sha1_file.c: add a function to release all packs
[git/git-svn.git] / sha1_file.c
blobd215e0c9bf0fdf0a7a28233ddccd87ba0c51fd68
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 "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
38 const unsigned char null_sha1[20];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git *last_found_pack;
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
71 int i;
72 struct cached_object *co = cached_objects;
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
83 int mkdir_in_gitdir(const char *path)
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
90 if (errno != EEXIST)
91 return -1;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
107 strbuf_release(&sb);
109 return adjust_shared_perm(path);
112 enum scld_error safe_create_leading_directories(char *path)
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
124 if (!*slash)
125 break;
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
159 *slash = slash_character;
161 return ret;
164 enum scld_error safe_create_leading_directories_const(const char *path)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
185 const char *sha1_file_name(const unsigned char *sha1)
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
191 objdir = get_object_directory();
192 len = strlen(objdir);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
226 buf = *name;
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
234 return *base;
237 char *sha1_pack_name(const unsigned char *sha1)
239 static char *name, *base;
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1)
246 static char *name, *base;
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
281 strbuf_addstr(&pathbuf, entry);
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
285 pfxlen = strlen(pathbuf.buf);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
335 ent->base[pfxlen] = '/';
337 return 0;
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
375 void read_info_alternates(const char * relative_base, int depth)
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 char *path;
381 int fd;
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
398 munmap(map, mapsz);
401 void add_to_alternates_file(const char *reference)
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
405 char *alt = mkpath("%s\n", reference);
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
413 int foreach_alt_odb(alt_odb_fn fn, void *cb)
415 struct alternate_object_database *ent;
416 int r = 0;
418 prepare_alt_odb();
419 for (ent = alt_odb_list; ent; ent = ent->next) {
420 r = fn(ent, cb);
421 if (r)
422 break;
424 return r;
427 void prepare_alt_odb(void)
429 const char *alt;
431 if (alt_odb_tail)
432 return;
434 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
435 if (!alt) alt = "";
437 alt_odb_tail = &alt_odb_list;
438 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
440 read_info_alternates(get_object_directory(), 0);
443 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
444 static int freshen_file(const char *fn)
446 struct utimbuf t;
447 t.actime = t.modtime = time(NULL);
448 return !utime(fn, &t);
452 * All of the check_and_freshen functions return 1 if the file exists and was
453 * freshened (if freshening was requested), 0 otherwise. If they return
454 * 0, you should not assume that it is safe to skip a write of the object (it
455 * either does not exist on disk, or has a stale mtime and may be subject to
456 * pruning).
458 static int check_and_freshen_file(const char *fn, int freshen)
460 if (access(fn, F_OK))
461 return 0;
462 if (freshen && !freshen_file(fn))
463 return 0;
464 return 1;
467 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
469 return check_and_freshen_file(sha1_file_name(sha1), freshen);
472 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
474 struct alternate_object_database *alt;
475 prepare_alt_odb();
476 for (alt = alt_odb_list; alt; alt = alt->next) {
477 fill_sha1_path(alt->name, sha1);
478 if (check_and_freshen_file(alt->base, freshen))
479 return 1;
481 return 0;
484 static int check_and_freshen(const unsigned char *sha1, int freshen)
486 return check_and_freshen_local(sha1, freshen) ||
487 check_and_freshen_nonlocal(sha1, freshen);
490 int has_loose_object_nonlocal(const unsigned char *sha1)
492 return check_and_freshen_nonlocal(sha1, 0);
495 static int has_loose_object(const unsigned char *sha1)
497 return check_and_freshen(sha1, 0);
500 static unsigned int pack_used_ctr;
501 static unsigned int pack_mmap_calls;
502 static unsigned int peak_pack_open_windows;
503 static unsigned int pack_open_windows;
504 static unsigned int pack_open_fds;
505 static unsigned int pack_max_fds;
506 static size_t peak_pack_mapped;
507 static size_t pack_mapped;
508 struct packed_git *packed_git;
510 void pack_report(void)
512 fprintf(stderr,
513 "pack_report: getpagesize() = %10" SZ_FMT "\n"
514 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
515 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
516 sz_fmt(getpagesize()),
517 sz_fmt(packed_git_window_size),
518 sz_fmt(packed_git_limit));
519 fprintf(stderr,
520 "pack_report: pack_used_ctr = %10u\n"
521 "pack_report: pack_mmap_calls = %10u\n"
522 "pack_report: pack_open_windows = %10u / %10u\n"
523 "pack_report: pack_mapped = "
524 "%10" SZ_FMT " / %10" SZ_FMT "\n",
525 pack_used_ctr,
526 pack_mmap_calls,
527 pack_open_windows, peak_pack_open_windows,
528 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
532 * Open and mmap the index file at path, perform a couple of
533 * consistency checks, then record its information to p. Return 0 on
534 * success.
536 static int check_packed_git_idx(const char *path, struct packed_git *p)
538 void *idx_map;
539 struct pack_idx_header *hdr;
540 size_t idx_size;
541 uint32_t version, nr, i, *index;
542 int fd = git_open_noatime(path);
543 struct stat st;
545 if (fd < 0)
546 return -1;
547 if (fstat(fd, &st)) {
548 close(fd);
549 return -1;
551 idx_size = xsize_t(st.st_size);
552 if (idx_size < 4 * 256 + 20 + 20) {
553 close(fd);
554 return error("index file %s is too small", path);
556 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
557 close(fd);
559 hdr = idx_map;
560 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
561 version = ntohl(hdr->idx_version);
562 if (version < 2 || version > 2) {
563 munmap(idx_map, idx_size);
564 return error("index file %s is version %"PRIu32
565 " and is not supported by this binary"
566 " (try upgrading GIT to a newer version)",
567 path, version);
569 } else
570 version = 1;
572 nr = 0;
573 index = idx_map;
574 if (version > 1)
575 index += 2; /* skip index header */
576 for (i = 0; i < 256; i++) {
577 uint32_t n = ntohl(index[i]);
578 if (n < nr) {
579 munmap(idx_map, idx_size);
580 return error("non-monotonic index %s", path);
582 nr = n;
585 if (version == 1) {
587 * Total size:
588 * - 256 index entries 4 bytes each
589 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
590 * - 20-byte SHA1 of the packfile
591 * - 20-byte SHA1 file checksum
593 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
594 munmap(idx_map, idx_size);
595 return error("wrong index v1 file size in %s", path);
597 } else if (version == 2) {
599 * Minimum size:
600 * - 8 bytes of header
601 * - 256 index entries 4 bytes each
602 * - 20-byte sha1 entry * nr
603 * - 4-byte crc entry * nr
604 * - 4-byte offset entry * nr
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 * And after the 4-byte offset table might be a
608 * variable sized table containing 8-byte entries
609 * for offsets larger than 2^31.
611 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
612 unsigned long max_size = min_size;
613 if (nr)
614 max_size += (nr - 1)*8;
615 if (idx_size < min_size || idx_size > max_size) {
616 munmap(idx_map, idx_size);
617 return error("wrong index v2 file size in %s", path);
619 if (idx_size != min_size &&
621 * make sure we can deal with large pack offsets.
622 * 31-bit signed offset won't be enough, neither
623 * 32-bit unsigned one will be.
625 (sizeof(off_t) <= 4)) {
626 munmap(idx_map, idx_size);
627 return error("pack too large for current definition of off_t in %s", path);
631 p->index_version = version;
632 p->index_data = idx_map;
633 p->index_size = idx_size;
634 p->num_objects = nr;
635 return 0;
638 int open_pack_index(struct packed_git *p)
640 char *idx_name;
641 int ret;
643 if (p->index_data)
644 return 0;
646 idx_name = xstrdup(p->pack_name);
647 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
648 ret = check_packed_git_idx(idx_name, p);
649 free(idx_name);
650 return ret;
653 static void scan_windows(struct packed_git *p,
654 struct packed_git **lru_p,
655 struct pack_window **lru_w,
656 struct pack_window **lru_l)
658 struct pack_window *w, *w_l;
660 for (w_l = NULL, w = p->windows; w; w = w->next) {
661 if (!w->inuse_cnt) {
662 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
663 *lru_p = p;
664 *lru_w = w;
665 *lru_l = w_l;
668 w_l = w;
672 static int unuse_one_window(struct packed_git *current)
674 struct packed_git *p, *lru_p = NULL;
675 struct pack_window *lru_w = NULL, *lru_l = NULL;
677 if (current)
678 scan_windows(current, &lru_p, &lru_w, &lru_l);
679 for (p = packed_git; p; p = p->next)
680 scan_windows(p, &lru_p, &lru_w, &lru_l);
681 if (lru_p) {
682 munmap(lru_w->base, lru_w->len);
683 pack_mapped -= lru_w->len;
684 if (lru_l)
685 lru_l->next = lru_w->next;
686 else
687 lru_p->windows = lru_w->next;
688 free(lru_w);
689 pack_open_windows--;
690 return 1;
692 return 0;
695 void release_pack_memory(size_t need)
697 size_t cur = pack_mapped;
698 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
699 ; /* nothing */
702 static void mmap_limit_check(size_t length)
704 static size_t limit = 0;
705 if (!limit) {
706 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
707 if (!limit)
708 limit = SIZE_MAX;
710 if (length > limit)
711 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
712 (uintmax_t)length, (uintmax_t)limit);
715 void *xmmap_gently(void *start, size_t length,
716 int prot, int flags, int fd, off_t offset)
718 void *ret;
720 mmap_limit_check(length);
721 ret = mmap(start, length, prot, flags, fd, offset);
722 if (ret == MAP_FAILED) {
723 if (!length)
724 return NULL;
725 release_pack_memory(length);
726 ret = mmap(start, length, prot, flags, fd, offset);
728 return ret;
731 void *xmmap(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
734 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
735 if (ret == MAP_FAILED)
736 die_errno("mmap failed");
737 return ret;
740 void close_pack_windows(struct packed_git *p)
742 while (p->windows) {
743 struct pack_window *w = p->windows;
745 if (w->inuse_cnt)
746 die("pack '%s' still has open windows to it",
747 p->pack_name);
748 munmap(w->base, w->len);
749 pack_mapped -= w->len;
750 pack_open_windows--;
751 p->windows = w->next;
752 free(w);
756 static int close_pack_fd(struct packed_git *p)
758 if (p->pack_fd < 0)
759 return 0;
761 close(p->pack_fd);
762 pack_open_fds--;
763 p->pack_fd = -1;
765 return 1;
768 static void close_pack(struct packed_git *p)
770 close_pack_windows(p);
771 close_pack_fd(p);
772 close_pack_index(p);
775 void close_all_packs(void)
777 struct packed_git *p;
779 for (p = packed_git; p; p = p->next)
780 if (p->do_not_close)
781 die("BUG! Want to close pack marked 'do-not-close'");
782 else
783 close_pack(p);
788 * The LRU pack is the one with the oldest MRU window, preferring packs
789 * with no used windows, or the oldest mtime if it has no windows allocated.
