACHE (256)
1550 static size_t delta_base_cached;
1552 static struct delta_base_cache_lru_list {
1553 struct delta_base_cache_lru_list *prev;
1554 struct delta_base_cache_lru_list *next;
1555 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1557 static struct delta_base_cache_entry {
1558 struct delta_base_cache_lru_list lru;
1559 void *data;
1560 struct packed_git *p;
1561 off_t base_offset;
1562 unsigned long size;
1563 enum object_type type;
1564 } delta_base_cache[MAX_DELTA_CACHE];
1566 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1568 unsigned long hash;
1570 hash = (unsigned long)p + (unsigned long)base_offset;
1571 hash += (hash >> 8) + (hash >> 16);
1572 return hash % MAX_DELTA_CACHE;
1575 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1576 unsigned long *base_size, enum object_type *type, int keep_cache)
1578 void *ret;
1579 unsigned long hash = pack_entry_hash(p, base_offset);
1580 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1582 ret = ent->data;
1583 if (!ret || ent->p != p || ent->base_offset != base_offset)
1584 return unpack_entry(p, base_offset, type, base_size);
1586 if (!keep_cache) {
1587 ent->data = NULL;
1588 ent->lru.next->prev = ent->lru.prev;
1589 ent->lru.prev->next = ent->lru.next;
1590 delta_base_cached -= ent->size;
1591 } else {
1592 ret = xmemdupz(ent->data, ent->size);
1594 *type = ent->type;
1595 *base_size = ent->size;
1596 return ret;
1599 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1601 if (ent->data) {
1602 free(ent->data);
1603 ent->data = NULL;
1604 ent->lru.next->prev = ent->lru.prev;
1605 ent->lru.prev->next = ent->lru.next;
1606 delta_base_cached -= ent->size;
1610 void clear_delta_base_cache(void)
1612 unsigned long p;
1613 for (p = 0; p < MAX_DELTA_CACHE; p++)
1614 release_delta_base_cache(&delta_base_cache[p]);
1617 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1618 void *base, unsigned long base_size, enum object_type type)
1620 unsigned long hash = pack_entry_hash(p, base_offset);
1621 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1622 struct delta_base_cache_lru_list *lru;
1624 release_delta_base_cache(ent);
1625 delta_base_cached += base_size;
1627 for (lru = delta_base_cache_lru.next;
1628 delta_base_cached > delta_base_cache_limit
1629 && lru != &delta_base_cache_lru;
1630 lru = lru->next) {
1631 struct delta_base_cache_entry *f = (void *)lru;
1632 if (f->type == OBJ_BLOB)
1633 release_delta_base_cache(f);
1635 for (lru = delta_base_cache_lru.next;
1636 delta_base_cached > delta_base_cache_limit
1637 && lru != &delta_base_cache_lru;
1638 lru = lru->next) {
1639 struct delta_base_cache_entry *f = (void *)lru;
1640 release_delta_base_cache(f);
1643 ent->p = p;
1644 ent->base_offset = base_offset;
1645 ent->type = type;
1646 ent->data = base;
1647 ent->size = base_size;
1648 ent->lru.next = &delta_base_cache_lru;
1649 ent->lru.prev = delta_base_cache_lru.prev;
1650 delta_base_cache_lru.prev->next = &ent->lru;
1651 delta_base_cache_lru.prev = &ent->lru;
1654 static void *read_object(const unsigned char *sha1, enum object_type *type,
1655 unsigned long *size);
1657 static void *unpack_delta_entry(struct packed_git *p,
1658 struct pack_window **w_curs,
1659 off_t curpos,
1660 unsigned long delta_size,
1661 off_t obj_offset,
1662 enum object_type *type,
1663 unsigned long *sizep)
1665 void *delta_data, *result, *base;
1666 unsigned long base_size;
1667 off_t base_offset;
1669 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1670 if (!base_offset) {
1671 error("failed to validate delta base reference "
1672 "at offset %"PRIuMAX" from %s",
1673 (uintmax_t)curpos, p->pack_name);
1674 return NULL;
1676 unuse_pack(w_curs);
1677 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1678 if (!base) {
1680 * We're probably in deep shit, but let's try to fetch
1681 * the required base anyway from another pack or loose.
