setup: prepare setup_discovered_git_dir() for the root directory
[git.git] / sha1_file.c
blobec957db5e1c2620b4a95e4f4d028f01794cef02a
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"
26 #include "mru.h"
27 #include "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
31 #define SZ_FMT PRIuMAX
32 static inline uintmax_t sz_fmt(size_t s) { return s; }
34 const unsigned char null_sha1[20];
35 const struct object_id null_oid;
36 const struct object_id empty_tree_oid = {
37 EMPTY_TREE_SHA1_BIN_LITERAL
39 const struct object_id empty_blob_oid = {
40 EMPTY_BLOB_SHA1_BIN_LITERAL
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
49 static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54 } *cached_objects;
55 static int cached_object_nr, cached_object_alloc;
57 static struct cached_object empty_tree = {
58 EMPTY_TREE_SHA1_BIN_LITERAL,
59 OBJ_TREE,
60 "",
64 static struct cached_object *find_cached_object(const unsigned char *sha1)
66 int i;
67 struct cached_object *co = cached_objects;
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
78 int mkdir_in_gitdir(const char *path)
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
85 if (errno != EEXIST)
86 return -1;
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
102 strbuf_release(&sb);
104 return adjust_shared_perm(path);
107 enum scld_error safe_create_leading_directories(char *path)
109 char *next_component = path + offset_1st_component(path);
110 enum scld_error ret = SCLD_OK;
112 while (ret == SCLD_OK && next_component) {
113 struct stat st;
114 char *slash = next_component, slash_character;
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
119 if (!*slash)
120 break;
122 next_component = slash + 1;
123 while (is_dir_sep(*next_component))
124 next_component++;
125 if (!*next_component)
126 break;
128 slash_character = *slash;
129 *slash = '\0';
130 if (!stat(path, &st)) {
131 /* path exists */
132 if (!S_ISDIR(st.st_mode))
133 ret = SCLD_EXISTS;
134 } else if (mkdir(path, 0777)) {
135 if (errno == EEXIST &&
136 !stat(path, &st) && S_ISDIR(st.st_mode))
137 ; /* somebody created it since we checked */
138 else if (errno == ENOENT)
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
148 ret = SCLD_VANISHED;
149 else
150 ret = SCLD_FAILED;
151 } else if (adjust_shared_perm(path)) {
152 ret = SCLD_PERMS;
154 *slash = slash_character;
156 return ret;
159 enum scld_error safe_create_leading_directories_const(const char *path)
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf = xstrdup(path);
163 enum scld_error result = safe_create_leading_directories(buf);
164 free(buf);
165 return result;
168 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
170 int i;
171 for (i = 0; i < 20; i++) {
172 static char hex[] = "0123456789abcdef";
173 unsigned int val = sha1[i];
174 strbuf_addch(buf, hex[val >> 4]);
175 strbuf_addch(buf, hex[val & 0xf]);
176 if (!i)
177 strbuf_addch(buf, '/');
181 const char *sha1_file_name(const unsigned char *sha1)
183 static struct strbuf buf = STRBUF_INIT;
185 strbuf_reset(&buf);
186 strbuf_addf(&buf, "%s/", get_object_directory());
188 fill_sha1_path(&buf, sha1);
189 return buf.buf;
192 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
194 strbuf_setlen(&alt->scratch, alt->base_len);
195 return &alt->scratch;
198 static const char *alt_sha1_path(struct alternate_object_database *alt,
199 const unsigned char *sha1)
201 struct strbuf *buf = alt_scratch_buf(alt);
202 fill_sha1_path(buf, sha1);
203 return buf->buf;
207 * Return the name of the pack or index file with the specified sha1
208 * in its filename. *base and *name are scratch space that must be
209 * provided by the caller. which should be "pack" or "idx".
211 static char *sha1_get_pack_name(const unsigned char *sha1,
212 struct strbuf *buf,
213 const char *which)
215 strbuf_reset(buf);
216 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
217 sha1_to_hex(sha1), which);
218 return buf->buf;
221 char *sha1_pack_name(const unsigned char *sha1)
223 static struct strbuf buf = STRBUF_INIT;
224 return sha1_get_pack_name(sha1, &buf, "pack");
227 char *sha1_pack_index_name(const unsigned char *sha1)
229 static struct strbuf buf = STRBUF_INIT;
230 return sha1_get_pack_name(sha1, &buf, "idx");
233 struct alternate_object_database *alt_odb_list;
234 static struct alternate_object_database **alt_odb_tail;
237 * Return non-zero iff the path is usable as an alternate object database.
239 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
241 struct alternate_object_database *alt;
243 /* Detect cases where alternate disappeared */
244 if (!is_directory(path->buf)) {
245 error("object directory %s does not exist; "
246 "check .git/objects/info/alternates.",
247 path->buf);
248 return 0;
252 * Prevent the common mistake of listing the same
253 * thing twice, or object directory itself.
255 for (alt = alt_odb_list; alt; alt = alt->next) {
256 if (!fspathcmp(path->buf, alt->path))
257 return 0;
259 if (!fspathcmp(path->buf, normalized_objdir))
260 return 0;
262 return 1;
266 * Prepare alternate object database registry.
268 * The variable alt_odb_list points at the list of struct
269 * alternate_object_database. The elements on this list come from
270 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
271 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
272 * whose contents is similar to that environment variable but can be
273 * LF separated. Its base points at a statically allocated buffer that
274 * contains "/the/directory/corresponding/to/.git/objects/...", while
275 * its name points just after the slash at the end of ".git/objects/"
276 * in the example above, and has enough space to hold 40-byte hex
277 * SHA1, an extra slash for the first level indirection, and the
278 * terminating NUL.
280 static int link_alt_odb_entry(const char *entry, const char *relative_base,
281 int depth, const char *normalized_objdir)
283 struct alternate_object_database *ent;
284 struct strbuf pathbuf = STRBUF_INIT;
286 if (!is_absolute_path(entry) && relative_base) {
287 strbuf_realpath(&pathbuf, relative_base, 1);
288 strbuf_addch(&pathbuf, '/');
290 strbuf_addstr(&pathbuf, entry);
292 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
293 error("unable to normalize alternate object path: %s",
294 pathbuf.buf);
295 strbuf_release(&pathbuf);
296 return -1;
300 * The trailing slash after the directory name is given by
301 * this function at the end. Remove duplicates.
303 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
304 strbuf_setlen(&pathbuf, pathbuf.len - 1);
306 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
307 strbuf_release(&pathbuf);
308 return -1;
311 ent = alloc_alt_odb(pathbuf.buf);
313 /* add the alternate entry */
314 *alt_odb_tail = ent;
315 alt_odb_tail = &(ent->next);
316 ent->next = NULL;
318 /* recursively add alternates */
319 read_info_alternates(pathbuf.buf, depth + 1);
321 strbuf_release(&pathbuf);
322 return 0;
325 static const char *parse_alt_odb_entry(const char *string,
326 int sep,
327 struct strbuf *out)
329 const char *end;
331 strbuf_reset(out);
333 if (*string == '#') {
334 /* comment; consume up to next separator */
335 end = strchrnul(string, sep);
336 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
338 * quoted path; unquote_c_style has copied the
339 * data for us and set "end". Broken quoting (e.g.,
340 * an entry that doesn't end with a quote) falls
341 * back to the unquoted case below.
343 } else {
344 /* normal, unquoted path */
345 end = strchrnul(string, sep);
346 strbuf_add(out, string, end - string);
349 if (*end)
350 end++;
351 return end;
354 static void link_alt_odb_entries(const char *alt, int len, int sep,
355 const char *relative_base, int depth)
357 struct strbuf objdirbuf = STRBUF_INIT;
358 struct strbuf entry = STRBUF_INIT;
360 if (depth > 5) {
361 error("%s: ignoring alternate object stores, nesting too deep.",
362 relative_base);
363 return;
366 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
367 if (strbuf_normalize_path(&objdirbuf) < 0)
368 die("unable to normalize object directory: %s",
369 objdirbuf.buf);
371 while (*alt) {
372 alt = parse_alt_odb_entry(alt, sep, &entry);
373 if (!entry.len)
374 continue;
375 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
377 strbuf_release(&entry);
378 strbuf_release(&objdirbuf);
381 void read_info_alternates(const char * relative_base, int depth)
383 char *map;
384 size_t mapsz;
385 struct stat st;
386 char *path;
387 int fd;
389 path = xstrfmt("%s/info/alternates", relative_base);
390 fd = git_open(path);
391 free(path);
392 if (fd < 0)
393 return;
394 if (fstat(fd, &st) || (st.st_size == 0)) {
395 close(fd);
396 return;
398 mapsz = xsize_t(st.st_size);
399 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
400 close(fd);
402 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
404 munmap(map, mapsz);
407 struct alternate_object_database *alloc_alt_odb(const char *dir)
409 struct alternate_object_database *ent;
411 FLEX_ALLOC_STR(ent, path, dir);
412 strbuf_init(&ent->scratch, 0);
413 strbuf_addf(&ent->scratch, "%s/", dir);
414 ent->base_len = ent->scratch.len;
416 return ent;
419 void add_to_alternates_file(const char *reference)
421 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
422 char *alts = git_pathdup("objects/info/alternates");
423 FILE *in, *out;
425 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
426 out = fdopen_lock_file(lock, "w");
427 if (!out)
428 die_errno("unable to fdopen alternates lockfile");
430 in = fopen(alts, "r");
431 if (in) {
432 struct strbuf line = STRBUF_INIT;
433 int found = 0;
435 while (strbuf_getline(&line, in) != EOF) {
436 if (!strcmp(reference, line.buf)) {
437 found = 1;
438 break;
440 fprintf_or_die(out, "%s\n", line.buf);
443 strbuf_release(&line);
444 fclose(in);
446 if (found) {
447 rollback_lock_file(lock);
448 lock = NULL;
451 else if (errno != ENOENT)
452 die_errno("unable to read alternates file");
454 if (lock) {
455 fprintf_or_die(out, "%s\n", reference);
456 if (commit_lock_file(lock))
457 die_errno("unable to move new alternates file into place");
458 if (alt_odb_tail)
459 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
461 free(alts);
464 void add_to_alternates_memory(const char *reference)
467 * Make sure alternates are initialized, or else our entry may be
468 * overwritten when they are.