791 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
793 struct pack_window *w, *this_mru_w;
794 int has_windows_inuse = 0;
797 * Reject this pack if it has windows and the previously selected
798 * one does not. If this pack does not have windows, reject
799 * it if the pack file is newer than the previously selected one.
801 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
802 return;
804 for (w = this_mru_w = p->windows; w; w = w->next) {
806 * Reject this pack if any of its windows are in use,
807 * but the previously selected pack did not have any
808 * inuse windows. Otherwise, record that this pack
809 * has windows in use.
811 if (w->inuse_cnt) {
812 if (*accept_windows_inuse)
813 has_windows_inuse = 1;
814 else
815 return;
818 if (w->last_used > this_mru_w->last_used)
819 this_mru_w = w;
822 * Reject this pack if it has windows that have been
823 * used more recently than the previously selected pack.
824 * If the previously selected pack had windows inuse and
825 * we have not encountered a window in this pack that is
826 * inuse, skip this check since we prefer a pack with no
827 * inuse windows to one that has inuse windows.
829 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
830 this_mru_w->last_used > (*mru_w)->last_used)
831 return;
835 * Select this pack.
837 *mru_w = this_mru_w;
838 *lru_p = p;
839 *accept_windows_inuse = has_windows_inuse;
842 static int close_one_pack(void)
844 struct packed_git *p, *lru_p = NULL;
845 struct pack_window *mru_w = NULL;
846 int accept_windows_inuse = 1;
848 for (p = packed_git; p; p = p->next) {
849 if (p->pack_fd == -1)
850 continue;
851 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
854 if (lru_p)
855 return close_pack_fd(lru_p);
857 return 0;
860 void unuse_pack(struct pack_window **w_cursor)
862 struct pack_window *w = *w_cursor;
863 if (w) {
864 w->inuse_cnt--;
865 *w_cursor = NULL;
869 void close_pack_index(struct packed_git *p)
871 if (p->index_data) {
872 munmap((void *)p->index_data, p->index_size);
873 p->index_data = NULL;
878 * This is used by git-repack in case a newly created pack happens to
879 * contain the same set of objects as an existing one. In that case
880 * the resulting file might be different even if its name would be the
881 * same. It is best to close any reference to the old pack before it is
882 * replaced on disk. Of course no index pointers or windows for given pack
883 * must subsist at this point. If ever objects from this pack are requested
884 * again, the new version of the pack will be reinitialized through
885 * reprepare_packed_git().
887 void free_pack_by_name(const char *pack_name)
889 struct packed_git *p, **pp = &packed_git;
891 while (*pp) {
892 p = *pp;
893 if (strcmp(pack_name, p->pack_name) == 0) {
894 clear_delta_base_cache();
895 close_pack(p);
896 free(p->bad_object_sha1);
897 *pp = p->next;
898 if (last_found_pack == p)
899 last_found_pack = NULL;
900 free(p);
901 return;
903 pp = &p->next;
907 static unsigned int get_max_fd_limit(void)
909 #ifdef RLIMIT_NOFILE
911 struct rlimit lim;
913 if (!getrlimit(RLIMIT_NOFILE, &lim))
914 return lim.rlim_cur;
916 #endif
918 #ifdef _SC_OPEN_MAX
920 long open_max = sysconf(_SC_OPEN_MAX);
921 if (0 < open_max)
922 return open_max;
924 * Otherwise, we got -1 for one of the two
925 * reasons:
927 * (1) sysconf() did not understand _SC_OPEN_MAX
928 * and signaled an error with -1; or
929 * (2) sysconf() said there is no limit.
931 * We _could_ clear errno before calling sysconf() to
932 * tell these two cases apart and return a huge number
933 * in the latter case to let the caller cap it to a
934 * value that is not so selfish, but letting the
935 * fallback OPEN_MAX codepath take care of these cases
936 * is a lot simpler.
939 #endif
941 #ifdef OPEN_MAX
942 return OPEN_MAX;
943 #else
944 return 1; /* see the caller ;-) */
945 #endif
949 * Do not call this directly as this leaks p->pack_fd on error return;
950 * call open_packed_git() instead.
952 static int open_packed_git_1(struct packed_git *p)
954 struct stat st;
955 struct pack_header hdr;
956 unsigned char sha1[20];
957 unsigned char *idx_sha1;
958 long fd_flag;
960 if (!p->index_data && open_pack_index(p))
961 return error("packfile %s index unavailable", p->pack_name);
963 if (!pack_max_fds) {
964 unsigned int max_fds = get_max_fd_limit();
966 /* Save 3 for stdin/stdout/stderr, 22 for work */
967 if (25 < max_fds)
968 pack_max_fds = max_fds - 25;
969 else
970 pack_max_fds = 1;
973 while (pack_max_fds <= pack_open_fds && close_one_pack())
974 ; /* nothing */
976 p->pack_fd = git_open_noatime(p->pack_name);
977 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
978 return -1;
979 pack_open_fds++;
981 /* If we created the struct before we had the pack we lack size. */
982 if (!p->pack_size) {
983 if (!S_ISREG(st.st_mode))
984 return error("packfile %s not a regular file", p->pack_name);
985 p->pack_size = st.st_size;
986 } else if (p->pack_size != st.st_size)
987 return error("packfile %s size changed", p->pack_name);
989 /* We leave these file descriptors open with sliding mmap;
990 * there is no point keeping them open across exec(), though.
992 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
993 if (fd_flag < 0)
994 return error("cannot determine file descriptor flags");
995 fd_flag |= FD_CLOEXEC;
996 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
997 return error("cannot set FD_CLOEXEC");
999 /* Verify we recognize this pack file format. */
1000 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1001 return error("file %s is far too short to be a packfile", p->pack_name);
1002 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1003 return error("file %s is not a GIT packfile", p->pack_name);
1004 if (!pack_version_ok(hdr.hdr_version))
1005 return error("packfile %s is version %"PRIu32" and not"
1006 " supported (try upgrading GIT to a newer version)",
1007 p->pack_name, ntohl(hdr.hdr_version));
1009 /* Verify the pack matches its index. */
1010 if (p->num_objects != ntohl(hdr.hdr_entries))
1011 return error("packfile %s claims to have %"PRIu32" objects"
1012 " while index indicates %"PRIu32" objects",
1013 p->pack_name, ntohl(hdr.hdr_entries),
1014 p->num_objects);
1015 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1016 return error("end of packfile %s is unavailable", p->pack_name);
1017 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1018 return error("packfile %s signature is unavailable", p->pack_name);
1019 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1020 if (hashcmp(sha1, idx_sha1))
1021 return error("packfile %s does not match index", p->pack_name);
1022 return 0;
1025 static int open_packed_git(struct packed_git *p)
1027 if (!open_packed_git_1(p))
1028 return 0;
1029 close_pack_fd(p);
1030 return -1;
1033 static int in_window(struct pack_window *win, off_t offset)
1035 /* We must promise at least 20 bytes (one hash) after the
1036 * offset is available from this window, otherwise the offset
1037 * is not actually in this window and a different window (which
1038 * has that one hash excess) must be used. This is to support
1039 * the object header and delta base parsing routines below.
1041 off_t win_off = win->offset;
1042 return win_off <= offset
1043 && (offset + 20) <= (win_off + win->len);
1046 unsigned char *use_pack(struct packed_git *p,
1047 struct pack_window **w_cursor,
1048 off_t offset,
1049 unsigned long *left)
1051 struct pack_window *win = *w_cursor;
1053 /* Since packfiles end in a hash of their content and it's
1054 * pointless to ask for an offset into the middle of that
1055 * hash, and the in_window function above wouldn't match
1056 * don't allow an offset too close to the end of the file.
1058 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1059 die("packfile %s cannot be accessed", p->pack_name);
1060 if (offset > (p->pack_size - 20))
1061 die("offset beyond end of packfile (truncated pack?)");
1063 if (!win || !in_window(win, offset)) {
1064 if (win)
1065 win->inuse_cnt--;
1066 for (win = p->windows; win; win = win->next) {
1067 if (in_window(win, offset))
1068 break;
1070 if (!win) {
1071 size_t window_align = packed_git_window_size / 2;
1072 off_t len;
1074 if (p->pack_fd == -1 && open_packed_git(p))
1075 die("packfile %s cannot be accessed", p->pack_name);
1077 win = xcalloc(1, sizeof(*win));
1078 win->offset = (offset / window_align) * window_align;
1079 len = p->pack_size - win->offset;
1080 if (len > packed_git_window_size)
1081 len = packed_git_window_size;
1082 win->len = (size_t)len;
1083 pack_mapped += win->len;
1084 while (packed_git_limit < pack_mapped
1085 && unuse_one_window(p))
1086 ; /* nothing */
1087 win->base = xmmap(NULL, win->len,
1088 PROT_READ, MAP_PRIVATE,
1089 p->pack_fd, win->offset);
1090 if (win->base == MAP_FAILED)
1091 die("packfile %s cannot be mapped: %s",
1092 p->pack_name,
1093 strerror(errno));
1094 if (!win->offset && win->len == p->pack_size
1095 && !p->do_not_close)
1096 close_pack_fd(p);
1097 pack_mmap_calls++;
1098 pack_open_windows++;
1099 if (pack_mapped > peak_pack_mapped)
1100 peak_pack_mapped = pack_mapped;
1101 if (pack_open_windows > peak_pack_open_windows)
1102 peak_pack_open_windows = pack_open_windows;
1103 win->next = p->windows;
1104 p->windows = win;
1107 if (win != *w_cursor) {
1108 win->last_used = pack_used_ctr++;
1109 win->inuse_cnt++;
1110 *w_cursor = win;
1112 offset -= win->offset;
1113 if (left)
1114 *left = win->len - xsize_t(offset);
1115 return win->base + offset;
1118 static struct packed_git *alloc_packed_git(int extra)
1120 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1121 memset(p, 0, sizeof(*p));
1122 p->pack_fd = -1;
1123 return p;
1126 static void try_to_free_pack_memory(size_t size)
1128 release_pack_memory(size);
1131 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1133 static int have_set_try_to_free_routine;
1134 struct stat st;
1135 struct packed_git *p = alloc_packed_git(path_len + 2);
1137 if (!have_set_try_to_free_routine) {
1138 have_set_try_to_free_routine = 1;
1139 set_try_to_free_routine(try_to_free_pack_memory);
1143 * Make sure a corresponding .pack file exists and that
1144 * the index looks sane.