1682 * This is costly but should happen only in the presence
1683 * of a corrupted pack, and is better than failing outright.
1685 struct revindex_entry *revidx;
1686 const unsigned char *base_sha1;
1687 revidx = find_pack_revindex(p, base_offset);
1688 if (!revidx)
1689 return NULL;
1690 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1691 error("failed to read delta base object %s"
1692 " at offset %"PRIuMAX" from %s",
1693 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1694 p->pack_name);
1695 mark_bad_packed_object(p, base_sha1);
1696 base = read_object(base_sha1, type, &base_size);
1697 if (!base)
1698 return NULL;
1701 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1702 if (!delta_data) {
1703 error("failed to unpack compressed delta "
1704 "at offset %"PRIuMAX" from %s",
1705 (uintmax_t)curpos, p->pack_name);
1706 free(base);
1707 return NULL;
1709 result = patch_delta(base, base_size,
1710 delta_data, delta_size,
1711 sizep);
1712 if (!result)
1713 die("failed to apply delta");
1714 free(delta_data);
1715 add_delta_base_cache(p, base_offset, base, base_size, *type);
1716 return result;
1719 int do_check_packed_object_crc;
1721 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1722 enum object_type *type, unsigned long *sizep)
1724 struct pack_window *w_curs = NULL;
1725 off_t curpos = obj_offset;
1726 void *data;
1728 if (do_check_packed_object_crc && p->index_version > 1) {
1729 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1730 unsigned long len = revidx[1].offset - obj_offset;
1731 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1732 const unsigned char *sha1 =
1733 nth_packed_object_sha1(p, revidx->nr);
1734 error("bad packed object CRC for %s",
1735 sha1_to_hex(sha1));
1736 mark_bad_packed_object(p, sha1);
1737 unuse_pack(&w_curs);
1738 return NULL;
1742 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1743 switch (*type) {
1744 case OBJ_OFS_DELTA:
1745 case OBJ_REF_DELTA:
1746 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1747 obj_offset, type, sizep);
1748 break;
1749 case OBJ_COMMIT:
1750 case OBJ_TREE:
1751 case OBJ_BLOB:
1752 case OBJ_TAG:
1753 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1754 break;
1755 default:
1756 data = NULL;
1757 error("unknown object type %i at offset %"PRIuMAX" in %s",
1758 *type, (uintmax_t)obj_offset, p->pack_name);
1760 unuse_pack(&w_curs);
1761 return data;
1764 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1765 uint32_t n)
1767 const unsigned char *index = p->index_data;
1768 if (!index) {
1769 if (open_pack_index(p))
1770 return NULL;
1771 index = p->index_data;
1773 if (n >= p->num_objects)
1774 return NULL;
1775 index += 4 * 256;
1776 if (p->index_version == 1) {
1777 return index + 24 * n + 4;
1778 } else {
1779 index += 8;
1780 return index + 20 * n;
1784 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1786 const unsigned char *index = p->index_data;
1787 index += 4 * 256;
1788 if (p->index_version == 1) {
1789 return ntohl(*((uint32_t *)(index + 24 * n)));
1790 } else {
1791 uint32_t off;
1792 index += 8 + p->num_objects * (20 + 4);
1793 off = ntohl(*((uint32_t *)(index + 4 * n)));
1794 if (!(off & 0x80000000))
1795 return off;
1796 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1797 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1798 ntohl(*((uint32_t *)(index + 4)));
1802 off_t find_pack_entry_one(const unsigned char *sha1,
1803 struct packed_git *p)
1805 const uint32_t *level1_ofs = p->index_data;
1806 const unsigned char *index = p->index_data;
1807 unsigned hi, lo, stride;
1808 static int use_lookup = -1;
1809 static int debug_lookup = -1;
1811 if (debug_lookup < 0)
1812 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1814 if (!index) {
1815 if (open_pack_index(p))
1816 return 0;
1817 level1_ofs = p->index_data;
1818 index = p->index_data;
1820 if (p->index_version > 1) {
1821 level1_ofs += 2;
1822 index += 8;
1824 index += 4 * 256;
1825 hi = ntohl(level1_ofs[*sha1]);