470 prepare_alt_odb();
472 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
476 * Compute the exact path an alternate is at and returns it. In case of
477 * error NULL is returned and the human readable error is added to `err`
478 * `path` may be relative and should point to $GITDIR.
479 * `err` must not be null.
481 char *compute_alternate_path(const char *path, struct strbuf *err)
483 char *ref_git = NULL;
484 const char *repo, *ref_git_s;
485 int seen_error = 0;
487 ref_git_s = real_path_if_valid(path);
488 if (!ref_git_s) {
489 seen_error = 1;
490 strbuf_addf(err, _("path '%s' does not exist"), path);
491 goto out;
492 } else
494 * Beware: read_gitfile(), real_path() and mkpath()
495 * return static buffer
497 ref_git = xstrdup(ref_git_s);
499 repo = read_gitfile(ref_git);
500 if (!repo)
501 repo = read_gitfile(mkpath("%s/.git", ref_git));
502 if (repo) {
503 free(ref_git);
504 ref_git = xstrdup(repo);
507 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
508 char *ref_git_git = mkpathdup("%s/.git", ref_git);
509 free(ref_git);
510 ref_git = ref_git_git;
511 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
512 struct strbuf sb = STRBUF_INIT;
513 seen_error = 1;
514 if (get_common_dir(&sb, ref_git)) {
515 strbuf_addf(err,
516 _("reference repository '%s' as a linked "
517 "checkout is not supported yet."),
518 path);
519 goto out;
522 strbuf_addf(err, _("reference repository '%s' is not a "
523 "local repository."), path);
524 goto out;
527 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
528 strbuf_addf(err, _("reference repository '%s' is shallow"),
529 path);
530 seen_error = 1;
531 goto out;
534 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
535 strbuf_addf(err,
536 _("reference repository '%s' is grafted"),
537 path);
538 seen_error = 1;
539 goto out;
542 out:
543 if (seen_error) {
544 free(ref_git);
545 ref_git = NULL;
548 return ref_git;
551 int foreach_alt_odb(alt_odb_fn fn, void *cb)
553 struct alternate_object_database *ent;
554 int r = 0;
556 prepare_alt_odb();
557 for (ent = alt_odb_list; ent; ent = ent->next) {
558 r = fn(ent, cb);
559 if (r)
560 break;
562 return r;
565 void prepare_alt_odb(void)
567 const char *alt;
569 if (alt_odb_tail)
570 return;
572 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
573 if (!alt) alt = "";
575 alt_odb_tail = &alt_odb_list;
576 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
578 read_info_alternates(get_object_directory(), 0);
581 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
582 static int freshen_file(const char *fn)
584 struct utimbuf t;
585 t.actime = t.modtime = time(NULL);
586 return !utime(fn, &t);
590 * All of the check_and_freshen functions return 1 if the file exists and was
591 * freshened (if freshening was requested), 0 otherwise. If they return
592 * 0, you should not assume that it is safe to skip a write of the object (it
593 * either does not exist on disk, or has a stale mtime and may be subject to
594 * pruning).
596 static int check_and_freshen_file(const char *fn, int freshen)
598 if (access(fn, F_OK))
599 return 0;
600 if (freshen && !freshen_file(fn))
601 return 0;
602 return 1;
605 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
607 return check_and_freshen_file(sha1_file_name(sha1), freshen);
610 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
612 struct alternate_object_database *alt;
613 prepare_alt_odb();
614 for (alt = alt_odb_list; alt; alt = alt->next) {
615 const char *path = alt_sha1_path(alt, sha1);
616 if (check_and_freshen_file(path, freshen))
617 return 1;
619 return 0;
622 static int check_and_freshen(const unsigned char *sha1, int freshen)
624 return check_and_freshen_local(sha1, freshen) ||
625 check_and_freshen_nonlocal(sha1, freshen);
628 int has_loose_object_nonlocal(const unsigned char *sha1)
630 return check_and_freshen_nonlocal(sha1, 0);
633 static int has_loose_object(const unsigned char *sha1)
635 return check_and_freshen(sha1, 0);
638 static unsigned int pack_used_ctr;
639 static unsigned int pack_mmap_calls;
640 static unsigned int peak_pack_open_windows;
641 static unsigned int pack_open_windows;
642 static unsigned int pack_open_fds;
643 static unsigned int pack_max_fds;
644 static size_t peak_pack_mapped;
645 static size_t pack_mapped;
646 struct packed_git *packed_git;
648 static struct mru packed_git_mru_storage;
649 struct mru *packed_git_mru = &packed_git_mru_storage;
651 void pack_report(void)
653 fprintf(stderr,
654 "pack_report: getpagesize() = %10" SZ_FMT "\n"
655 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
656 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
657 sz_fmt(getpagesize()),
658 sz_fmt(packed_git_window_size),
659 sz_fmt(packed_git_limit));
660 fprintf(stderr,
661 "pack_report: pack_used_ctr = %10u\n"
662 "pack_report: pack_mmap_calls = %10u\n"
663 "pack_report: pack_open_windows = %10u / %10u\n"
664 "pack_report: pack_mapped = "
665 "%10" SZ_FMT " / %10" SZ_FMT "\n",
666 pack_used_ctr,
667 pack_mmap_calls,
668 pack_open_windows, peak_pack_open_windows,
669 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
673 * Open and mmap the index file at path, perform a couple of
674 * consistency checks, then record its information to p. Return 0 on
675 * success.
677 static int check_packed_git_idx(const char *path, struct packed_git *p)
679 void *idx_map;
680 struct pack_idx_header *hdr;
681 size_t idx_size;
682 uint32_t version, nr, i, *index;
683 int fd = git_open(path);
684 struct stat st;
686 if (fd < 0)
687 return -1;
688 if (fstat(fd, &st)) {
689 close(fd);
690 return -1;
692 idx_size = xsize_t(st.st_size);
693 if (idx_size < 4 * 256 + 20 + 20) {
694 close(fd);
695 return error("index file %s is too small", path);
697 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
698 close(fd);
700 hdr = idx_map;
701 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
702 version = ntohl(hdr->idx_version);
703 if (version < 2 || version > 2) {
704 munmap(idx_map, idx_size);
705 return error("index file %s is version %"PRIu32
706 " and is not supported by this binary"
707 " (try upgrading GIT to a newer version)",
708 path, version);
710 } else
711 version = 1;
713 nr = 0;
714 index = idx_map;
715 if (version > 1)
716 index += 2; /* skip index header */
717 for (i = 0; i < 256; i++) {
718 uint32_t n = ntohl(index[i]);
719 if (n < nr) {
720 munmap(idx_map, idx_size);
721 return error("non-monotonic index %s", path);
723 nr = n;
726 if (version == 1) {
728 * Total size:
729 * - 256 index entries 4 bytes each
730 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
731 * - 20-byte SHA1 of the packfile
732 * - 20-byte SHA1 file checksum
734 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
735 munmap(idx_map, idx_size);
736 return error("wrong index v1 file size in %s", path);
738 } else if (version == 2) {
740 * Minimum size:
741 * - 8 bytes of header
742 * - 256 index entries 4 bytes each
743 * - 20-byte sha1 entry * nr
744 * - 4-byte crc entry * nr
745 * - 4-byte offset entry * nr
746 * - 20-byte SHA1 of the packfile
747 * - 20-byte SHA1 file checksum
748 * And after the 4-byte offset table might be a
749 * variable sized table containing 8-byte entries
750 * for offsets larger than 2^31.