1146 path_len -= strlen(".idx");
1147 if (path_len < 1) {
1148 free(p);
1149 return NULL;
1151 memcpy(p->pack_name, path, path_len);
1153 strcpy(p->pack_name + path_len, ".keep");
1154 if (!access(p->pack_name, F_OK))
1155 p->pack_keep = 1;
1157 strcpy(p->pack_name + path_len, ".pack");
1158 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1159 free(p);
1160 return NULL;
1163 /* ok, it looks sane as far as we can check without
1164 * actually mapping the pack file.
1166 p->pack_size = st.st_size;
1167 p->pack_local = local;
1168 p->mtime = st.st_mtime;
1169 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1170 hashclr(p->sha1);
1171 return p;
1174 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1176 const char *path = sha1_pack_name(sha1);
1177 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1179 strcpy(p->pack_name, path);
1180 hashcpy(p->sha1, sha1);
1181 if (check_packed_git_idx(idx_path, p)) {
1182 free(p);
1183 return NULL;
1186 return p;
1189 void install_packed_git(struct packed_git *pack)
1191 if (pack->pack_fd != -1)
1192 pack_open_fds++;
1194 pack->next = packed_git;
1195 packed_git = pack;
1198 void (*report_garbage)(const char *desc, const char *path);
1200 static void report_helper(const struct string_list *list,
1201 int seen_bits, int first, int last)
1203 const char *msg;
1204 switch (seen_bits) {
1205 case 0:
1206 msg = "no corresponding .idx or .pack";
1207 break;
1208 case 1:
1209 msg = "no corresponding .idx";
1210 break;
1211 case 2:
1212 msg = "no corresponding .pack";
1213 break;
1214 default:
1215 return;
1217 for (; first < last; first++)
1218 report_garbage(msg, list->items[first].string);
1221 static void report_pack_garbage(struct string_list *list)
1223 int i, baselen = -1, first = 0, seen_bits = 0;
1225 if (!report_garbage)
1226 return;
1228 string_list_sort(list);
1230 for (i = 0; i < list->nr; i++) {
1231 const char *path = list->items[i].string;
1232 if (baselen != -1 &&
1233 strncmp(path, list->items[first].string, baselen)) {
1234 report_helper(list, seen_bits, first, i);
1235 baselen = -1;
1236 seen_bits = 0;
1238 if (baselen == -1) {
1239 const char *dot = strrchr(path, '.');
1240 if (!dot) {
1241 report_garbage("garbage found", path);
1242 continue;
1244 baselen = dot - path + 1;
1245 first = i;
1247 if (!strcmp(path + baselen, "pack"))
1248 seen_bits |= 1;
1249 else if (!strcmp(path + baselen, "idx"))
1250 seen_bits |= 2;
1252 report_helper(list, seen_bits, first, list->nr);
1255 static void prepare_packed_git_one(char *objdir, int local)
1257 struct strbuf path = STRBUF_INIT;
1258 size_t dirnamelen;
1259 DIR *dir;
1260 struct dirent *de;
1261 struct string_list garbage = STRING_LIST_INIT_DUP;
1263 strbuf_addstr(&path, objdir);
1264 strbuf_addstr(&path, "/pack");
1265 dir = opendir(path.buf);
1266 if (!dir) {
1267 if (errno != ENOENT)
1268 error("unable to open object pack directory: %s: %s",
1269 path.buf, strerror(errno));
1270 strbuf_release(&path);
1271 return;
1273 strbuf_addch(&path, '/');
1274 dirnamelen = path.len;
1275 while ((de = readdir(dir)) != NULL) {
1276 struct packed_git *p;
1277 size_t base_len;
1279 if (is_dot_or_dotdot(de->d_name))
1280 continue;
1282 strbuf_setlen(&path, dirnamelen);
1283 strbuf_addstr(&path, de->d_name);
1285 base_len = path.len;
1286 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1287 /* Don't reopen a pack we already have. */
1288 for (p = packed_git; p; p = p->next) {
1289 size_t len;
1290 if (strip_suffix(p->pack_name, ".pack", &len) &&
1291 len == base_len &&
1292 !memcmp(p->pack_name, path.buf, len))
1293 break;
1295 if (p == NULL &&
1297 * See if it really is a valid .idx file with
1298 * corresponding .pack file that we can map.
1300 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1301 install_packed_git(p);
1304 if (!report_garbage)
1305 continue;
1307 if (ends_with(de->d_name, ".idx") ||
1308 ends_with(de->d_name, ".pack") ||
1309 ends_with(de->d_name, ".bitmap") ||
1310 ends_with(de->d_name, ".keep"))
1311 string_list_append(&garbage, path.buf);
1312 else
1313 report_garbage("garbage found", path.buf);
1315 closedir(dir);
1316 report_pack_garbage(&garbage);
1317 string_list_clear(&garbage, 0);
1318 strbuf_release(&path);
1321 static int sort_pack(const void *a_, const void *b_)
1323 struct packed_git *a = *((struct packed_git **)a_);
1324 struct packed_git *b = *((struct packed_git **)b_);
1325 int st;
1328 * Local packs tend to contain objects specific to our
1329 * variant of the project than remote ones. In addition,
1330 * remote ones could be on a network mounted filesystem.
1331 * Favor local ones for these reasons.
1333 st = a->pack_local - b->pack_local;
1334 if (st)
1335 return -st;
1338 * Younger packs tend to contain more recent objects,
1339 * and more recent objects tend to get accessed more
1340 * often.
1342 if (a->mtime < b->mtime)
1343 return 1;
1344 else if (a->mtime == b->mtime)
1345 return 0;
1346 return -1;
1349 static void rearrange_packed_git(void)
1351 struct packed_git **ary, *p;
1352 int i, n;
1354 for (n = 0, p = packed_git; p; p = p->next)
1355 n++;
1356 if (n < 2)
1357 return;
1359 /* prepare an array of packed_git for easier sorting */
1360 ary = xcalloc(n, sizeof(struct packed_git *));
1361 for (n = 0, p = packed_git; p; p = p->next)
1362 ary[n++] = p;
1364 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1366 /* link them back again */
1367 for (i = 0; i < n - 1; i++)
1368 ary[i]->next = ary[i + 1];
1369 ary[n - 1]->next = NULL;
1370 packed_git = ary[0];
1372 free(ary);
1375 static int prepare_packed_git_run_once = 0;
1376 void prepare_packed_git(void)
1378 struct alternate_object_database *alt;
1380 if (prepare_packed_git_run_once)
1381 return;
1382 prepare_packed_git_one(get_object_directory(), 1);
1383 prepare_alt_odb();
1384 for (alt = alt_odb_list; alt; alt = alt->next) {
1385 alt->name[-1] = 0;
1386 prepare_packed_git_one(alt->base, 0);
1387 alt->name[-1] = '/';
1389 rearrange_packed_git();
1390 prepare_packed_git_run_once = 1;
1393 void reprepare_packed_git(void)
1395 prepare_packed_git_run_once = 0;
1396 prepare_packed_git();
1399 static void mark_bad_packed_object(struct packed_git *p,
1400 const unsigned char *sha1)
1402 unsigned i;
1403 for (i = 0; i < p->num_bad_objects; i++)
1404 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1405 return;
1406 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1407 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1408 p->num_bad_objects++;
1411 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1413 struct packed_git *p;
1414 unsigned i;
1416 for (p = packed_git; p; p = p->next)
1417 for (i = 0; i < p->num_bad_objects; i++)
1418 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1419 return p;
1420 return NULL;
1424 * With an in-core object data in "map", rehash it to make sure the
1425 * object name actually matches "sha1" to detect object corruption.
1426 * With "map" == NULL, try reading the object named with "sha1" using
1427 * the streaming interface and rehash it to do the same.
1429 int check_sha1_signature(const unsigned char *sha1, void *map,
1430 unsigned long size, const char *type)
1432 unsigned char real_sha1[20];
1433 enum object_type obj_type;
1434 struct git_istream *st;
1435 git_SHA_CTX c;
1436 char hdr[32];
1437 int hdrlen;
1439 if (map) {
1440 hash_sha1_file(map, size, type, real_sha1);
1441 return hashcmp(sha1, real_sha1) ? -1 : 0;
1444 st = open_istream(sha1, &obj_type, &size, NULL);
1445 if (!st)
1446 return -1;
1448 /* Generate the header */
1449 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1451 /* Sha1.. */
1452 git_SHA1_Init(&c);
1453 git_SHA1_Update(&c, hdr, hdrlen);
1454 for (;;) {
1455 char buf[1024 * 16];
1456 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1458 if (readlen < 0) {
1459 close_istream(st);
1460 return -1;
1462 if (!readlen)
1463 break;
1464 git_SHA1_Update(&c, buf, readlen);
1466 git_SHA1_Final(real_sha1, &c);
1467 close_istream(st);
1468 return hashcmp(sha1, real_sha1) ? -1 : 0;
1471 int git_open_noatime(const char *name)
1473 static int sha1_file_open_flag = O_NOATIME;
1475 for (;;) {
1476 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1477 if (fd >= 0)
1478 return fd;
1480 /* Might the failure be due to O_NOATIME? */
1481 if (errno != ENOENT && sha1_file_open_flag) {
1482 sha1_file_open_flag = 0;
1483 continue;
1486 return -1;
1490 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1492 struct alternate_object_database *alt;
1494 if (!lstat(sha1_file_name(sha1), st))
1495 return 0;
1497 prepare_alt_odb();
1498 errno = ENOENT;
1499 for (alt = alt_odb_list; alt; alt = alt->next) {
1500 fill_sha1_path(alt->name, sha1);
1501 if (!lstat(alt->base, st))
1502 return 0;
1505 return -1;
1508 static int open_sha1_file(const unsigned char *sha1)
1510 int fd;
1511 struct alternate_object_database *alt;
1512 int most_interesting_errno;
1514 fd = git_open_noatime(sha1_file_name(sha1));
1515 if (fd >= 0)
1516 return fd;
1517 most_interesting_errno = errno;
1519 prepare_alt_odb();
1520 for (alt = alt_odb_list; alt; alt = alt->next) {
1521 fill_sha1_path(alt->name, sha1);
1522 fd = git_open_noatime(alt->base);
1523 if (fd >= 0)
1524 return fd;
1525 if (most_interesting_errno == ENOENT)
1526 most_interesting_errno = errno;
1528 errno = most_interesting_errno;
1529 return -1;
1532 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1534 void *map;
1535 int fd;
1537 fd = open_sha1_file(sha1);
1538 map = NULL;
1539 if (fd >= 0) {
1540 struct stat st;
1542 if (!fstat(fd, &st)) {
1543 *size = xsize_t(st.st_size);
1544 if (!*size) {
1545 /* mmap() is forbidden on empty files */
1546 error("object file %s is empty", sha1_file_name(sha1));
1547 return NULL;
1549 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1551 close(fd);
1553 return map;
1556 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1557 unsigned long len, enum object_type *type, unsigned long *sizep)
1559 unsigned shift;
1560 unsigned long size, c;
1561 unsigned long used = 0;
1563 c = buf[used++];
1564 *type = (c >> 4) & 7;
1565 size = c & 15;
1566 shift = 4;
1567 while (c & 0x80) {
1568 if (len <= used || bitsizeof(long) <= shift) {
1569 error("bad object header");
1570 size = used = 0;
1571 break;
1573 c = buf[used++];
1574 size += (c & 0x7f) << shift;
1575 shift += 7;
1577 *sizep = size;
1578 return used;
1581 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1583 /* Get the data stream */
1584 memset(stream, 0, sizeof(*stream));
1585 stream->next_in = map;
1586 stream->avail_in = mapsize;
1587 stream->next_out = buffer;
1588 stream->avail_out = bufsiz;
1590 git_inflate_init(stream);
1591 return git_inflate(stream, 0);
1594 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1596 int bytes = strlen(buffer) + 1;
1597 unsigned char *buf = xmallocz(size);
1598 unsigned long n;
1599 int status = Z_OK;
1601 n = stream->total_out - bytes;
1602 if (n > size)
1603 n = size;
1604 memcpy(buf, (char *) buffer + bytes, n);
1605 bytes = n;
1606 if (bytes <= size) {
1608 * The above condition must be (bytes <= size), not
1609 * (bytes < size). In other words, even though we
1610 * expect no more output and set avail_out to zero,
1611 * the input zlib stream may have bytes that express
1612 * "this concludes the stream", and we *do* want to
1613 * eat that input.