1826 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1827 if (p->index_version > 1) {
1828 stride = 20;
1829 } else {
1830 stride = 24;
1831 index += 4;
1834 if (debug_lookup)
1835 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1836 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1838 if (use_lookup < 0)
1839 use_lookup = !!getenv("GIT_USE_LOOKUP");
1840 if (use_lookup) {
1841 int pos = sha1_entry_pos(index, stride, 0,
1842 lo, hi, p->num_objects, sha1);
1843 if (pos < 0)
1844 return 0;
1845 return nth_packed_object_offset(p, pos);
1848 do {
1849 unsigned mi = (lo + hi) / 2;
1850 int cmp = hashcmp(index + mi * stride, sha1);
1852 if (debug_lookup)
1853 printf("lo %u hi %u rg %u mi %u\n",
1854 lo, hi, hi - lo, mi);
1855 if (!cmp)
1856 return nth_packed_object_offset(p, mi);
1857 if (cmp > 0)
1858 hi = mi;
1859 else
1860 lo = mi+1;
1861 } while (lo < hi);
1862 return 0;
1865 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1867 static struct packed_git *last_found = (void *)1;
1868 struct packed_git *p;
1869 off_t offset;
1871 prepare_packed_git();
1872 if (!packed_git)
1873 return 0;
1874 p = (last_found == (void *)1) ? packed_git : last_found;
1876 do {
1877 if (p->num_bad_objects) {
1878 unsigned i;
1879 for (i = 0; i < p->num_bad_objects; i++)
1880 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1881 goto next;
1884 offset = find_pack_entry_one(sha1, p);
1885 if (offset) {
1887 * We are about to tell the caller where they can
1888 * locate the requested object. We better make
1889 * sure the packfile is still here and can be
1890 * accessed before supplying that answer, as
1891 * it may have been deleted since the index
1892 * was loaded!
1894 if (p->pack_fd == -1 && open_packed_git(p)) {
1895 error("packfile %s cannot be accessed", p->pack_name);
1896 goto next;
1898 e->offset = offset;
1899 e->p = p;
1900 hashcpy(e->sha1, sha1);
1901 last_found = p;
1902 return 1;
1905 next:
1906 if (p == last_found)
1907 p = packed_git;
1908 else
1909 p = p->next;
1910 if (p == last_found)
1911 p = p->next;
1912 } while (p);
1913 return 0;
1916 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1917 struct packed_git *packs)
1919 struct packed_git *p;
1921 for (p = packs; p; p = p->next) {
1922 if (find_pack_entry_one(sha1, p))
1923 return p;
1925 return NULL;
1929 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1931 int status;
1932 unsigned long mapsize, size;
1933 void *map;
1934 z_stream stream;
1935 char hdr[32];
1937 map = map_sha1_file(sha1, &mapsize);
1938 if (!map)
1939 return error("unable to find %s", sha1_to_hex(sha1));
1940 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1941 status = error("unable to unpack %s header",
1942 sha1_to_hex(sha1));
1943 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1944 status = error("unable to parse %s header", sha1_to_hex(sha1));
1945 else if (sizep)
1946 *sizep = size;
1947 git_inflate_end(&stream);
1948 munmap(map, mapsize);
1949 return status;
1952 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1954 struct pack_entry e;
1955 int status;
1957 if (!find_pack_entry(sha1, &e)) {
1958 /* Most likely it's a loose object. */
1959 status = sha1_loose_object_info(sha1, sizep);
1960 if (status >= 0)
1961 return status;
1963 /* Not a loose object; someone else may have just packed it. */
1964 reprepare_packed_git();
1965 if (!find_pack_entry(sha1, &e))
1966 return status;
1969 status = packed_object_info(e.p, e.offset, sizep);
1970 if (status < 0) {
1971 mark_bad_packed_object(e.p, sha1);
1972 status = sha1_object_info(sha1, sizep);
1975 return status;
1978 static void *read_packed_sha1(const unsigned char *sha1,
1979 enum object_type *type, unsigned long *size)
1981 struct pack_entry e;
1982 void *data;
1984 if (!find_pack_entry(sha1, &e))
1985 return NULL;
1986 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1987 if (!data) {
1989 * We're probably in deep shit, but let's try to fetch
1990 * the required object anyway from another pack or loose.