752 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
753 unsigned long max_size = min_size;
754 if (nr)
755 max_size += (nr - 1)*8;
756 if (idx_size < min_size || idx_size > max_size) {
757 munmap(idx_map, idx_size);
758 return error("wrong index v2 file size in %s", path);
760 if (idx_size != min_size &&
762 * make sure we can deal with large pack offsets.
763 * 31-bit signed offset won't be enough, neither
764 * 32-bit unsigned one will be.
766 (sizeof(off_t) <= 4)) {
767 munmap(idx_map, idx_size);
768 return error("pack too large for current definition of off_t in %s", path);
772 p->index_version = version;
773 p->index_data = idx_map;
774 p->index_size = idx_size;
775 p->num_objects = nr;
776 return 0;
779 int open_pack_index(struct packed_git *p)
781 char *idx_name;
782 size_t len;
783 int ret;
785 if (p->index_data)
786 return 0;
788 if (!strip_suffix(p->pack_name, ".pack", &len))
789 die("BUG: pack_name does not end in .pack");
790 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
791 ret = check_packed_git_idx(idx_name, p);
792 free(idx_name);
793 return ret;
796 static void scan_windows(struct packed_git *p,
797 struct packed_git **lru_p,
798 struct pack_window **lru_w,
799 struct pack_window **lru_l)
801 struct pack_window *w, *w_l;
803 for (w_l = NULL, w = p->windows; w; w = w->next) {
804 if (!w->inuse_cnt) {
805 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
806 *lru_p = p;
807 *lru_w = w;
808 *lru_l = w_l;
811 w_l = w;
815 static int unuse_one_window(struct packed_git *current)
817 struct packed_git *p, *lru_p = NULL;
818 struct pack_window *lru_w = NULL, *lru_l = NULL;
820 if (current)
821 scan_windows(current, &lru_p, &lru_w, &lru_l);
822 for (p = packed_git; p; p = p->next)
823 scan_windows(p, &lru_p, &lru_w, &lru_l);
824 if (lru_p) {
825 munmap(lru_w->base, lru_w->len);
826 pack_mapped -= lru_w->len;
827 if (lru_l)
828 lru_l->next = lru_w->next;
829 else
830 lru_p->windows = lru_w->next;
831 free(lru_w);
832 pack_open_windows--;
833 return 1;
835 return 0;
838 void release_pack_memory(size_t need)
840 size_t cur = pack_mapped;
841 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
842 ; /* nothing */
845 static void mmap_limit_check(size_t length)
847 static size_t limit = 0;
848 if (!limit) {
849 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
850 if (!limit)
851 limit = SIZE_MAX;
853 if (length > limit)
854 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
855 (uintmax_t)length, (uintmax_t)limit);
858 void *xmmap_gently(void *start, size_t length,
859 int prot, int flags, int fd, off_t offset)
861 void *ret;
863 mmap_limit_check(length);
864 ret = mmap(start, length, prot, flags, fd, offset);
865 if (ret == MAP_FAILED) {
866 if (!length)
867 return NULL;
868 release_pack_memory(length);
869 ret = mmap(start, length, prot, flags, fd, offset);
871 return ret;
874 void *xmmap(void *start, size_t length,
875 int prot, int flags, int fd, off_t offset)
877 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
878 if (ret == MAP_FAILED)
879 die_errno("mmap failed");
880 return ret;
883 void close_pack_windows(struct packed_git *p)
885 while (p->windows) {
886 struct pack_window *w = p->windows;
888 if (w->inuse_cnt)
889 die("pack '%s' still has open windows to it",
890 p->pack_name);
891 munmap(w->base, w->len);
892 pack_mapped -= w->len;
893 pack_open_windows--;
894 p->windows = w->next;
895 free(w);
899 static int close_pack_fd(struct packed_git *p)
901 if (p->pack_fd < 0)
902 return 0;
904 close(p->pack_fd);
905 pack_open_fds--;
906 p->pack_fd = -1;
908 return 1;
911 static void close_pack(struct packed_git *p)
913 close_pack_windows(p);
914 close_pack_fd(p);
915 close_pack_index(p);
918 void close_all_packs(void)
920 struct packed_git *p;
922 for (p = packed_git; p; p = p->next)
923 if (p->do_not_close)
924 die("BUG: want to close pack marked 'do-not-close'");
925 else
926 close_pack(p);
931 * The LRU pack is the one with the oldest MRU window, preferring packs
932 * with no used windows, or the oldest mtime if it has no windows allocated.
934 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
936 struct pack_window *w, *this_mru_w;
937 int has_windows_inuse = 0;
940 * Reject this pack if it has windows and the previously selected
941 * one does not. If this pack does not have windows, reject
942 * it if the pack file is newer than the previously selected one.
944 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
945 return;
947 for (w = this_mru_w = p->windows; w; w = w->next) {
949 * Reject this pack if any of its windows are in use,
950 * but the previously selected pack did not have any
951 * inuse windows. Otherwise, record that this pack
952 * has windows in use.
954 if (w->inuse_cnt) {
955 if (*accept_windows_inuse)
956 has_windows_inuse = 1;
957 else
958 return;
961 if (w->last_used > this_mru_w->last_used)
962 this_mru_w = w;
965 * Reject this pack if it has windows that have been
966 * used more recently than the previously selected pack.
967 * If the previously selected pack had windows inuse and
968 * we have not encountered a window in this pack that is
969 * inuse, skip this check since we prefer a pack with no
970 * inuse windows to one that has inuse windows.
972 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
973 this_mru_w->last_used > (*mru_w)->last_used)
974 return;
978 * Select this pack.
980 *mru_w = this_mru_w;
981 *lru_p = p;
982 *accept_windows_inuse = has_windows_inuse;
985 static int close_one_pack(void)
987 struct packed_git *p, *lru_p = NULL;
988 struct pack_window *mru_w = NULL;
989 int accept_windows_inuse = 1;
991 for (p = packed_git; p; p = p->next) {
992 if (p->pack_fd == -1)
993 continue;
994 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
997 if (lru_p)
998 return close_pack_fd(lru_p);
1000 return 0;
1003 void unuse_pack(struct pack_window **w_cursor)
1005 struct pack_window *w = *w_cursor;
1006 if (w) {
1007 w->inuse_cnt--;
1008 *w_cursor = NULL;
1012 void close_pack_index(struct packed_git *p)
1014 if (p->index_data) {
1015 munmap((void *)p->index_data, p->index_size);
1016 p->index_data = NULL;
1020 static unsigned int get_max_fd_limit(void)
1022 #ifdef RLIMIT_NOFILE
1024 struct rlimit lim;
1026 if (!getrlimit(RLIMIT_NOFILE, &lim))
1027 return lim.rlim_cur;
1029 #endif
1031 #ifdef _SC_OPEN_MAX
1033 long open_max = sysconf(_SC_OPEN_MAX);
1034 if (0 < open_max)
1035 return open_max;
1037 * Otherwise, we got -1 for one of the two
1038 * reasons:
1040 * (1) sysconf() did not understand _SC_OPEN_MAX
1041 * and signaled an error with -1; or
1042 * (2) sysconf() said there is no limit.
1044 * We _could_ clear errno before calling sysconf() to
1045 * tell these two cases apart and return a huge number
1046 * in the latter case to let the caller cap it to a
1047 * value that is not so selfish, but letting the
1048 * fallback OPEN_MAX codepath take care of these cases
1049 * is a lot simpler.
1052 #endif
1054 #ifdef OPEN_MAX
1055 return OPEN_MAX;
1056 #else
1057 return 1; /* see the caller ;-) */
1058 #endif
1062 * Do not call this directly as this leaks p->pack_fd on error return;
1063 * call open_packed_git() instead.
1065 static int open_packed_git_1(struct packed_git *p)
1067 struct stat st;
1068 struct pack_header hdr;
1069 unsigned char sha1[20];
1070 unsigned char *idx_sha1;
1071 long fd_flag;
1073 if (!p->index_data && open_pack_index(p))
1074 return error("packfile %s index unavailable", p->pack_name);
1076 if (!pack_max_fds) {
1077 unsigned int max_fds = get_max_fd_limit();
1079 /* Save 3 for stdin/stdout/stderr, 22 for work */
1080 if (25 < max_fds)
1081 pack_max_fds = max_fds - 25;
1082 else
1083 pack_max_fds = 1;
1086 while (pack_max_fds <= pack_open_fds && close_one_pack())
1087 ; /* nothing */
1089 p->pack_fd = git_open(p->pack_name);
1090 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1091 return -1;
1092 pack_open_fds++;
1094 /* If we created the struct before we had the pack we lack size. */
1095 if (!p->pack_size) {
1096 if (!S_ISREG(st.st_mode))
1097 return error("packfile %s not a regular file", p->pack_name);
1098 p->pack_size = st.st_size;
1099 } else if (p->pack_size != st.st_size)
1100 return error("packfile %s size changed", p->pack_name);
1102 /* We leave these file descriptors open with sliding mmap;
1103 * there is no point keeping them open across exec(), though.