1615 * Otherwise we would not be able to test that we
1616 * consumed all the input to reach the expected size;
1617 * we also want to check that zlib tells us that all
1618 * went well with status == Z_STREAM_END at the end.
1620 stream->next_out = buf + bytes;
1621 stream->avail_out = size - bytes;
1622 while (status == Z_OK)
1623 status = git_inflate(stream, Z_FINISH);
1625 if (status == Z_STREAM_END && !stream->avail_in) {
1626 git_inflate_end(stream);
1627 return buf;
1630 if (status < 0)
1631 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1632 else if (stream->avail_in)
1633 error("garbage at end of loose object '%s'",
1634 sha1_to_hex(sha1));
1635 free(buf);
1636 return NULL;
1640 * We used to just use "sscanf()", but that's actually way
1641 * too permissive for what we want to check. So do an anal
1642 * object header parse by hand.
1644 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1646 char type[10];
1647 int i;
1648 unsigned long size;
1651 * The type can be at most ten bytes (including the
1652 * terminating '\0' that we add), and is followed by
1653 * a space.
1655 i = 0;
1656 for (;;) {
1657 char c = *hdr++;
1658 if (c == ' ')
1659 break;
1660 type[i++] = c;
1661 if (i >= sizeof(type))
1662 return -1;
1664 type[i] = 0;
1667 * The length must follow immediately, and be in canonical
1668 * decimal format (ie "010" is not valid).
1670 size = *hdr++ - '0';
1671 if (size > 9)
1672 return -1;
1673 if (size) {
1674 for (;;) {
1675 unsigned long c = *hdr - '0';
1676 if (c > 9)
1677 break;
1678 hdr++;
1679 size = size * 10 + c;
1682 *sizep = size;
1685 * The length must be followed by a zero byte
1687 return *hdr ? -1 : type_from_string(type);
1690 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1692 int ret;
1693 git_zstream stream;
1694 char hdr[8192];
1696 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1697 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1698 return NULL;
1700 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1703 unsigned long get_size_from_delta(struct packed_git *p,
1704 struct pack_window **w_curs,
1705 off_t curpos)
1707 const unsigned char *data;
1708 unsigned char delta_head[20], *in;
1709 git_zstream stream;
1710 int st;
1712 memset(&stream, 0, sizeof(stream));
1713 stream.next_out = delta_head;
1714 stream.avail_out = sizeof(delta_head);
1716 git_inflate_init(&stream);
1717 do {
1718 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1719 stream.next_in = in;
1720 st = git_inflate(&stream, Z_FINISH);
1721 curpos += stream.next_in - in;
1722 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1723 stream.total_out < sizeof(delta_head));
1724 git_inflate_end(&stream);
1725 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1726 error("delta data unpack-initial failed");
1727 return 0;
1730 /* Examine the initial part of the delta to figure out
1731 * the result size.
1733 data = delta_head;
1735 /* ignore base size */
1736 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1738 /* Read the result size */
1739 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1742 static off_t get_delta_base(struct packed_git *p,
1743 struct pack_window **w_curs,
1744 off_t *curpos,
1745 enum object_type type,
1746 off_t delta_obj_offset)
1748 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1749 off_t base_offset;
1751 /* use_pack() assured us we have [base_info, base_info + 20)
1752 * as a range that we can look at without walking off the
1753 * end of the mapped window. Its actually the hash size
1754 * that is assured. An OFS_DELTA longer than the hash size
1755 * is stupid, as then a REF_DELTA would be smaller to store.
1757 if (type == OBJ_OFS_DELTA) {
1758 unsigned used = 0;
1759 unsigned char c = base_info[used++];
1760 base_offset = c & 127;
1761 while (c & 128) {
1762 base_offset += 1;
1763 if (!base_offset || MSB(base_offset, 7))
1764 return 0; /* overflow */
1765 c = base_info[used++];
1766 base_offset = (base_offset << 7) + (c & 127);
1768 base_offset = delta_obj_offset - base_offset;
1769 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1770 return 0; /* out of bound */
1771 *curpos += used;
1772 } else if (type == OBJ_REF_DELTA) {
1773 /* The base entry _must_ be in the same pack */
1774 base_offset = find_pack_entry_one(base_info, p);
1775 *curpos += 20;
1776 } else
1777 die("I am totally screwed");
1778 return base_offset;
1782 * Like get_delta_base above, but we return the sha1 instead of the pack
1783 * offset. This means it is cheaper for REF deltas (we do not have to do
1784 * the final object lookup), but more expensive for OFS deltas (we
1785 * have to load the revidx to convert the offset back into a sha1).
1787 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1788 struct pack_window **w_curs,
1789 off_t curpos,
1790 enum object_type type,
1791 off_t delta_obj_offset)
1793 if (type == OBJ_REF_DELTA) {
1794 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1795 return base;
1796 } else if (type == OBJ_OFS_DELTA) {
1797 struct revindex_entry *revidx;
1798 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1799 type, delta_obj_offset);
1801 if (!base_offset)
1802 return NULL;
1804 revidx = find_pack_revindex(p, base_offset);
1805 if (!revidx)
1806 return NULL;
1808 return nth_packed_object_sha1(p, revidx->nr);
1809 } else
1810 return NULL;
1813 int unpack_object_header(struct packed_git *p,
1814 struct pack_window **w_curs,
1815 off_t *curpos,
1816 unsigned long *sizep)
1818 unsigned char *base;
1819 unsigned long left;
1820 unsigned long used;
1821 enum object_type type;
1823 /* use_pack() assures us we have [base, base + 20) available
1824 * as a range that we can look at. (Its actually the hash
1825 * size that is assured.) With our object header encoding
1826 * the maximum deflated object size is 2^137, which is just
1827 * insane, so we know won't exceed what we have been given.
1829 base = use_pack(p, w_curs, *curpos, &left);
1830 used = unpack_object_header_buffer(base, left, &type, sizep);
1831 if (!used) {
1832 type = OBJ_BAD;
1833 } else
1834 *curpos += used;
1836 return type;
1839 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1841 int type;
1842 struct revindex_entry *revidx;
1843 const unsigned char *sha1;
1844 revidx = find_pack_revindex(p, obj_offset);
1845 if (!revidx)
1846 return OBJ_BAD;
1847 sha1 = nth_packed_object_sha1(p, revidx->nr);
1848 mark_bad_packed_object(p, sha1);
1849 type = sha1_object_info(sha1, NULL);
1850 if (type <= OBJ_NONE)
1851 return OBJ_BAD;
1852 return type;
1855 #define POI_STACK_PREALLOC 64
1857 static enum object_type packed_to_object_type(struct packed_git *p,
1858 off_t obj_offset,
1859 enum object_type type,
1860 struct pack_window **w_curs,
1861 off_t curpos)
1863 off_t small_poi_stack[POI_STACK_PREALLOC];
1864 off_t *poi_stack = small_poi_stack;
1865 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1867 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1868 off_t base_offset;
1869 unsigned long size;
1870 /* Push the object we're going to leave behind */
1871 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1872 poi_stack_alloc = alloc_nr(poi_stack_nr);
1873 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1874 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1875 } else {
1876 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1878 poi_stack[poi_stack_nr++] = obj_offset;
1879 /* If parsing the base offset fails, just unwind */
1880 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1881 if (!base_offset)
1882 goto unwind;
1883 curpos = obj_offset = base_offset;
1884 type = unpack_object_header(p, w_curs, &curpos, &size);
1885 if (type <= OBJ_NONE) {
1886 /* If getting the base itself fails, we first
1887 * retry the base, otherwise unwind */
1888 type = retry_bad_packed_offset(p, base_offset);
1889 if (type > OBJ_NONE)
1890 goto out;
1891 goto unwind;
1895 switch (type) {
1896 case OBJ_BAD:
1897 case OBJ_COMMIT:
1898 case OBJ_TREE:
1899 case OBJ_BLOB:
1900 case OBJ_TAG:
1901 break;
1902 default:
1903 error("unknown object type %i at offset %"PRIuMAX" in %s",
1904 type, (uintmax_t)obj_offset, p->pack_name);
1905 type = OBJ_BAD;
1908 out:
1909 if (poi_stack != small_poi_stack)
1910 free(poi_stack);
1911 return type;
1913 unwind:
1914 while (poi_stack_nr) {
1915 obj_offset = poi_stack[--poi_stack_nr];
1916 type = retry_bad_packed_offset(p, obj_offset);
1917 if (type > OBJ_NONE)
1918 goto out;
1920 type = OBJ_BAD;
1921 goto out;
1924 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1925 struct object_info *oi)
1927 struct pack_window *w_curs = NULL;
1928 unsigned long size;
1929 off_t curpos = obj_offset;
1930 enum object_type type;
1933 * We always get the representation type, but only convert it to
1934 * a "real" type later if the caller is interested.