1991 * This should happen only in the presence of a corrupted
1992 * pack, and is better than failing outright.
1994 error("failed to read object %s at offset %"PRIuMAX" from %s",
1995 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1996 mark_bad_packed_object(e.p, sha1);
1997 data = read_object(sha1, type, size);
1999 return data;
2003 * This is meant to hold a *small* number of objects that you would
2004 * want read_sha1_file() to be able to return, but yet you do not want
2005 * to write them into the object store (e.g. a browse-only
2006 * application).
2008 static struct cached_object {
2009 unsigned char sha1[20];
2010 enum object_type type;
2011 void *buf;
2012 unsigned long size;
2013 } *cached_objects;
2014 static int cached_object_nr, cached_object_alloc;
2016 static struct cached_object empty_tree = {
2017 EMPTY_TREE_SHA1_BIN,
2018 OBJ_TREE,
2023 static struct cached_object *find_cached_object(const unsigned char *sha1)
2025 int i;
2026 struct cached_object *co = cached_objects;
2028 for (i = 0; i < cached_object_nr; i++, co++) {
2029 if (!hashcmp(co->sha1, sha1))
2030 return co;
2032 if (!hashcmp(sha1, empty_tree.sha1))
2033 return &empty_tree;
2034 return NULL;
2037 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2038 unsigned char *sha1)
2040 struct cached_object *co;
2042 hash_sha1_file(buf, len, typename(type), sha1);
2043 if (has_sha1_file(sha1) || find_cached_object(sha1))
2044 return 0;
2045 if (cached_object_alloc <= cached_object_nr) {
2046 cached_object_alloc = alloc_nr(cached_object_alloc);
2047 cached_objects = xrealloc(cached_objects,
2048 sizeof(*cached_objects) *
2049 cached_object_alloc);
2051 co = &cached_objects[cached_object_nr++];
2052 co->size = len;
2053 co->type = type;
2054 co->buf = xmalloc(len);
2055 memcpy(co->buf, buf, len);
2056 hashcpy(co->sha1, sha1);
2057 return 0;
2060 static void *read_object(const unsigned char *sha1, enum object_type *type,
2061 unsigned long *size)
2063 unsigned long mapsize;
2064 void *map, *buf;
2065 struct cached_object *co;
2067 co = find_cached_object(sha1);
2068 if (co) {
2069 *type = co->type;
2070 *size = co->size;
2071 return xmemdupz(co->buf, co->size);
2074 buf = read_packed_sha1(sha1, type, size);
2075 if (buf)
2076 return buf;
2077 map = map_sha1_file(sha1, &mapsize);
2078 if (map) {
2079 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2080 munmap(map, mapsize);
2081 return buf;
2083 reprepare_packed_git();
2084 return read_packed_sha1(sha1, type, size);
2087 void *read_sha1_file_repl(const unsigned char *sha1,
2088 enum object_type *type,
2089 unsigned long *size,
2090 const unsigned char **replacement)
2092 const unsigned char *repl = lookup_replace_object(sha1);
2093 void *data = read_object(repl, type, size);
2095 /* die if we replaced an object with one that does not exist */
2096 if (!data && repl != sha1)
2097 die("replacement %s not found for %s",
2098 sha1_to_hex(repl), sha1_to_hex(sha1));
2100 /* legacy behavior is to die on corrupted objects */
2101 if (!data && (has_loose_object(repl) || has_packed_and_bad(repl)))
2102 die("object %s is corrupted", sha1_to_hex(repl));
2104 if (replacement)
2105 *replacement = repl;
2107 return data;
2110 void *read_object_with_reference(const unsigned char *sha1,
2111 const char *required_type_name,
2112 unsigned long *size,
2113 unsigned char *actual_sha1_return)
2115 enum object_type type, required_type;
2116 void *buffer;
2117 unsigned long isize;
2118 unsigned char actual_sha1[20];
2120 required_type = type_from_string(required_type_name);