1105 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1106 if (fd_flag < 0)
1107 return error("cannot determine file descriptor flags");
1108 fd_flag |= FD_CLOEXEC;
1109 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1110 return error("cannot set FD_CLOEXEC");
1112 /* Verify we recognize this pack file format. */
1113 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1114 return error("file %s is far too short to be a packfile", p->pack_name);
1115 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1116 return error("file %s is not a GIT packfile", p->pack_name);
1117 if (!pack_version_ok(hdr.hdr_version))
1118 return error("packfile %s is version %"PRIu32" and not"
1119 " supported (try upgrading GIT to a newer version)",
1120 p->pack_name, ntohl(hdr.hdr_version));
1122 /* Verify the pack matches its index. */
1123 if (p->num_objects != ntohl(hdr.hdr_entries))
1124 return error("packfile %s claims to have %"PRIu32" objects"
1125 " while index indicates %"PRIu32" objects",
1126 p->pack_name, ntohl(hdr.hdr_entries),
1127 p->num_objects);
1128 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1129 return error("end of packfile %s is unavailable", p->pack_name);
1130 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1131 return error("packfile %s signature is unavailable", p->pack_name);
1132 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1133 if (hashcmp(sha1, idx_sha1))
1134 return error("packfile %s does not match index", p->pack_name);
1135 return 0;
1138 static int open_packed_git(struct packed_git *p)
1140 if (!open_packed_git_1(p))
1141 return 0;
1142 close_pack_fd(p);
1143 return -1;
1146 static int in_window(struct pack_window *win, off_t offset)
1148 /* We must promise at least 20 bytes (one hash) after the
1149 * offset is available from this window, otherwise the offset
1150 * is not actually in this window and a different window (which
1151 * has that one hash excess) must be used. This is to support
1152 * the object header and delta base parsing routines below.
1154 off_t win_off = win->offset;
1155 return win_off <= offset
1156 && (offset + 20) <= (win_off + win->len);
1159 unsigned char *use_pack(struct packed_git *p,
1160 struct pack_window **w_cursor,
1161 off_t offset,
1162 unsigned long *left)
1164 struct pack_window *win = *w_cursor;
1166 /* Since packfiles end in a hash of their content and it's
1167 * pointless to ask for an offset into the middle of that
1168 * hash, and the in_window function above wouldn't match
1169 * don't allow an offset too close to the end of the file.
1171 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1172 die("packfile %s cannot be accessed", p->pack_name);
1173 if (offset > (p->pack_size - 20))
1174 die("offset beyond end of packfile (truncated pack?)");
1175 if (offset < 0)
1176 die(_("offset before end of packfile (broken .idx?)"));
1178 if (!win || !in_window(win, offset)) {
1179 if (win)
1180 win->inuse_cnt--;
1181 for (win = p->windows; win; win = win->next) {
1182 if (in_window(win, offset))
1183 break;
1185 if (!win) {
1186 size_t window_align = packed_git_window_size / 2;
1187 off_t len;
1189 if (p->pack_fd == -1 && open_packed_git(p))
1190 die("packfile %s cannot be accessed", p->pack_name);
1192 win = xcalloc(1, sizeof(*win));
1193 win->offset = (offset / window_align) * window_align;
1194 len = p->pack_size - win->offset;
1195 if (len > packed_git_window_size)
1196 len = packed_git_window_size;
1197 win->len = (size_t)len;
1198 pack_mapped += win->len;
1199 while (packed_git_limit < pack_mapped
1200 && unuse_one_window(p))
1201 ; /* nothing */
1202 win->base = xmmap(NULL, win->len,
1203 PROT_READ, MAP_PRIVATE,
1204 p->pack_fd, win->offset);
1205 if (win->base == MAP_FAILED)
1206 die_errno("packfile %s cannot be mapped",
1207 p->pack_name);
1208 if (!win->offset && win->len == p->pack_size
1209 && !p->do_not_close)
1210 close_pack_fd(p);
1211 pack_mmap_calls++;
1212 pack_open_windows++;
1213 if (pack_mapped > peak_pack_mapped)
1214 peak_pack_mapped = pack_mapped;
1215 if (pack_open_windows > peak_pack_open_windows)
1216 peak_pack_open_windows = pack_open_windows;
1217 win->next = p->windows;
1218 p->windows = win;
1221 if (win != *w_cursor) {
1222 win->last_used = pack_used_ctr++;
1223 win->inuse_cnt++;
1224 *w_cursor = win;
1226 offset -= win->offset;
1227 if (left)
1228 *left = win->len - xsize_t(offset);
1229 return win->base + offset;
1232 static struct packed_git *alloc_packed_git(int extra)
1234 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1235 memset(p, 0, sizeof(*p));
1236 p->pack_fd = -1;
1237 return p;
1240 static void try_to_free_pack_memory(size_t size)
1242 release_pack_memory(size);
1245 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1247 static int have_set_try_to_free_routine;
1248 struct stat st;
1249 size_t alloc;
1250 struct packed_git *p;
1252 if (!have_set_try_to_free_routine) {
1253 have_set_try_to_free_routine = 1;
1254 set_try_to_free_routine(try_to_free_pack_memory);
1258 * Make sure a corresponding .pack file exists and that
1259 * the index looks sane.
1261 if (!strip_suffix_mem(path, &path_len, ".idx"))
1262 return NULL;
1265 * ".pack" is long enough to hold any suffix we're adding (and
1266 * the use xsnprintf double-checks that)
1268 alloc = st_add3(path_len, strlen(".pack"), 1);
1269 p = alloc_packed_git(alloc);
1270 memcpy(p->pack_name, path, path_len);
1272 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1273 if (!access(p->pack_name, F_OK))
1274 p->pack_keep = 1;
1276 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1277 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1278 free(p);
1279 return NULL;
1282 /* ok, it looks sane as far as we can check without
1283 * actually mapping the pack file.
1285 p->pack_size = st.st_size;
1286 p->pack_local = local;
1287 p->mtime = st.st_mtime;
1288 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1289 hashclr(p->sha1);
1290 return p;
1293 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1295 const char *path = sha1_pack_name(sha1);
1296 size_t alloc = st_add(strlen(path), 1);
1297 struct packed_git *p = alloc_packed_git(alloc);
1299 memcpy(p->pack_name, path, alloc); /* includes NUL */
1300 hashcpy(p->sha1, sha1);
1301 if (check_packed_git_idx(idx_path, p)) {
1302 free(p);
1303 return NULL;
1306 return p;
1309 void install_packed_git(struct packed_git *pack)
1311 if (pack->pack_fd != -1)
1312 pack_open_fds++;
1314 pack->next = packed_git;
1315 packed_git = pack;
1318 void (*report_garbage)(unsigned seen_bits, const char *path);
1320 static void report_helper(const struct string_list *list,
1321 int seen_bits, int first, int last)
1323 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1324 return;
1326 for (; first < last; first++)
1327 report_garbage(seen_bits, list->items[first].string);
1330 static void report_pack_garbage(struct string_list *list)
1332 int i, baselen = -1, first = 0, seen_bits = 0;
1334 if (!report_garbage)
1335 return;
1337 string_list_sort(list);
1339 for (i = 0; i < list->nr; i++) {
1340 const char *path = list->items[i].string;
1341 if (baselen != -1 &&
1342 strncmp(path, list->items[first].string, baselen)) {
1343 report_helper(list, seen_bits, first, i);
1344 baselen = -1;
1345 seen_bits = 0;
1347 if (baselen == -1) {
1348 const char *dot = strrchr(path, '.');
1349 if (!dot) {
1350 report_garbage(PACKDIR_FILE_GARBAGE, path);
1351 continue;
1353 baselen = dot - path + 1;
1354 first = i;
1356 if (!strcmp(path + baselen, "pack"))
1357 seen_bits |= 1;
1358 else if (!strcmp(path + baselen, "idx"))
1359 seen_bits |= 2;
1361 report_helper(list, seen_bits, first, list->nr);
1364 static void prepare_packed_git_one(char *objdir, int local)
1366 struct strbuf path = STRBUF_INIT;
1367 size_t dirnamelen;
1368 DIR *dir;
1369 struct dirent *de;
1370 struct string_list garbage = STRING_LIST_INIT_DUP;
1372 strbuf_addstr(&path, objdir);
1373 strbuf_addstr(&path, "/pack");
1374 dir = opendir(path.buf);
1375 if (!dir) {
1376 if (errno != ENOENT)
1377 error_errno("unable to open object pack directory: %s",
1378 path.buf);
1379 strbuf_release(&path);
1380 return;
1382 strbuf_addch(&path, '/');
1383 dirnamelen = path.len;
1384 while ((de = readdir(dir)) != NULL) {
1385 struct packed_git *p;
1386 size_t base_len;
1388 if (is_dot_or_dotdot(de->d_name))
1389 continue;
1391 strbuf_setlen(&path, dirnamelen);
1392 strbuf_addstr(&path, de->d_name);
1394 base_len = path.len;
1395 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1396 /* Don't reopen a pack we already have. */
1397 for (p = packed_git; p; p = p->next) {
1398 size_t len;
1399 if (strip_suffix(p->pack_name, ".pack", &len) &&
1400 len == base_len &&
1401 !memcmp(p->pack_name, path.buf, len))
1402 break;
1404 if (p == NULL &&
1406 * See if it really is a valid .idx file with
1407 * corresponding .pack file that we can map.