1936 type = unpack_object_header(p, &w_curs, &curpos, &size);
1938 if (oi->sizep) {
1939 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1940 off_t tmp_pos = curpos;
1941 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1942 type, obj_offset);
1943 if (!base_offset) {
1944 type = OBJ_BAD;
1945 goto out;
1947 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1948 if (*oi->sizep == 0) {
1949 type = OBJ_BAD;
1950 goto out;
1952 } else {
1953 *oi->sizep = size;
1957 if (oi->disk_sizep) {
1958 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1959 *oi->disk_sizep = revidx[1].offset - obj_offset;
1962 if (oi->typep) {
1963 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1964 if (*oi->typep < 0) {
1965 type = OBJ_BAD;
1966 goto out;
1970 if (oi->delta_base_sha1) {
1971 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1972 const unsigned char *base;
1974 base = get_delta_base_sha1(p, &w_curs, curpos,
1975 type, obj_offset);
1976 if (!base) {
1977 type = OBJ_BAD;
1978 goto out;
1981 hashcpy(oi->delta_base_sha1, base);
1982 } else
1983 hashclr(oi->delta_base_sha1);
1986 out:
1987 unuse_pack(&w_curs);
1988 return type;
1991 static void *unpack_compressed_entry(struct packed_git *p,
1992 struct pack_window **w_curs,
1993 off_t curpos,
1994 unsigned long size)
1996 int st;
1997 git_zstream stream;
1998 unsigned char *buffer, *in;
2000 buffer = xmallocz_gently(size);
2001 if (!buffer)
2002 return NULL;
2003 memset(&stream, 0, sizeof(stream));
2004 stream.next_out = buffer;
2005 stream.avail_out = size + 1;
2007 git_inflate_init(&stream);
2008 do {
2009 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2010 stream.next_in = in;
2011 st = git_inflate(&stream, Z_FINISH);
2012 if (!stream.avail_out)
2013 break; /* the payload is larger than it should be */
2014 curpos += stream.next_in - in;
2015 } while (st == Z_OK || st == Z_BUF_ERROR);
2016 git_inflate_end(&stream);
2017 if ((st != Z_STREAM_END) || stream.total_out != size) {
2018 free(buffer);
2019 return NULL;
2022 return buffer;
2025 #define MAX_DELTA_CACHE (256)
2027 static size_t delta_base_cached;
2029 static struct delta_base_cache_lru_list {
2030 struct delta_base_cache_lru_list *prev;
2031 struct delta_base_cache_lru_list *next;
2032 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2034 static struct delta_base_cache_entry {
2035 struct delta_base_cache_lru_list lru;
2036 void *data;
2037 struct packed_git *p;
2038 off_t base_offset;
2039 unsigned long size;
2040 enum object_type type;
2041 } delta_base_cache[MAX_DELTA_CACHE];
2043 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2045 unsigned long hash;
2047 hash = (unsigned long)p + (unsigned long)base_offset;
2048 hash += (hash >> 8) + (hash >> 16);
2049 return hash % MAX_DELTA_CACHE;
2052 static struct delta_base_cache_entry *
2053 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2055 unsigned long hash = pack_entry_hash(p, base_offset);
2056 return delta_base_cache + hash;
2059 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2060 struct packed_git *p, off_t base_offset)
2062 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2065 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2067 struct delta_base_cache_entry *ent;
2068 ent = get_delta_base_cache_entry(p, base_offset);
2069 return eq_delta_base_cache_entry(ent, p, base_offset);
2072 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2074 ent->data = NULL;
2075 ent->lru.next->prev = ent->lru.prev;
2076 ent->lru.prev->next = ent->lru.next;
2077 delta_base_cached -= ent->size;
2080 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2081 unsigned long *base_size, enum object_type *type, int keep_cache)
2083 struct delta_base_cache_entry *ent;
2084 void *ret;
2086 ent = get_delta_base_cache_entry(p, base_offset);
2088 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2089 return unpack_entry(p, base_offset, type, base_size);
2091 ret = ent->data;
2093 if (!keep_cache)
2094 clear_delta_base_cache_entry(ent);
2095 else
2096 ret = xmemdupz(ent->data, ent->size);
2097 *type = ent->type;
2098 *base_size = ent->size;
2099 return ret;
2102 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2104 if (ent->data) {
2105 free(ent->data);
2106 ent->data = NULL;
2107 ent->lru.next->prev = ent->lru.prev;
2108 ent->lru.prev->next = ent->lru.next;
2109 delta_base_cached -= ent->size;
2113 void clear_delta_base_cache(void)
2115 unsigned long p;
2116 for (p = 0; p < MAX_DELTA_CACHE; p++)
2117 release_delta_base_cache(&delta_base_cache[p]);
2120 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2121 void *base, unsigned long base_size, enum object_type type)
2123 unsigned long hash = pack_entry_hash(p, base_offset);
2124 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2125 struct delta_base_cache_lru_list *lru;
2127 release_delta_base_cache(ent);
2128 delta_base_cached += base_size;
2130 for (lru = delta_base_cache_lru.next;
2131 delta_base_cached > delta_base_cache_limit
2132 && lru != &delta_base_cache_lru;
2133 lru = lru->next) {
2134 struct delta_base_cache_entry *f = (void *)lru;
2135 if (f->type == OBJ_BLOB)
2136 release_delta_base_cache(f);
2138 for (lru = delta_base_cache_lru.next;
2139 delta_base_cached > delta_base_cache_limit
2140 && lru != &delta_base_cache_lru;
2141 lru = lru->next) {
2142 struct delta_base_cache_entry *f = (void *)lru;
2143 release_delta_base_cache(f);
2146 ent->p = p;
2147 ent->base_offset = base_offset;
2148 ent->type = type;
2149 ent->data = base;
2150 ent->size = base_size;
2151 ent->lru.next = &delta_base_cache_lru;
2152 ent->lru.prev = delta_base_cache_lru.prev;
2153 delta_base_cache_lru.prev->next = &ent->lru;
2154 delta_base_cache_lru.prev = &ent->lru;
2157 static void *read_object(const unsigned char *sha1, enum object_type *type,
2158 unsigned long *size);
2160 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2162 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2163 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2164 p->pack_name, (uintmax_t)obj_offset);
2167 int do_check_packed_object_crc;
2169 #define UNPACK_ENTRY_STACK_PREALLOC 64
2170 struct unpack_entry_stack_ent {
2171 off_t obj_offset;
2172 off_t curpos;
2173 unsigned long size;
2176 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2177 enum object_type *final_type, unsigned long *final_size)
2179 struct pack_window *w_curs = NULL;
2180 off_t curpos = obj_offset;
2181 void *data = NULL;
2182 unsigned long size;
2183 enum object_type type;
2184 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2185 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2186 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2187 int base_from_cache = 0;
2189 write_pack_access_log(p, obj_offset);
2191 /* PHASE 1: drill down to the innermost base object */
2192 for (;;) {
2193 off_t base_offset;
2194 int i;
2195 struct delta_base_cache_entry *ent;
2197 ent = get_delta_base_cache_entry(p, curpos);
2198 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2199 type = ent->type;
2200 data = ent->data;
2201 size = ent->size;
2202 clear_delta_base_cache_entry(ent);
2203 base_from_cache = 1;
2204 break;
2207 if (do_check_packed_object_crc && p->index_version > 1) {
2208 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2209 unsigned long len = revidx[1].offset - obj_offset;
2210 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2211 const unsigned char *sha1 =
2212 nth_packed_object_sha1(p, revidx->nr);
2213 error("bad packed object CRC for %s",
2214 sha1_to_hex(sha1));
2215 mark_bad_packed_object(p, sha1);
2216 unuse_pack(&w_curs);
2217 return NULL;
2221 type = unpack_object_header(p, &w_curs, &curpos, &size);
2222 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2223 break;
2225 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2226 if (!base_offset) {
2227 error("failed to validate delta base reference "
2228 "at offset %"PRIuMAX" from %s",
2229 (uintmax_t)curpos, p->pack_name);
2230 /* bail to phase 2, in hopes of recovery */
2231 data = NULL;
2232 break;
2235 /* push object, proceed to base */
2236 if (delta_stack_nr >= delta_stack_alloc
2237 && delta_stack == small_delta_stack) {
2238 delta_stack_alloc = alloc_nr(delta_stack_nr);
2239 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2240 memcpy(delta_stack, small_delta_stack,
2241 sizeof(*delta_stack)*delta_stack_nr);
2242 } else {
2243 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2245 i = delta_stack_nr++;
2246 delta_stack[i].obj_offset = obj_offset;
2247 delta_stack[i].curpos = curpos;
2248 delta_stack[i].size = size;
2250 curpos = obj_offset = base_offset;
2253 /* PHASE 2: handle the base */
2254 switch (type) {
2255 case OBJ_OFS_DELTA:
2256 case OBJ_REF_DELTA:
2257 if (data)
2258 die("BUG in unpack_entry: left loop at a valid delta");
2259 break;
2260 case OBJ_COMMIT:
2261 case OBJ_TREE:
2262 case OBJ_BLOB:
2263 case OBJ_TAG:
2264 if (!base_from_cache)
2265 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2266 break;
2267 default:
2268 data = NULL;
2269 error("unknown object type %i at offset %"PRIuMAX" in %s",
2270 type, (uintmax_t)obj_offset, p->pack_name);
2273 /* PHASE 3: apply deltas in order */
2275 /* invariants:
2276 * 'data' holds the base data, or NULL if there was corruption
2278 while (delta_stack_nr) {
2279 void *delta_data;
2280 void *base = data;
2281 unsigned long delta_size, base_size = size;
2282 int i;
2284 data = NULL;
2286 if (base)
2287 add_delta_base_cache(p, obj_offset, base, base_size, type);
2289 if (!base) {
2291 * We're probably in deep shit, but let's try to fetch
2292 * the required base anyway from another pack or loose.
2293 * This is costly but should happen only in the presence
2294 * of a corrupted pack, and is better than failing outright.
2296 struct revindex_entry *revidx;
2297 const unsigned char *base_sha1;
2298 revidx = find_pack_revindex(p, obj_offset);
2299 if (revidx) {
2300 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2301 error("failed to read delta base object %s"
2302 " at offset %"PRIuMAX" from %s",
2303 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2304 p->pack_name);
2305 mark_bad_packed_object(p, base_sha1);
2306 base = read_object(base_sha1, &type, &base_size);
2310 i = --delta_stack_nr;
2311 obj_offset = delta_stack[i].obj_offset;
2312 curpos = delta_stack[i].curpos;
2313 delta_size = delta_stack[i].size;
2315 if (!base)
2316 continue;
2318 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2320 if (!delta_data) {
2321 error("failed to unpack compressed delta "
2322 "at offset %"PRIuMAX" from %s",
2323 (uintmax_t)curpos, p->pack_name);
2324 data = NULL;
2325 continue;
2328 data = patch_delta(base, base_size,
2329 delta_data, delta_size,
2330 &size);
2333 * We could not apply the delta; warn the user, but keep going.