2121 hashcpy(actual_sha1, sha1);
2122 while (1) {
2123 int ref_length = -1;
2124 const char *ref_type = NULL;
2126 buffer = read_sha1_file(actual_sha1, &type, &isize);
2127 if (!buffer)
2128 return NULL;
2129 if (type == required_type) {
2130 *size = isize;
2131 if (actual_sha1_return)
2132 hashcpy(actual_sha1_return, actual_sha1);
2133 return buffer;
2135 /* Handle references */
2136 else if (type == OBJ_COMMIT)
2137 ref_type = "tree ";
2138 else if (type == OBJ_TAG)
2139 ref_type = "object ";
2140 else {
2141 free(buffer);
2142 return NULL;
2144 ref_length = strlen(ref_type);
2146 if (ref_length + 40 > isize ||
2147 memcmp(buffer, ref_type, ref_length) ||
2148 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2149 free(buffer);
2150 return NULL;
2152 free(buffer);
2153 /* Now we have the ID of the referred-to object in
2154 * actual_sha1. Check again. */
2158 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2159 const char *type, unsigned char *sha1,
2160 char *hdr, int *hdrlen)
2162 git_SHA_CTX c;
2164 /* Generate the header */
2165 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2167 /* Sha1.. */
2168 git_SHA1_Init(&c);
2169 git_SHA1_Update(&c, hdr, *hdrlen);
2170 git_SHA1_Update(&c, buf, len);
2171 git_SHA1_Final(sha1, &c);
2175 * Move the just written object into its final resting place.
2176 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2177 * "moving" is only a part of what it does, when no patch between
2178 * master to pu changes the call sites of this function.
2180 int move_temp_to_file(const char *tmpfile, const char *filename)
2182 int ret = 0;
2184 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2185 goto try_rename;
2186 else if (link(tmpfile, filename))
2187 ret = errno;
2190 * Coda hack - coda doesn't like cross-directory links,
2191 * so we fall back to a rename, which will mean that it
2192 * won't be able to check collisions, but that's not a
2193 * big deal.
2195 * The same holds for FAT formatted media.
2197 * When this succeeds, we just return. We have nothing
2198 * left to unlink.
2200 if (ret && ret != EEXIST) {
2201 try_rename:
2202 if (!rename(tmpfile, filename))
2203 goto out;
2204 ret = errno;
2206 unlink_or_warn(tmpfile);
2207 if (ret) {
2208 if (ret != EEXIST) {
2209 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2211 /* FIXME!!! Collision check here ? */
2214 out:
2215 if (adjust_shared_perm(filename))
2216 return error("unable to set permission to '%s'", filename);
2217 return 0;
2220 static int write_buffer(int fd, const void *buf, size_t len)
2222 if (write_in_full(fd, buf, len) < 0)
2223 return error("file write error (%s)", strerror(errno));
2224 return 0;
2227 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2228 unsigned char *sha1)
2230 char hdr[32];
2231 int hdrlen;
2232 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2233 return 0;
2236 /* Finalize a file on disk, and close it. */
2237 static void close_sha1_file(int fd)
2239 if (fsync_object_files)
2240 fsync_or_die(fd, "sha1 file");
2241 if (close(fd) != 0)
2242 die_errno("error when closing sha1 file");
2245 /* Size of directory component, including the ending '/' */
2246 static inline int directory_size(const char *filename)
2248 const char *s = strrchr(filename, '/');
2249 if (!s)
2250 return 0;
2251 return s - filename + 1;
2255 * This creates a temporary file in the same directory as the final
2256 * 'filename'
2258 * We want to avoid cross-directory filename renames, because those
2259 * can have problems on various filesystems (FAT, NFS, Coda).