1409 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1410 install_packed_git(p);
1413 if (!report_garbage)
1414 continue;
1416 if (ends_with(de->d_name, ".idx") ||
1417 ends_with(de->d_name, ".pack") ||
1418 ends_with(de->d_name, ".bitmap") ||
1419 ends_with(de->d_name, ".keep"))
1420 string_list_append(&garbage, path.buf);
1421 else
1422 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1424 closedir(dir);
1425 report_pack_garbage(&garbage);
1426 string_list_clear(&garbage, 0);
1427 strbuf_release(&path);
1430 static int approximate_object_count_valid;
1433 * Give a fast, rough count of the number of objects in the repository. This
1434 * ignores loose objects completely. If you have a lot of them, then either
1435 * you should repack because your performance will be awful, or they are
1436 * all unreachable objects about to be pruned, in which case they're not really
1437 * interesting as a measure of repo size in the first place.
1439 unsigned long approximate_object_count(void)
1441 static unsigned long count;
1442 if (!approximate_object_count_valid) {
1443 struct packed_git *p;
1445 prepare_packed_git();
1446 count = 0;
1447 for (p = packed_git; p; p = p->next) {
1448 if (open_pack_index(p))
1449 continue;
1450 count += p->num_objects;
1453 return count;
1456 static void *get_next_packed_git(const void *p)
1458 return ((const struct packed_git *)p)->next;
1461 static void set_next_packed_git(void *p, void *next)
1463 ((struct packed_git *)p)->next = next;
1466 static int sort_pack(const void *a_, const void *b_)
1468 const struct packed_git *a = a_;
1469 const struct packed_git *b = b_;
1470 int st;
1473 * Local packs tend to contain objects specific to our
1474 * variant of the project than remote ones. In addition,
1475 * remote ones could be on a network mounted filesystem.
1476 * Favor local ones for these reasons.
1478 st = a->pack_local - b->pack_local;
1479 if (st)
1480 return -st;
1483 * Younger packs tend to contain more recent objects,
1484 * and more recent objects tend to get accessed more
1485 * often.
1487 if (a->mtime < b->mtime)
1488 return 1;
1489 else if (a->mtime == b->mtime)
1490 return 0;
1491 return -1;
1494 static void rearrange_packed_git(void)
1496 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1497 set_next_packed_git, sort_pack);
1500 static void prepare_packed_git_mru(void)
1502 struct packed_git *p;
1504 mru_clear(packed_git_mru);
1505 for (p = packed_git; p; p = p->next)
1506 mru_append(packed_git_mru, p);
1509 static int prepare_packed_git_run_once = 0;
1510 void prepare_packed_git(void)
1512 struct alternate_object_database *alt;
1514 if (prepare_packed_git_run_once)
1515 return;
1516 prepare_packed_git_one(get_object_directory(), 1);
1517 prepare_alt_odb();
1518 for (alt = alt_odb_list; alt; alt = alt->next)
1519 prepare_packed_git_one(alt->path, 0);
1520 rearrange_packed_git();
1521 prepare_packed_git_mru();
1522 prepare_packed_git_run_once = 1;
1525 void reprepare_packed_git(void)
1527 approximate_object_count_valid = 0;
1528 prepare_packed_git_run_once = 0;
1529 prepare_packed_git();
1532 static void mark_bad_packed_object(struct packed_git *p,
1533 const unsigned char *sha1)
1535 unsigned i;
1536 for (i = 0; i < p->num_bad_objects; i++)
1537 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1538 return;
1539 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1540 st_mult(GIT_SHA1_RAWSZ,
1541 st_add(p->num_bad_objects, 1)));
1542 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1543 p->num_bad_objects++;
1546 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1548 struct packed_git *p;
1549 unsigned i;
1551 for (p = packed_git; p; p = p->next)
1552 for (i = 0; i < p->num_bad_objects; i++)
1553 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1554 return p;
1555 return NULL;
1559 * With an in-core object data in "map", rehash it to make sure the
1560 * object name actually matches "sha1" to detect object corruption.
1561 * With "map" == NULL, try reading the object named with "sha1" using
1562 * the streaming interface and rehash it to do the same.
1564 int check_sha1_signature(const unsigned char *sha1, void *map,
1565 unsigned long size, const char *type)
1567 unsigned char real_sha1[20];
1568 enum object_type obj_type;
1569 struct git_istream *st;
1570 git_SHA_CTX c;
1571 char hdr[32];
1572 int hdrlen;
1574 if (map) {
1575 hash_sha1_file(map, size, type, real_sha1);
1576 return hashcmp(sha1, real_sha1) ? -1 : 0;
1579 st = open_istream(sha1, &obj_type, &size, NULL);
1580 if (!st)
1581 return -1;
1583 /* Generate the header */
1584 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1586 /* Sha1.. */
1587 git_SHA1_Init(&c);
1588 git_SHA1_Update(&c, hdr, hdrlen);
1589 for (;;) {
1590 char buf[1024 * 16];
1591 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1593 if (readlen < 0) {
1594 close_istream(st);
1595 return -1;
1597 if (!readlen)
1598 break;
1599 git_SHA1_Update(&c, buf, readlen);
1601 git_SHA1_Final(real_sha1, &c);
1602 close_istream(st);
1603 return hashcmp(sha1, real_sha1) ? -1 : 0;
1606 int git_open_cloexec(const char *name, int flags)
1608 int fd;
1609 static int o_cloexec = O_CLOEXEC;
1611 fd = open(name, flags | o_cloexec);
1612 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1613 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1614 o_cloexec &= ~O_CLOEXEC;
1615 fd = open(name, flags | o_cloexec);
1618 #if defined(F_GETFL) && defined(F_SETFL) && defined(FD_CLOEXEC)
1620 static int fd_cloexec = FD_CLOEXEC;
1622 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1623 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1624 int flags = fcntl(fd, F_GETFL);
1625 if (fcntl(fd, F_SETFL, flags | fd_cloexec))
1626 fd_cloexec = 0;
1629 #endif
1630 return fd;
1634 * Find "sha1" as a loose object in the local repository or in an alternate.
1635 * Returns 0 on success, negative on failure.
1637 * The "path" out-parameter will give the path of the object we found (if any).
1638 * Note that it may point to static storage and is only valid until another
1639 * call to sha1_file_name(), etc.
1641 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1642 const char **path)
1644 struct alternate_object_database *alt;
1646 *path = sha1_file_name(sha1);
1647 if (!lstat(*path, st))
1648 return 0;
1650 prepare_alt_odb();
1651 errno = ENOENT;
1652 for (alt = alt_odb_list; alt; alt = alt->next) {
1653 *path = alt_sha1_path(alt, sha1);
1654 if (!lstat(*path, st))
1655 return 0;
1658 return -1;
1662 * Like stat_sha1_file(), but actually open the object and return the
1663 * descriptor. See the caveats on the "path" parameter above.
1665 static int open_sha1_file(const unsigned char *sha1, const char **path)
1667 int fd;
1668 struct alternate_object_database *alt;
1669 int most_interesting_errno;
1671 *path = sha1_file_name(sha1);
1672 fd = git_open(*path);
1673 if (fd >= 0)
1674 return fd;
1675 most_interesting_errno = errno;
1677 prepare_alt_odb();
1678 for (alt = alt_odb_list; alt; alt = alt->next) {
1679 *path = alt_sha1_path(alt, sha1);
1680 fd = git_open(*path);
1681 if (fd >= 0)
1682 return fd;
1683 if (most_interesting_errno == ENOENT)
1684 most_interesting_errno = errno;
1686 errno = most_interesting_errno;
1687 return -1;
1691 * Map the loose object at "path" if it is not NULL, or the path found by
1692 * searching for a loose object named "sha1".