2334 * Our failure will be noticed either in the next iteration of
2335 * the loop, or if this is the final delta, in the caller when
2336 * we return NULL. Those code paths will take care of making
2337 * a more explicit warning and retrying with another copy of
2338 * the object.
2340 if (!data)
2341 error("failed to apply delta");
2343 free(delta_data);
2346 *final_type = type;
2347 *final_size = size;
2349 unuse_pack(&w_curs);
2351 if (delta_stack != small_delta_stack)
2352 free(delta_stack);
2354 return data;
2357 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2358 uint32_t n)
2360 const unsigned char *index = p->index_data;
2361 if (!index) {
2362 if (open_pack_index(p))
2363 return NULL;
2364 index = p->index_data;
2366 if (n >= p->num_objects)
2367 return NULL;
2368 index += 4 * 256;
2369 if (p->index_version == 1) {
2370 return index + 24 * n + 4;
2371 } else {
2372 index += 8;
2373 return index + 20 * n;
2377 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2379 const unsigned char *index = p->index_data;
2380 index += 4 * 256;
2381 if (p->index_version == 1) {
2382 return ntohl(*((uint32_t *)(index + 24 * n)));
2383 } else {
2384 uint32_t off;
2385 index += 8 + p->num_objects * (20 + 4);
2386 off = ntohl(*((uint32_t *)(index + 4 * n)));
2387 if (!(off & 0x80000000))
2388 return off;
2389 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2390 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2391 ntohl(*((uint32_t *)(index + 4)));
2395 off_t find_pack_entry_one(const unsigned char *sha1,
2396 struct packed_git *p)
2398 const uint32_t *level1_ofs = p->index_data;
2399 const unsigned char *index = p->index_data;
2400 unsigned hi, lo, stride;
2401 static int use_lookup = -1;
2402 static int debug_lookup = -1;
2404 if (debug_lookup < 0)
2405 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2407 if (!index) {
2408 if (open_pack_index(p))
2409 return 0;
2410 level1_ofs = p->index_data;
2411 index = p->index_data;
2413 if (p->index_version > 1) {
2414 level1_ofs += 2;
2415 index += 8;
2417 index += 4 * 256;
2418 hi = ntohl(level1_ofs[*sha1]);
2419 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2420 if (p->index_version > 1) {
2421 stride = 20;
2422 } else {
2423 stride = 24;
2424 index += 4;
2427 if (debug_lookup)
2428 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2429 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2431 if (use_lookup < 0)
2432 use_lookup = !!getenv("GIT_USE_LOOKUP");
2433 if (use_lookup) {
2434 int pos = sha1_entry_pos(index, stride, 0,
2435 lo, hi, p->num_objects, sha1);
2436 if (pos < 0)
2437 return 0;
2438 return nth_packed_object_offset(p, pos);
2441 do {
2442 unsigned mi = (lo + hi) / 2;
2443 int cmp = hashcmp(index + mi * stride, sha1);
2445 if (debug_lookup)
2446 printf("lo %u hi %u rg %u mi %u\n",
2447 lo, hi, hi - lo, mi);
2448 if (!cmp)
2449 return nth_packed_object_offset(p, mi);
2450 if (cmp > 0)
2451 hi = mi;
2452 else
2453 lo = mi+1;
2454 } while (lo < hi);
2455 return 0;
2458 int is_pack_valid(struct packed_git *p)
2460 /* An already open pack is known to be valid. */
2461 if (p->pack_fd != -1)
2462 return 1;
2464 /* If the pack has one window completely covering the
2465 * file size, the pack is known to be valid even if
2466 * the descriptor is not currently open.
2468 if (p->windows) {
2469 struct pack_window *w = p->windows;
2471 if (!w->offset && w->len == p->pack_size)
2472 return 1;
2475 /* Force the pack to open to prove its valid. */
2476 return !open_packed_git(p);
2479 static int fill_pack_entry(const unsigned char *sha1,
2480 struct pack_entry *e,
2481 struct packed_git *p)
2483 off_t offset;
2485 if (p->num_bad_objects) {
2486 unsigned i;
2487 for (i = 0; i < p->num_bad_objects; i++)
2488 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2489 return 0;
2492 offset = find_pack_entry_one(sha1, p);
2493 if (!offset)
2494 return 0;
2497 * We are about to tell the caller where they can locate the
2498 * requested object. We better make sure the packfile is
2499 * still here and can be accessed before supplying that
2500 * answer, as it may have been deleted since the index was
2501 * loaded!
2503 if (!is_pack_valid(p))
2504 return 0;
2505 e->offset = offset;
2506 e->p = p;
2507 hashcpy(e->sha1, sha1);
2508 return 1;
2512 * Iff a pack file contains the object named by sha1, return true and
2513 * store its location to e.
2515 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2517 struct packed_git *p;
2519 prepare_packed_git();
2520 if (!packed_git)
2521 return 0;
2523 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2524 return 1;
2526 for (p = packed_git; p; p = p->next) {
2527 if (p == last_found_pack)
2528 continue; /* we already checked this one */
2530 if (fill_pack_entry(sha1, e, p)) {
2531 last_found_pack = p;
2532 return 1;
2535 return 0;
2538 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2539 struct packed_git *packs)
2541 struct packed_git *p;
2543 for (p = packs; p; p = p->next) {
2544 if (find_pack_entry_one(sha1, p))
2545 return p;
2547 return NULL;
2551 static int sha1_loose_object_info(const unsigned char *sha1,
2552 struct object_info *oi)
2554 int status;
2555 unsigned long mapsize, size;
2556 void *map;
2557 git_zstream stream;
2558 char hdr[32];
2560 if (oi->delta_base_sha1)
2561 hashclr(oi->delta_base_sha1);
2564 * If we don't care about type or size, then we don't
2565 * need to look inside the object at all. Note that we
2566 * do not optimize out the stat call, even if the
2567 * caller doesn't care about the disk-size, since our
2568 * return value implicitly indicates whether the
2569 * object even exists.
2571 if (!oi->typep && !oi->sizep) {
2572 struct stat st;
2573 if (stat_sha1_file(sha1, &st) < 0)
2574 return -1;
2575 if (oi->disk_sizep)
2576 *oi->disk_sizep = st.st_size;
2577 return 0;
2580 map = map_sha1_file(sha1, &mapsize);
2581 if (!map)
2582 return -1;
2583 if (oi->disk_sizep)
2584 *oi->disk_sizep = mapsize;
2585 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2586 status = error("unable to unpack %s header",
2587 sha1_to_hex(sha1));
2588 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2589 status = error("unable to parse %s header", sha1_to_hex(sha1));
2590 else if (oi->sizep)
2591 *oi->sizep = size;
2592 git_inflate_end(&stream);
2593 munmap(map, mapsize);
2594 if (oi->typep)
2595 *oi->typep = status;
2596 return 0;
2599 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2601 struct cached_object *co;
2602 struct pack_entry e;
2603 int rtype;
2604 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2606 co = find_cached_object(real);
2607 if (co) {
2608 if (oi->typep)
2609 *(oi->typep) = co->type;
2610 if (oi->sizep)
2611 *(oi->sizep) = co->size;
2612 if (oi->disk_sizep)
2613 *(oi->disk_sizep) = 0;
2614 if (oi->delta_base_sha1)
2615 hashclr(oi->delta_base_sha1);
2616 oi->whence = OI_CACHED;
2617 return 0;
2620 if (!find_pack_entry(real, &e)) {
2621 /* Most likely it's a loose object. */
2622 if (!sha1_loose_object_info(real, oi)) {
2623 oi->whence = OI_LOOSE;
2624 return 0;
2627 /* Not a loose object; someone else may have just packed it. */
2628 reprepare_packed_git();
2629 if (!find_pack_entry(real, &e))
2630 return -1;
2633 rtype = packed_object_info(e.p, e.offset, oi);
2634 if (rtype < 0) {
2635 mark_bad_packed_object(e.p, real);
2636 return sha1_object_info_extended(real, oi, 0);
2637 } else if (in_delta_base_cache(e.p, e.offset)) {
2638 oi->whence = OI_DBCACHED;
2639 } else {
2640 oi->whence = OI_PACKED;
2641 oi->u.packed.offset = e.offset;
2642 oi->u.packed.pack = e.p;
2643 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2644 rtype == OBJ_OFS_DELTA);
2647 return 0;
2650 /* returns enum object_type or negative */
2651 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2653 enum object_type type;
2654 struct object_info oi = {NULL};
2656 oi.typep = &type;
2657 oi.sizep = sizep;
2658 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2659 return -1;
2660 return type;
2663 static void *read_packed_sha1(const unsigned char *sha1,
2664 enum object_type *type, unsigned long *size)
2666 struct pack_entry e;
2667 void *data;
2669 if (!find_pack_entry(sha1, &e))
2670 return NULL;
2671 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2672 if (!data) {
2674 * We're probably in deep shit, but let's try to fetch
2675 * the required object anyway from another pack or loose.
2676 * This should happen only in the presence of a corrupted
2677 * pack, and is better than failing outright.
2679 error("failed to read object %s at offset %"PRIuMAX" from %s",
2680 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2681 mark_bad_packed_object(e.p, sha1);
2682 data = read_object(sha1, type, size);
2684 return data;
2687 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2688 unsigned char *sha1)
2690 struct cached_object *co;
2692 hash_sha1_file(buf, len, typename(type), sha1);
2693 if (has_sha1_file(sha1) || find_cached_object(sha1))
2694 return 0;
2695 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2696 co = &cached_objects[cached_object_nr++];
2697 co->size = len;
2698 co->type = type;
2699 co->buf = xmalloc(len);
2700 memcpy(co->buf, buf, len);
2701 hashcpy(co->sha1, sha1);
2702 return 0;
2705 static void *read_object(const unsigned char *sha1, enum object_type *type,
2706 unsigned long *size)
2708 unsigned long mapsize;
2709 void *map, *buf;
2710 struct cached_object *co;
2712 co = find_cached_object(sha1);
2713 if (co) {
2714 *type = co->type;
2715 *size = co->size;
2716 return xmemdupz(co->buf, co->size);
2719 buf = read_packed_sha1(sha1, type, size);
2720 if (buf)
2721 return buf;
2722 map = map_sha1_file(sha1, &mapsize);
2723 if (map) {
2724 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2725 munmap(map, mapsize);
2726 return buf;
2728 reprepare_packed_git();
2729 return read_packed_sha1(sha1, type, size);
2733 * This function dies on corrupt objects; the callers who want to
2734 * deal with them should arrange to call read_object() and give error
2735 * messages themselves.