2261 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2263 int fd, dirlen = directory_size(filename);
2265 if (dirlen + 20 > bufsiz) {
2266 errno = ENAMETOOLONG;
2267 return -1;
2269 memcpy(buffer, filename, dirlen);
2270 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2271 fd = git_mkstemp_mode(buffer, 0444);
2272 if (fd < 0 && dirlen && errno == ENOENT) {
2273 /* Make sure the directory exists */
2274 memcpy(buffer, filename, dirlen);
2275 buffer[dirlen-1] = 0;
2276 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2277 return -1;
2279 /* Try again */
2280 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2281 fd = git_mkstemp_mode(buffer, 0444);
2283 return fd;
2286 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2287 void *buf, unsigned long len, time_t mtime)
2289 int fd, ret;
2290 unsigned char compressed[4096];
2291 z_stream stream;
2292 git_SHA_CTX c;
2293 unsigned char parano_sha1[20];
2294 char *filename;
2295 static char tmpfile[PATH_MAX];
2297 filename = sha1_file_name(sha1);
2298 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2299 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2300 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2301 if (fd < 0) {
2302 if (errno == EACCES)
2303 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2304 else
2305 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2308 /* Set it up */
2309 memset(&stream, 0, sizeof(stream));
2310 deflateInit(&stream, zlib_compression_level);
2311 stream.next_out = compressed;
2312 stream.avail_out = sizeof(compressed);
2313 git_SHA1_Init(&c);
2315 /* First header.. */
2316 stream.next_in = (unsigned char *)hdr;
2317 stream.avail_in = hdrlen;
2318 while (deflate(&stream, 0) == Z_OK)
2319 /* nothing */;
2320 git_SHA1_Update(&c, hdr, hdrlen);
2322 /* Then the data itself.. */
2323 stream.next_in = buf;
2324 stream.avail_in = len;
2325 do {
2326 unsigned char *in0 = stream.next_in;
2327 ret = deflate(&stream, Z_FINISH);
2328 git_SHA1_Update(&c, in0, stream.next_in - in0);
2329 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2330 die("unable to write sha1 file");
2331 stream.next_out = compressed;
2332 stream.avail_out = sizeof(compressed);
2333 } while (ret == Z_OK);
2335 if (ret != Z_STREAM_END)
2336 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2337 ret = deflateEnd(&stream);
2338 if (ret != Z_OK)
2339 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2340 git_SHA1_Final(parano_sha1, &c);
2341 if (hashcmp(sha1, parano_sha1) != 0)
2342 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2344 close_sha1_file(fd);
2346 if (mtime) {
2347 struct utimbuf utb;
2348 utb.actime = mtime;
2349 utb.modtime = mtime;
2350 if (utime(tmpfile, &utb) < 0)
2351 warning("failed utime() on %s: %s",
2352 tmpfile, strerror(errno));
2355 return move_temp_to_file(tmpfile, filename);
2358 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2360 unsigned char sha1[20];
2361 char hdr[32];
2362 int hdrlen;
2364 /* Normally if we have it in the pack then we do not bother writing
2365 * it out into .git/objects/??/?{38} file.