1694 static void *map_sha1_file_1(const char *path,
1695 const unsigned char *sha1,
1696 unsigned long *size)
1698 void *map;
1699 int fd;
1701 if (path)
1702 fd = git_open(path);
1703 else
1704 fd = open_sha1_file(sha1, &path);
1705 map = NULL;
1706 if (fd >= 0) {
1707 struct stat st;
1709 if (!fstat(fd, &st)) {
1710 *size = xsize_t(st.st_size);
1711 if (!*size) {
1712 /* mmap() is forbidden on empty files */
1713 error("object file %s is empty", path);
1714 return NULL;
1716 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1718 close(fd);
1720 return map;
1723 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1725 return map_sha1_file_1(NULL, sha1, size);
1728 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1729 unsigned long len, enum object_type *type, unsigned long *sizep)
1731 unsigned shift;
1732 unsigned long size, c;
1733 unsigned long used = 0;
1735 c = buf[used++];
1736 *type = (c >> 4) & 7;
1737 size = c & 15;
1738 shift = 4;
1739 while (c & 0x80) {
1740 if (len <= used || bitsizeof(long) <= shift) {
1741 error("bad object header");
1742 size = used = 0;
1743 break;
1745 c = buf[used++];
1746 size += (c & 0x7f) << shift;
1747 shift += 7;
1749 *sizep = size;
1750 return used;
1753 static int unpack_sha1_short_header(git_zstream *stream,
1754 unsigned char *map, unsigned long mapsize,
1755 void *buffer, unsigned long bufsiz)
1757 /* Get the data stream */
1758 memset(stream, 0, sizeof(*stream));
1759 stream->next_in = map;
1760 stream->avail_in = mapsize;
1761 stream->next_out = buffer;
1762 stream->avail_out = bufsiz;
1764 git_inflate_init(stream);
1765 return git_inflate(stream, 0);
1768 int unpack_sha1_header(git_zstream *stream,
1769 unsigned char *map, unsigned long mapsize,
1770 void *buffer, unsigned long bufsiz)
1772 int status = unpack_sha1_short_header(stream, map, mapsize,
1773 buffer, bufsiz);
1775 if (status < Z_OK)
1776 return status;
1778 /* Make sure we have the terminating NUL */
1779 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1780 return -1;
1781 return 0;
1784 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1785 unsigned long mapsize, void *buffer,
1786 unsigned long bufsiz, struct strbuf *header)
1788 int status;
1790 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1791 if (status < Z_OK)
1792 return -1;
1795 * Check if entire header is unpacked in the first iteration.
1797 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1798 return 0;
1801 * buffer[0..bufsiz] was not large enough. Copy the partial
1802 * result out to header, and then append the result of further
1803 * reading the stream.
1805 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1806 stream->next_out = buffer;
1807 stream->avail_out = bufsiz;
1809 do {
1810 status = git_inflate(stream, 0);
1811 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1812 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1813 return 0;
1814 stream->next_out = buffer;
1815 stream->avail_out = bufsiz;
1816 } while (status != Z_STREAM_END);
1817 return -1;
1820 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1822 int bytes = strlen(buffer) + 1;
1823 unsigned char *buf = xmallocz(size);
1824 unsigned long n;
1825 int status = Z_OK;
1827 n = stream->total_out - bytes;
1828 if (n > size)
1829 n = size;
1830 memcpy(buf, (char *) buffer + bytes, n);
1831 bytes = n;
1832 if (bytes <= size) {
1834 * The above condition must be (bytes <= size), not
1835 * (bytes < size). In other words, even though we
1836 * expect no more output and set avail_out to zero,
1837 * the input zlib stream may have bytes that express
1838 * "this concludes the stream", and we *do* want to
1839 * eat that input.
1841 * Otherwise we would not be able to test that we
1842 * consumed all the input to reach the expected size;
1843 * we also want to check that zlib tells us that all
1844 * went well with status == Z_STREAM_END at the end.
1846 stream->next_out = buf + bytes;
1847 stream->avail_out = size - bytes;
1848 while (status == Z_OK)
1849 status = git_inflate(stream, Z_FINISH);
1851 if (status == Z_STREAM_END && !stream->avail_in) {
1852 git_inflate_end(stream);
1853 return buf;
1856 if (status < 0)
1857 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1858 else if (stream->avail_in)
1859 error("garbage at end of loose object '%s'",
1860 sha1_to_hex(sha1));
1861 free(buf);
1862 return NULL;
1866 * We used to just use "sscanf()", but that's actually way
1867 * too permissive for what we want to check. So do an anal
1868 * object header parse by hand.
1870 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1871 unsigned int flags)
1873 const char *type_buf = hdr;
1874 unsigned long size;
1875 int type, type_len = 0;
1878 * The type can be of any size but is followed by
1879 * a space.
1881 for (;;) {
1882 char c = *hdr++;
1883 if (!c)
1884 return -1;
1885 if (c == ' ')
1886 break;
1887 type_len++;
1890 type = type_from_string_gently(type_buf, type_len, 1);
1891 if (oi->typename)
1892 strbuf_add(oi->typename, type_buf, type_len);
1894 * Set type to 0 if its an unknown object and
1895 * we're obtaining the type using '--allow-unknown-type'
1896 * option.
1898 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1899 type = 0;
1900 else if (type < 0)
1901 die("invalid object type");
1902 if (oi->typep)
1903 *oi->typep = type;
1906 * The length must follow immediately, and be in canonical
1907 * decimal format (ie "010" is not valid).
1909 size = *hdr++ - '0';
1910 if (size > 9)
1911 return -1;
1912 if (size) {
1913 for (;;) {
1914 unsigned long c = *hdr - '0';
1915 if (c > 9)
1916 break;
1917 hdr++;
1918 size = size * 10 + c;
1922 if (oi->sizep)
1923 *oi->sizep = size;
1926 * The length must be followed by a zero byte
1928 return *hdr ? -1 : type;
1931 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1933 struct object_info oi = OBJECT_INFO_INIT;
1935 oi.sizep = sizep;
1936 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1939 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1941 int ret;
1942 git_zstream stream;
1943 char hdr[8192];
1945 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1946 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1947 return NULL;
1949 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1952 unsigned long get_size_from_delta(struct packed_git *p,
1953 struct pack_window **w_curs,
1954 off_t curpos)
1956 const unsigned char *data;
1957 unsigned char delta_head[20], *in;
1958 git_zstream stream;
1959 int st;
1961 memset(&stream, 0, sizeof(stream));
1962 stream.next_out = delta_head;
1963 stream.avail_out = sizeof(delta_head);
1965 git_inflate_init(&stream);
1966 do {
1967 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1968 stream.next_in = in;
1969 st = git_inflate(&stream, Z_FINISH);
1970 curpos += stream.next_in - in;
1971 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1972 stream.total_out < sizeof(delta_head));
1973 git_inflate_end(&stream);
1974 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1975 error("delta data unpack-initial failed");
1976 return 0;
1979 /* Examine the initial part of the delta to figure out
1980 * the result size.
1982 data = delta_head;
1984 /* ignore base size */
1985 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1987 /* Read the result size */
1988 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1991 static off_t get_delta_base(struct packed_git *p,
1992 struct pack_window **w_curs,
1993 off_t *curpos,
1994 enum object_type type,
1995 off_t delta_obj_offset)
1997 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1998 off_t base_offset;
2000 /* use_pack() assured us we have [base_info, base_info + 20)
2001 * as a range that we can look at without walking off the
2002 * end of the mapped window. Its actually the hash size
2003 * that is assured. An OFS_DELTA longer than the hash size
2004 * is stupid, as then a REF_DELTA would be smaller to store.
2006 if (type == OBJ_OFS_DELTA) {
2007 unsigned used = 0;
2008 unsigned char c = base_info[used++];
2009 base_offset = c & 127;
2010 while (c & 128) {
2011 base_offset += 1;
2012 if (!base_offset || MSB(base_offset, 7))
2013 return 0; /* overflow */
2014 c = base_info[used++];
2015 base_offset = (base_offset << 7) + (c & 127);
2017 base_offset = delta_obj_offset - base_offset;
2018 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2019 return 0; /* out of bound */
2020 *curpos += used;
2021 } else if (type == OBJ_REF_DELTA) {
2022 /* The base entry _must_ be in the same pack */
2023 base_offset = find_pack_entry_one(base_info, p);
2024 *curpos += 20;
2025 } else
2026 die("I am totally screwed");
2027 return base_offset;
2031 * Like get_delta_base above, but we return the sha1 instead of the pack
2032 * offset. This means it is cheaper for REF deltas (we do not have to do
2033 * the final object lookup), but more expensive for OFS deltas (we
2034 * have to load the revidx to convert the offset back into a sha1).
2036 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2037 struct pack_window **w_curs,
2038 off_t curpos,
2039 enum object_type type,
2040 off_t delta_obj_offset)
2042 if (type == OBJ_REF_DELTA) {
2043 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2044 return base;
2045 } else if (type == OBJ_OFS_DELTA) {
2046 struct revindex_entry *revidx;
2047 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2048 type, delta_obj_offset);
2050 if (!base_offset)
2051 return NULL;
2053 revidx = find_pack_revindex(p, base_offset);
2054 if (!revidx)
2055 return NULL;
2057 return nth_packed_object_sha1(p, revidx->nr);
2058 } else
2059 return NULL;
2062 int unpack_object_header(struct packed_git *p,
2063 struct pack_window **w_curs,
2064 off_t *curpos,
2065 unsigned long *sizep)
2067 unsigned char *base;
2068 unsigned long left;
2069 unsigned long used;
2070 enum object_type type;
2072 /* use_pack() assures us we have [base, base + 20) available
2073 * as a range that we can look at. (Its actually the hash
2074 * size that is assured.) With our object header encoding
2075 * the maximum deflated object size is 2^137, which is just
2076 * insane, so we know won't exceed what we have been given.