2737 void *read_sha1_file_extended(const unsigned char *sha1,
2738 enum object_type *type,
2739 unsigned long *size,
2740 unsigned flag)
2742 void *data;
2743 const struct packed_git *p;
2744 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2746 errno = 0;
2747 data = read_object(repl, type, size);
2748 if (data)
2749 return data;
2751 if (errno && errno != ENOENT)
2752 die_errno("failed to read object %s", sha1_to_hex(sha1));
2754 /* die if we replaced an object with one that does not exist */
2755 if (repl != sha1)
2756 die("replacement %s not found for %s",
2757 sha1_to_hex(repl), sha1_to_hex(sha1));
2759 if (has_loose_object(repl)) {
2760 const char *path = sha1_file_name(sha1);
2762 die("loose object %s (stored in %s) is corrupt",
2763 sha1_to_hex(repl), path);
2766 if ((p = has_packed_and_bad(repl)) != NULL)
2767 die("packed object %s (stored in %s) is corrupt",
2768 sha1_to_hex(repl), p->pack_name);
2770 return NULL;
2773 void *read_object_with_reference(const unsigned char *sha1,
2774 const char *required_type_name,
2775 unsigned long *size,
2776 unsigned char *actual_sha1_return)
2778 enum object_type type, required_type;
2779 void *buffer;
2780 unsigned long isize;
2781 unsigned char actual_sha1[20];
2783 required_type = type_from_string(required_type_name);
2784 hashcpy(actual_sha1, sha1);
2785 while (1) {
2786 int ref_length = -1;
2787 const char *ref_type = NULL;
2789 buffer = read_sha1_file(actual_sha1, &type, &isize);
2790 if (!buffer)
2791 return NULL;
2792 if (type == required_type) {
2793 *size = isize;
2794 if (actual_sha1_return)
2795 hashcpy(actual_sha1_return, actual_sha1);
2796 return buffer;
2798 /* Handle references */
2799 else if (type == OBJ_COMMIT)
2800 ref_type = "tree ";
2801 else if (type == OBJ_TAG)
2802 ref_type = "object ";
2803 else {
2804 free(buffer);
2805 return NULL;
2807 ref_length = strlen(ref_type);
2809 if (ref_length + 40 > isize ||
2810 memcmp(buffer, ref_type, ref_length) ||
2811 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2812 free(buffer);
2813 return NULL;
2815 free(buffer);
2816 /* Now we have the ID of the referred-to object in
2817 * actual_sha1. Check again. */
2821 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2822 const char *type, unsigned char *sha1,
2823 char *hdr, int *hdrlen)
2825 git_SHA_CTX c;
2827 /* Generate the header */
2828 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2830 /* Sha1.. */
2831 git_SHA1_Init(&c);
2832 git_SHA1_Update(&c, hdr, *hdrlen);
2833 git_SHA1_Update(&c, buf, len);
2834 git_SHA1_Final(sha1, &c);
2838 * Move the just written object into its final resting place.
2839 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2840 * "moving" is only a part of what it does, when no patch between
2841 * master to pu changes the call sites of this function.
2843 int move_temp_to_file(const char *tmpfile, const char *filename)
2845 int ret = 0;
2847 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2848 goto try_rename;
2849 else if (link(tmpfile, filename))
2850 ret = errno;
2853 * Coda hack - coda doesn't like cross-directory links,
2854 * so we fall back to a rename, which will mean that it
2855 * won't be able to check collisions, but that's not a
2856 * big deal.
2858 * The same holds for FAT formatted media.
2860 * When this succeeds, we just return. We have nothing
2861 * left to unlink.
2863 if (ret && ret != EEXIST) {
2864 try_rename:
2865 if (!rename(tmpfile, filename))
2866 goto out;
2867 ret = errno;
2869 unlink_or_warn(tmpfile);
2870 if (ret) {
2871 if (ret != EEXIST) {
2872 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2874 /* FIXME!!! Collision check here ? */
2877 out:
2878 if (adjust_shared_perm(filename))
2879 return error("unable to set permission to '%s'", filename);
2880 return 0;
2883 static int write_buffer(int fd, const void *buf, size_t len)
2885 if (write_in_full(fd, buf, len) < 0)
2886 return error("file write error (%s)", strerror(errno));
2887 return 0;
2890 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2891 unsigned char *sha1)
2893 char hdr[32];
2894 int hdrlen;
2895 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2896 return 0;
2899 /* Finalize a file on disk, and close it. */
2900 static void close_sha1_file(int fd)
2902 if (fsync_object_files)
2903 fsync_or_die(fd, "sha1 file");
2904 if (close(fd) != 0)
2905 die_errno("error when closing sha1 file");
2908 /* Size of directory component, including the ending '/' */
2909 static inline int directory_size(const char *filename)
2911 const char *s = strrchr(filename, '/');
2912 if (!s)
2913 return 0;
2914 return s - filename + 1;
2918 * This creates a temporary file in the same directory as the final
2919 * 'filename'
2921 * We want to avoid cross-directory filename renames, because those
2922 * can have problems on various filesystems (FAT, NFS, Coda).
2924 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2926 int fd, dirlen = directory_size(filename);
2928 if (dirlen + 20 > bufsiz) {
2929 errno = ENAMETOOLONG;
2930 return -1;
2932 memcpy(buffer, filename, dirlen);
2933 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2934 fd = git_mkstemp_mode(buffer, 0444);
2935 if (fd < 0 && dirlen && errno == ENOENT) {
2936 /* Make sure the directory exists */
2937 memcpy(buffer, filename, dirlen);
2938 buffer[dirlen-1] = 0;
2939 if (mkdir(buffer, 0777) && errno != EEXIST)
2940 return -1;
2941 if (adjust_shared_perm(buffer))
2942 return -1;
2944 /* Try again */
2945 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2946 fd = git_mkstemp_mode(buffer, 0444);
2948 return fd;
2951 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2952 const void *buf, unsigned long len, time_t mtime)
2954 int fd, ret;
2955 unsigned char compressed[4096];
2956 git_zstream stream;
2957 git_SHA_CTX c;
2958 unsigned char parano_sha1[20];
2959 static char tmp_file[PATH_MAX];
2960 const char *filename = sha1_file_name(sha1);
2962 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2963 if (fd < 0) {
2964 if (errno == EACCES)
2965 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2966 else
2967 return error("unable to create temporary file: %s", strerror(errno));
2970 /* Set it up */
2971 git_deflate_init(&stream, zlib_compression_level);
2972 stream.next_out = compressed;
2973 stream.avail_out = sizeof(compressed);
2974 git_SHA1_Init(&c);
2976 /* First header.. */
2977 stream.next_in = (unsigned char *)hdr;
2978 stream.avail_in = hdrlen;
2979 while (git_deflate(&stream, 0) == Z_OK)
2980 ; /* nothing */
2981 git_SHA1_Update(&c, hdr, hdrlen);
2983 /* Then the data itself.. */
2984 stream.next_in = (void *)buf;
2985 stream.avail_in = len;
2986 do {
2987 unsigned char *in0 = stream.next_in;
2988 ret = git_deflate(&stream, Z_FINISH);
2989 git_SHA1_Update(&c, in0, stream.next_in - in0);
2990 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2991 die("unable to write sha1 file");
2992 stream.next_out = compressed;
2993 stream.avail_out = sizeof(compressed);
2994 } while (ret == Z_OK);
2996 if (ret != Z_STREAM_END)
2997 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2998 ret = git_deflate_end_gently(&stream);
2999 if (ret != Z_OK)
3000 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3001 git_SHA1_Final(parano_sha1, &c);
3002 if (hashcmp(sha1, parano_sha1) != 0)
3003 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3005 close_sha1_file(fd);
3007 if (mtime) {
3008 struct utimbuf utb;
3009 utb.actime = mtime;
3010 utb.modtime = mtime;
3011 if (utime(tmp_file, &utb) < 0)
3012 warning("failed utime() on %s: %s",
3013 tmp_file, strerror(errno));
3016 return move_temp_to_file(tmp_file, filename);
3019 static int freshen_loose_object(const unsigned char *sha1)
3021 return check_and_freshen(sha1, 1);
3024 static int freshen_packed_object(const unsigned char *sha1)
3026 struct pack_entry e;
3027 if (!find_pack_entry(sha1, &e))
3028 return 0;
3029 if (e.p->freshened)
3030 return 1;
3031 if (!freshen_file(e.p->pack_name))
3032 return 0;
3033 e.p->freshened = 1;
3034 return 1;
3037 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3039 char hdr[32];
3040 int hdrlen;
3042 /* Normally if we have it in the pack then we do not bother writing
3043 * it out into .git/objects/??/?{38} file.