2367 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2368 if (returnsha1)
2369 hashcpy(returnsha1, sha1);
2370 if (has_sha1_file(sha1))
2371 return 0;
2372 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2375 int force_object_loose(const unsigned char *sha1, time_t mtime)
2377 void *buf;
2378 unsigned long len;
2379 enum object_type type;
2380 char hdr[32];
2381 int hdrlen;
2382 int ret;
2384 if (has_loose_object(sha1))
2385 return 0;
2386 buf = read_packed_sha1(sha1, &type, &len);
2387 if (!buf)
2388 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2389 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2390 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2391 free(buf);
2393 return ret;
2396 int has_pack_index(const unsigned char *sha1)
2398 struct stat st;
2399 if (stat(sha1_pack_index_name(sha1), &st))
2400 return 0;
2401 return 1;
2404 int has_sha1_pack(const unsigned char *sha1)
2406 struct pack_entry e;
2407 return find_pack_entry(sha1, &e);
2410 int has_sha1_file(const unsigned char *sha1)
2412 struct pack_entry e;
2414 if (find_pack_entry(sha1, &e))
2415 return 1;
2416 return has_loose_object(sha1);
2419 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2420 int write_object, enum object_type type, const char *path)
2422 int ret, re_allocated = 0;
2424 if (!type)
2425 type = OBJ_BLOB;
2428 * Convert blobs to git internal format
2430 if ((type == OBJ_BLOB) && path) {
2431 struct strbuf nbuf = STRBUF_INIT;
2432 if (convert_to_git(path, buf, size, &nbuf,
2433 write_object ? safe_crlf : 0)) {
2434 buf = strbuf_detach(&nbuf, &size);
2435 re_allocated = 1;
2439 if (write_object)
2440 ret = write_sha1_file(buf, size, typename(type), sha1);
2441 else
2442 ret = hash_sha1_file(buf, size, typename(type), sha1);
2443 if (re_allocated)
2444 free(buf);
2445 return ret;
2448 #define SMALL_FILE_SIZE (32*1024)
2450 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2451 enum object_type type, const char *path)
2453 int ret;
2454 size_t size = xsize_t(st->st_size);
2456 if (!S_ISREG(st->st_mode)) {
2457 struct strbuf sbuf = STRBUF_INIT;
2458 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2459 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2460 type, path);
2461 else
2462 ret = -1;
2463 strbuf_release(&sbuf);
2464 } else if (size <= SMALL_FILE_SIZE) {
2465 char *buf = xmalloc(size);
2466 if (size == read_in_full(fd, buf, size))
2467 ret = index_mem(sha1, buf, size, write_object, type,
2468 path);
2469 else
2470 ret = error("short read %s", strerror(errno));
2471 free(buf);
2472 } else if (size) {
2473 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2474 ret = index_mem(sha1, buf, size, write_object, type, path);
2475 munmap(buf, size);
2476 } else
2477 ret = index_mem(sha1, NULL, size, write_object, type, path);
2478 close(fd);
2479 return ret;
2482 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2484 int fd;
2485 struct strbuf sb = STRBUF_INIT;
2487 switch (st->st_mode & S_IFMT) {
2488 case S_IFREG:
2489 fd = open(path, O_RDONLY);
2490 if (fd < 0)
2491 return error("open(\"%s\"): %s", path,
2492 strerror(errno));
2493 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2494 return error("%s: failed to insert into database",
2495 path);
2496 break;
2497 case S_IFLNK:
2498 if (strbuf_readlink(&sb, path, st->st_size)) {
2499 char *errstr = strerror(errno);
2500 return error("readlink(\"%s\"): %s", path,
2501 errstr);
2503 if (!write_object)
2504 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2505 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2506 return error("%s: failed to insert into database",
2507 path);
2508 strbuf_release(&sb);
2509 break;
2510 case S_IFDIR:
2511 return resolve_gitlink_ref(path, "HEAD", sha1);
2512 default:
2513 return error("%s: unsupported file type", path);
2515 return 0;
2518 int read_pack_header(int fd, struct pack_header *header)
2520 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2521 /* "eof before pack header was fully read" */
2522 return PH_ERROR_EOF;
2524 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2525 /* "protocol error (pack signature mismatch detected)" */
2526 return PH_ERROR_PACK_SIGNATURE;
2527 if (!pack_version_ok(header->hdr_version))
2528 /* "protocol error (pack version unsupported)" */
2529 return PH_ERROR_PROTOCOL;
2530 return 0;