2078 base = use_pack(p, w_curs, *curpos, &left);
2079 used = unpack_object_header_buffer(base, left, &type, sizep);
2080 if (!used) {
2081 type = OBJ_BAD;
2082 } else
2083 *curpos += used;
2085 return type;
2088 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2090 int type;
2091 struct revindex_entry *revidx;
2092 const unsigned char *sha1;
2093 revidx = find_pack_revindex(p, obj_offset);
2094 if (!revidx)
2095 return OBJ_BAD;
2096 sha1 = nth_packed_object_sha1(p, revidx->nr);
2097 mark_bad_packed_object(p, sha1);
2098 type = sha1_object_info(sha1, NULL);
2099 if (type <= OBJ_NONE)
2100 return OBJ_BAD;
2101 return type;
2104 #define POI_STACK_PREALLOC 64
2106 static enum object_type packed_to_object_type(struct packed_git *p,
2107 off_t obj_offset,
2108 enum object_type type,
2109 struct pack_window **w_curs,
2110 off_t curpos)
2112 off_t small_poi_stack[POI_STACK_PREALLOC];
2113 off_t *poi_stack = small_poi_stack;
2114 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2116 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2117 off_t base_offset;
2118 unsigned long size;
2119 /* Push the object we're going to leave behind */
2120 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2121 poi_stack_alloc = alloc_nr(poi_stack_nr);
2122 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2123 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2124 } else {
2125 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2127 poi_stack[poi_stack_nr++] = obj_offset;
2128 /* If parsing the base offset fails, just unwind */
2129 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2130 if (!base_offset)
2131 goto unwind;
2132 curpos = obj_offset = base_offset;
2133 type = unpack_object_header(p, w_curs, &curpos, &size);
2134 if (type <= OBJ_NONE) {
2135 /* If getting the base itself fails, we first
2136 * retry the base, otherwise unwind */
2137 type = retry_bad_packed_offset(p, base_offset);
2138 if (type > OBJ_NONE)
2139 goto out;
2140 goto unwind;
2144 switch (type) {
2145 case OBJ_BAD:
2146 case OBJ_COMMIT:
2147 case OBJ_TREE:
2148 case OBJ_BLOB:
2149 case OBJ_TAG:
2150 break;
2151 default:
2152 error("unknown object type %i at offset %"PRIuMAX" in %s",
2153 type, (uintmax_t)obj_offset, p->pack_name);
2154 type = OBJ_BAD;
2157 out:
2158 if (poi_stack != small_poi_stack)
2159 free(poi_stack);
2160 return type;
2162 unwind:
2163 while (poi_stack_nr) {
2164 obj_offset = poi_stack[--poi_stack_nr];
2165 type = retry_bad_packed_offset(p, obj_offset);
2166 if (type > OBJ_NONE)
2167 goto out;
2169 type = OBJ_BAD;
2170 goto out;
2173 int packed_object_info(struct packed_git *p, off_t obj_offset,
2174 struct object_info *oi)
2176 struct pack_window *w_curs = NULL;
2177 unsigned long size;
2178 off_t curpos = obj_offset;
2179 enum object_type type;
2182 * We always get the representation type, but only convert it to
2183 * a "real" type later if the caller is interested.
2185 type = unpack_object_header(p, &w_curs, &curpos, &size);
2187 if (oi->sizep) {
2188 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2189 off_t tmp_pos = curpos;
2190 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2191 type, obj_offset);
2192 if (!base_offset) {
2193 type = OBJ_BAD;
2194 goto out;
2196 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2197 if (*oi->sizep == 0) {
2198 type = OBJ_BAD;
2199 goto out;
2201 } else {
2202 *oi->sizep = size;
2206 if (oi->disk_sizep) {
2207 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2208 *oi->disk_sizep = revidx[1].offset - obj_offset;
2211 if (oi->typep) {
2212 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2213 if (*oi->typep < 0) {
2214 type = OBJ_BAD;
2215 goto out;
2219 if (oi->delta_base_sha1) {
2220 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2221 const unsigned char *base;
2223 base = get_delta_base_sha1(p, &w_curs, curpos,
2224 type, obj_offset);
2225 if (!base) {
2226 type = OBJ_BAD;
2227 goto out;
2230 hashcpy(oi->delta_base_sha1, base);
2231 } else
2232 hashclr(oi->delta_base_sha1);
2235 out:
2236 unuse_pack(&w_curs);
2237 return type;
2240 static void *unpack_compressed_entry(struct packed_git *p,
2241 struct pack_window **w_curs,
2242 off_t curpos,
2243 unsigned long size)
2245 int st;
2246 git_zstream stream;
2247 unsigned char *buffer, *in;
2249 buffer = xmallocz_gently(size);
2250 if (!buffer)
2251 return NULL;
2252 memset(&stream, 0, sizeof(stream));
2253 stream.next_out = buffer;
2254 stream.avail_out = size + 1;
2256 git_inflate_init(&stream);
2257 do {
2258 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2259 stream.next_in = in;
2260 st = git_inflate(&stream, Z_FINISH);
2261 if (!stream.avail_out)
2262 break; /* the payload is larger than it should be */
2263 curpos += stream.next_in - in;
2264 } while (st == Z_OK || st == Z_BUF_ERROR);
2265 git_inflate_end(&stream);
2266 if ((st != Z_STREAM_END) || stream.total_out != size) {
2267 free(buffer);
2268 return NULL;
2271 return buffer;
2274 static struct hashmap delta_base_cache;
2275 static size_t delta_base_cached;
2277 static LIST_HEAD(delta_base_cache_lru);
2279 struct delta_base_cache_key {
2280 struct packed_git *p;
2281 off_t base_offset;
2284 struct delta_base_cache_entry {
2285 struct hashmap hash;
2286 struct delta_base_cache_key key;
2287 struct list_head lru;
2288 void *data;
2289 unsigned long size;
2290 enum object_type type;
2293 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2295 unsigned int hash;
2297 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2298 hash += (hash >> 8) + (hash >> 16);
2299 return hash;
2302 static struct delta_base_cache_entry *
2303 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2305 struct hashmap_entry entry;
2306 struct delta_base_cache_key key;
2308 if (!delta_base_cache.cmpfn)
2309 return NULL;
2311 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2312 key.p = p;
2313 key.base_offset = base_offset;
2314 return hashmap_get(&delta_base_cache, &entry, &key);
2317 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2318 const struct delta_base_cache_key *b)
2320 return a->p == b->p && a->base_offset == b->base_offset;
2323 static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2324 const void *vkey)
2326 const struct delta_base_cache_entry *a = va, *b = vb;
2327 const struct delta_base_cache_key *key = vkey;
2328 if (key)
2329 return !delta_base_cache_key_eq(&a->key, key);
2330 else
2331 return !delta_base_cache_key_eq(&a->key, &b->key);
2334 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2336 return !!get_delta_base_cache_entry(p, base_offset);
2340 * Remove the entry from the cache, but do _not_ free the associated
2341 * entry data. The caller takes ownership of the "data" buffer, and
2342 * should copy out any fields it wants before detaching.