3045 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3046 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3047 return 0;
3048 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3051 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3052 unsigned char *sha1, unsigned flags)
3054 char *header;
3055 int hdrlen, status = 0;
3057 /* type string, SP, %lu of the length plus NUL must fit this */
3058 header = xmalloc(strlen(type) + 32);
3059 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3061 if (!(flags & HASH_WRITE_OBJECT))
3062 goto cleanup;
3063 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3064 goto cleanup;
3065 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3067 cleanup:
3068 free(header);
3069 return status;
3072 int force_object_loose(const unsigned char *sha1, time_t mtime)
3074 void *buf;
3075 unsigned long len;
3076 enum object_type type;
3077 char hdr[32];
3078 int hdrlen;
3079 int ret;
3081 if (has_loose_object(sha1))
3082 return 0;
3083 buf = read_packed_sha1(sha1, &type, &len);
3084 if (!buf)
3085 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3086 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3087 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3088 free(buf);
3090 return ret;
3093 int has_pack_index(const unsigned char *sha1)
3095 struct stat st;
3096 if (stat(sha1_pack_index_name(sha1), &st))
3097 return 0;
3098 return 1;
3101 int has_sha1_pack(const unsigned char *sha1)
3103 struct pack_entry e;
3104 return find_pack_entry(sha1, &e);
3107 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3109 struct pack_entry e;
3111 if (find_pack_entry(sha1, &e))
3112 return 1;
3113 if (has_loose_object(sha1))
3114 return 1;
3115 if (flags & HAS_SHA1_QUICK)
3116 return 0;
3117 reprepare_packed_git();
3118 return find_pack_entry(sha1, &e);
3121 static void check_tree(const void *buf, size_t size)
3123 struct tree_desc desc;
3124 struct name_entry entry;
3126 init_tree_desc(&desc, buf, size);
3127 while (tree_entry(&desc, &entry))
3128 /* do nothing
3129 * tree_entry() will die() on malformed entries */
3133 static void check_commit(const void *buf, size_t size)
3135 struct commit c;
3136 memset(&c, 0, sizeof(c));
3137 if (parse_commit_buffer(&c, buf, size))
3138 die("corrupt commit");
3141 static void check_tag(const void *buf, size_t size)
3143 struct tag t;
3144 memset(&t, 0, sizeof(t));
3145 if (parse_tag_buffer(&t, buf, size))
3146 die("corrupt tag");
3149 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3150 enum object_type type,
3151 const char *path, unsigned flags)
3153 int ret, re_allocated = 0;
3154 int write_object = flags & HASH_WRITE_OBJECT;
3156 if (!type)
3157 type = OBJ_BLOB;
3160 * Convert blobs to git internal format
3162 if ((type == OBJ_BLOB) && path) {
3163 struct strbuf nbuf = STRBUF_INIT;
3164 if (convert_to_git(path, buf, size, &nbuf,
3165 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3166 buf = strbuf_detach(&nbuf, &size);
3167 re_allocated = 1;
3170 if (flags & HASH_FORMAT_CHECK) {
3171 if (type == OBJ_TREE)
3172 check_tree(buf, size);
3173 if (type == OBJ_COMMIT)
3174 check_commit(buf, size);
3175 if (type == OBJ_TAG)
3176 check_tag(buf, size);
3179 if (write_object)
3180 ret = write_sha1_file(buf, size, typename(type), sha1);
3181 else
3182 ret = hash_sha1_file(buf, size, typename(type), sha1);
3183 if (re_allocated)
3184 free(buf);
3185 return ret;
3188 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3189 const char *path, unsigned flags)
3191 int ret;
3192 const int write_object = flags & HASH_WRITE_OBJECT;
3193 struct strbuf sbuf = STRBUF_INIT;
3195 assert(path);
3196 assert(would_convert_to_git_filter_fd(path));
3198 convert_to_git_filter_fd(path, fd, &sbuf,
3199 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3201 if (write_object)
3202 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3203 sha1);
3204 else
3205 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3206 sha1);
3207 strbuf_release(&sbuf);
3208 return ret;
3211 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3212 const char *path, unsigned flags)
3214 struct strbuf sbuf = STRBUF_INIT;
3215 int ret;
3217 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3218 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3219 else
3220 ret = -1;
3221 strbuf_release(&sbuf);
3222 return ret;
3225 #define SMALL_FILE_SIZE (32*1024)
3227 static int index_core(unsigned char *sha1, int fd, size_t size,
3228 enum object_type type, const char *path,
3229 unsigned flags)
3231 int ret;
3233 if (!size) {
3234 ret = index_mem(sha1, "", size, type, path, flags);
3235 } else if (size <= SMALL_FILE_SIZE) {
3236 char *buf = xmalloc(size);
3237 if (size == read_in_full(fd, buf, size))
3238 ret = index_mem(sha1, buf, size, type, path, flags);
3239 else
3240 ret = error("short read %s", strerror(errno));
3241 free(buf);
3242 } else {
3243 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3244 ret = index_mem(sha1, buf, size, type, path, flags);
3245 munmap(buf, size);
3247 return ret;
3251 * This creates one packfile per large blob unless bulk-checkin
3252 * machinery is "plugged".
3254 * This also bypasses the usual "convert-to-git" dance, and that is on
3255 * purpose. We could write a streaming version of the converting
3256 * functions and insert that before feeding the data to fast-import
3257 * (or equivalent in-core API described above). However, that is
3258 * somewhat complicated, as we do not know the size of the filter
3259 * result, which we need to know beforehand when writing a git object.
3260 * Since the primary motivation for trying to stream from the working
3261 * tree file and to avoid mmaping it in core is to deal with large
3262 * binary blobs, they generally do not want to get any conversion, and
3263 * callers should avoid this code path when filters are requested.
3265 static int index_stream(unsigned char *sha1, int fd, size_t size,
3266 enum object_type type, const char *path,
3267 unsigned flags)
3269 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3272 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3273 enum object_type type, const char *path, unsigned flags)
3275 int ret;
3278 * Call xsize_t() only when needed to avoid potentially unnecessary
3279 * die() for large files.
3281 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3282 ret = index_stream_convert_blob(sha1, fd, path, flags);
3283 else if (!S_ISREG(st->st_mode))
3284 ret = index_pipe(sha1, fd, type, path, flags);
3285 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3286 (path && would_convert_to_git(path)))
3287 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3288 flags);
3289 else
3290 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3291 flags);
3292 close(fd);
3293 return ret;
3296 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3298 int fd;
3299 struct strbuf sb = STRBUF_INIT;
3301 switch (st->st_mode & S_IFMT) {
3302 case S_IFREG:
3303 fd = open(path, O_RDONLY);
3304 if (fd < 0)
3305 return error("open(\"%s\"): %s", path,
3306 strerror(errno));
3307 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3308 return error("%s: failed to insert into database",
3309 path);
3310 break;
3311 case S_IFLNK:
3312 if (strbuf_readlink(&sb, path, st->st_size)) {
3313 char *errstr = strerror(errno);
3314 return error("readlink(\"%s\"): %s", path,
3315 errstr);
3317 if (!(flags & HASH_WRITE_OBJECT))
3318 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3319 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3320 return error("%s: failed to insert into database",
3321 path);
3322 strbuf_release(&sb);
3323 break;
3324 case S_IFDIR:
3325 return resolve_gitlink_ref(path, "HEAD", sha1);
3326 default:
3327 return error("%s: unsupported file type", path);
3329 return 0;
3332 int read_pack_header(int fd, struct pack_header *header)
3334 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3335 /* "eof before pack header was fully read" */
3336 return PH_ERROR_EOF;
3338 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3339 /* "protocol error (pack signature mismatch detected)" */
3340 return PH_ERROR_PACK_SIGNATURE;
3341 if (!pack_version_ok(header->hdr_version))
3342 /* "protocol error (pack version unsupported)" */
3343 return PH_ERROR_PROTOCOL;
3344 return 0;
3347 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3349 enum object_type type = sha1_object_info(sha1, NULL);
3350 if (type < 0)
3351 die("%s is not a valid object", sha1_to_hex(sha1));
3352 if (type != expect)
3353 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3354 typename(expect));
3357 static int for_each_file_in_obj_subdir(int subdir_nr,
3358 struct strbuf *path,
3359 each_loose_object_fn obj_cb,
3360 each_loose_cruft_fn cruft_cb,
3361 each_loose_subdir_fn subdir_cb,
3362 void *data)
3364 size_t baselen = path->len;
3365 DIR *dir = opendir(path->buf);
3366 struct dirent *de;
3367 int r = 0;
3369 if (!dir) {
3370 if (errno == ENOENT)
3371 return 0;
3372 return error("unable to open %s: %s", path->buf, strerror(errno));
3375 while ((de = readdir(dir))) {
3376 if (is_dot_or_dotdot(de->d_name))
3377 continue;
3379 strbuf_setlen(path, baselen);
3380 strbuf_addf(path, "/%s", de->d_name);
3382 if (strlen(de->d_name) == 38) {
3383 char hex[41];
3384 unsigned char sha1[20];
3386 snprintf(hex, sizeof(hex), "%02x%s",
3387 subdir_nr, de->d_name);
3388 if (!get_sha1_hex(hex, sha1)) {
3389 if (obj_cb) {
3390 r = obj_cb(sha1, path->buf, data);
3391 if (r)
3392 break;
3394 continue;
3398 if (cruft_cb) {
3399 r = cruft_cb(de->d_name, path->buf, data);
3400 if (r)
3401 break;
3404 strbuf_setlen(path, baselen);
3406 if (!r && subdir_cb)
3407 r = subdir_cb(subdir_nr, path->buf, data);
3409 closedir(dir);
3410 return r;
3413 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3414 each_loose_object_fn obj_cb,
3415 each_loose_cruft_fn cruft_cb,
3416 each_loose_subdir_fn subdir_cb,
3417 void *data)
3419 size_t baselen = path->len;
3420 int r = 0;
3421 int i;
3423 for (i = 0; i < 256; i++) {
3424 strbuf_addf(path, "/%02x", i);
3425 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3426 subdir_cb, data);
3427 strbuf_setlen(path, baselen);
3428 if (r)
3429 break;
3432 return r;
3435 int for_each_loose_file_in_objdir(const char *path,
3436 each_loose_object_fn obj_cb,
3437 each_loose_cruft_fn cruft_cb,
3438 each_loose_subdir_fn subdir_cb,
3439 void *data)
3441 struct strbuf buf = STRBUF_INIT;
3442 int r;
3444 strbuf_addstr(&buf, path);
3445 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3446 subdir_cb, data);
3447 strbuf_release(&buf);
3449 return r;
3452 struct loose_alt_odb_data {
3453 each_loose_object_fn *cb;
3454 void *data;
3457 static int loose_from_alt_odb(struct alternate_object_database *alt,
3458 void *vdata)
3460 struct loose_alt_odb_data *data = vdata;
3461 struct strbuf buf = STRBUF_INIT;
3462 int r;
3464 /* copy base not including trailing '/' */
3465 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3466 r = for_each_loose_file_in_objdir_buf(&buf,
3467 data->cb, NULL, NULL,
3468 data->data);
3469 strbuf_release(&buf);
3470 return r;
3473 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3475 struct loose_alt_odb_data alt;
3476 int r;
3478 r = for_each_loose_file_in_objdir(get_object_directory(),
3479 cb, NULL, NULL, data);
3480 if (r)
3481 return r;
3483 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3484 return 0;
3486 alt.cb = cb;
3487 alt.data = data;
3488 return foreach_alt_odb(loose_from_alt_odb, &alt);
3491 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3493 uint32_t i;
3494 int r = 0;
3496 for (i = 0; i < p->num_objects; i++) {
3497 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3499 if (!sha1)
3500 return error("unable to get sha1 of object %u in %s",
3501 i, p->pack_name);
3503 r = cb(sha1, p, i, data);
3504 if (r)
3505 break;
3507 return r;
3510 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3512 struct packed_git *p;
3513 int r = 0;
3515 prepare_packed_git();
3516 for (p = packed_git; p; p = p->next) {
3517 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3518 continue;
3519 r = for_each_object_in_pack(p, cb, data);
3520 if (r)
3521 break;
3523 return r;