2344 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2346 hashmap_remove(&delta_base_cache, ent, &ent->key);
2347 list_del(&ent->lru);
2348 delta_base_cached -= ent->size;
2349 free(ent);
2352 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2353 unsigned long *base_size, enum object_type *type)
2355 struct delta_base_cache_entry *ent;
2357 ent = get_delta_base_cache_entry(p, base_offset);
2358 if (!ent)
2359 return unpack_entry(p, base_offset, type, base_size);
2361 *type = ent->type;
2362 *base_size = ent->size;
2363 return xmemdupz(ent->data, ent->size);
2366 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2368 free(ent->data);
2369 detach_delta_base_cache_entry(ent);
2372 void clear_delta_base_cache(void)
2374 struct list_head *lru, *tmp;
2375 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2376 struct delta_base_cache_entry *entry =
2377 list_entry(lru, struct delta_base_cache_entry, lru);
2378 release_delta_base_cache(entry);
2382 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2383 void *base, unsigned long base_size, enum object_type type)
2385 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2386 struct list_head *lru, *tmp;
2388 delta_base_cached += base_size;
2390 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2391 struct delta_base_cache_entry *f =
2392 list_entry(lru, struct delta_base_cache_entry, lru);
2393 if (delta_base_cached <= delta_base_cache_limit)
2394 break;
2395 release_delta_base_cache(f);
2398 ent->key.p = p;
2399 ent->key.base_offset = base_offset;
2400 ent->type = type;
2401 ent->data = base;
2402 ent->size = base_size;
2403 list_add_tail(&ent->lru, &delta_base_cache_lru);
2405 if (!delta_base_cache.cmpfn)
2406 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2407 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2408 hashmap_add(&delta_base_cache, ent);
2411 static void *read_object(const unsigned char *sha1, enum object_type *type,
2412 unsigned long *size);
2414 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2416 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2417 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2418 p->pack_name, (uintmax_t)obj_offset);
2421 int do_check_packed_object_crc;
2423 #define UNPACK_ENTRY_STACK_PREALLOC 64
2424 struct unpack_entry_stack_ent {
2425 off_t obj_offset;
2426 off_t curpos;
2427 unsigned long size;
2430 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2431 enum object_type *final_type, unsigned long *final_size)
2433 struct pack_window *w_curs = NULL;
2434 off_t curpos = obj_offset;
2435 void *data = NULL;
2436 unsigned long size;
2437 enum object_type type;
2438 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2439 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2440 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2441 int base_from_cache = 0;
2443 write_pack_access_log(p, obj_offset);
2445 /* PHASE 1: drill down to the innermost base object */
2446 for (;;) {
2447 off_t base_offset;
2448 int i;
2449 struct delta_base_cache_entry *ent;
2451 ent = get_delta_base_cache_entry(p, curpos);
2452 if (ent) {
2453 type = ent->type;
2454 data = ent->data;
2455 size = ent->size;
2456 detach_delta_base_cache_entry(ent);
2457 base_from_cache = 1;
2458 break;
2461 if (do_check_packed_object_crc && p->index_version > 1) {
2462 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2463 off_t len = revidx[1].offset - obj_offset;
2464 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2465 const unsigned char *sha1 =
2466 nth_packed_object_sha1(p, revidx->nr);
2467 error("bad packed object CRC for %s",
2468 sha1_to_hex(sha1));
2469 mark_bad_packed_object(p, sha1);
2470 unuse_pack(&w_curs);
2471 return NULL;
2475 type = unpack_object_header(p, &w_curs, &curpos, &size);
2476 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2477 break;
2479 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2480 if (!base_offset) {
2481 error("failed to validate delta base reference "
2482 "at offset %"PRIuMAX" from %s",
2483 (uintmax_t)curpos, p->pack_name);
2484 /* bail to phase 2, in hopes of recovery */
2485 data = NULL;
2486 break;
2489 /* push object, proceed to base */
2490 if (delta_stack_nr >= delta_stack_alloc
2491 && delta_stack == small_delta_stack) {
2492 delta_stack_alloc = alloc_nr(delta_stack_nr);
2493 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2494 memcpy(delta_stack, small_delta_stack,
2495 sizeof(*delta_stack)*delta_stack_nr);
2496 } else {
2497 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2499 i = delta_stack_nr++;
2500 delta_stack[i].obj_offset = obj_offset;
2501 delta_stack[i].curpos = curpos;
2502 delta_stack[i].size = size;
2504 curpos = obj_offset = base_offset;
2507 /* PHASE 2: handle the base */
2508 switch (type) {
2509 case OBJ_OFS_DELTA:
2510 case OBJ_REF_DELTA:
2511 if (data)
2512 die("BUG: unpack_entry: left loop at a valid delta");
2513 break;
2514 case OBJ_COMMIT:
2515 case OBJ_TREE:
2516 case OBJ_BLOB:
2517 case OBJ_TAG:
2518 if (!base_from_cache)
2519 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2520 break;
2521 default:
2522 data = NULL;
2523 error("unknown object type %i at offset %"PRIuMAX" in %s",
2524 type, (uintmax_t)obj_offset, p->pack_name);
2527 /* PHASE 3: apply deltas in order */
2529 /* invariants:
2530 * 'data' holds the base data, or NULL if there was corruption
2532 while (delta_stack_nr) {
2533 void *delta_data;
2534 void *base = data;
2535 unsigned long delta_size, base_size = size;
2536 int i;
2538 data = NULL;
2540 if (base)
2541 add_delta_base_cache(p, obj_offset, base, base_size, type);
2543 if (!base) {
2545 * We're probably in deep shit, but let's try to fetch
2546 * the required base anyway from another pack or loose.
2547 * This is costly but should happen only in the presence
2548 * of a corrupted pack, and is better than failing outright.
2550 struct revindex_entry *revidx;
2551 const unsigned char *base_sha1;
2552 revidx = find_pack_revindex(p, obj_offset);
2553 if (revidx) {
2554 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2555 error("failed to read delta base object %s"
2556 " at offset %"PRIuMAX" from %s",
2557 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2558 p->pack_name);
2559 mark_bad_packed_object(p, base_sha1);
2560 base = read_object(base_sha1, &type, &base_size);
2564 i = --delta_stack_nr;
2565 obj_offset = delta_stack[i].obj_offset;
2566 curpos = delta_stack[i].curpos;
2567 delta_size = delta_stack[i].size;
2569 if (!base)
2570 continue;
2572 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2574 if (!delta_data) {
2575 error("failed to unpack compressed delta "
2576 "at offset %"PRIuMAX" from %s",
2577 (uintmax_t)curpos, p->pack_name);
2578 data = NULL;
2579 continue;
2582 data = patch_delta(base, base_size,
2583 delta_data, delta_size,
2584 &size);
2587 * We could not apply the delta; warn the user, but keep going.
2588 * Our failure will be noticed either in the next iteration of
2589 * the loop, or if this is the final delta, in the caller when
2590 * we return NULL. Those code paths will take care of making
2591 * a more explicit warning and retrying with another copy of
2592 * the object.
2594 if (!data)
2595 error("failed to apply delta");
2597 free(delta_data);
2600 *final_type = type;
2601 *final_size = size;
2603 unuse_pack(&w_curs);
2605 if (delta_stack != small_delta_stack)
2606 free(delta_stack);
2608 return data;
2611 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2612 uint32_t n)
2614 const unsigned char *index = p->index_data;
2615 if (!index) {
2616 if (open_pack_index(p))
2617 return NULL;
2618 index = p->index_data;
2620 if (n >= p->num_objects)
2621 return NULL;
2622 index += 4 * 256;
2623 if (p->index_version == 1) {
2624 return index + 24 * n + 4;
2625 } else {
2626 index += 8;
2627 return index + 20 * n;
2631 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2633 const unsigned char *ptr = vptr;
2634 const unsigned char *start = p->index_data;
2635 const unsigned char *end = start + p->index_size;
2636 if (ptr < start)
2637 die(_("offset before start of pack index for %s (corrupt index?)"),
2638 p->pack_name);
2639 /* No need to check for underflow; .idx files must be at least 8 bytes */
2640 if (ptr >= end - 8)
2641 die(_("offset beyond end of pack index for %s (truncated index?)"),
2642 p->pack_name);
2645 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2647 const unsigned char *index = p->index_data;
2648 index += 4 * 256;
2649 if (p->index_version == 1) {
2650 return ntohl(*((uint32_t *)(index + 24 * n)));
2651 } else {
2652 uint32_t off;
2653 index += 8 + p->num_objects * (20 + 4);
2654 off = ntohl(*((uint32_t *)(index + 4 * n)));
2655 if (!(off & 0x80000000))
2656 return off;
2657 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2658 check_pack_index_ptr(p, index);
2659 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2660 ntohl(*((uint32_t *)(index + 4)));
2664 off_t find_pack_entry_one(const unsigned char *sha1,
2665 struct packed_git *p)
2667 const uint32_t *level1_ofs = p->index_data;
2668 const unsigned char *index = p->index_data;
2669 unsigned hi, lo, stride;
2670 static int use_lookup = -1;
2671 static int debug_lookup = -1;
2673 if (debug_lookup < 0)
2674 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2676 if (!index) {
2677 if (open_pack_index(p))
2678 return 0;
2679 level1_ofs = p->index_data;
2680 index = p->index_data;
2682 if (p->index_version > 1) {
2683 level1_ofs += 2;
2684 index += 8;
2686 index += 4 * 256;
2687 hi = ntohl(level1_ofs[*sha1]);
2688 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2689 if (p->index_version > 1) {
2690 stride = 20;
2691 } else {
2692 stride = 24;
2693 index += 4;
2696 if (debug_lookup)
2697 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2698 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2700 if (use_lookup < 0)
2701 use_lookup = !!getenv("GIT_USE_LOOKUP");
2702 if (use_lookup) {
2703 int pos = sha1_entry_pos(index, stride, 0,
2704 lo, hi, p->num_objects, sha1);
2705 if (pos < 0)
2706 return 0;
2707 return nth_packed_object_offset(p, pos);
2710 do {
2711 unsigned mi = (lo + hi) / 2;
2712 int cmp = hashcmp(index + mi * stride, sha1);
2714 if (debug_lookup)
2715 printf("lo %u hi %u rg %u mi %u\n",
2716 lo, hi, hi - lo, mi);
2717 if (!cmp)
2718 return nth_packed_object_offset(p, mi);
2719 if (cmp > 0)
2720 hi = mi;
2721 else
2722 lo = mi+1;
2723 } while (lo < hi);
2724 return 0;
2727 int is_pack_valid(struct packed_git *p)
2729 /* An already open pack is known to be valid. */
2730 if (p->pack_fd != -1)
2731 return 1;
2733 /* If the pack has one window completely covering the
2734 * file size, the pack is known to be valid even if
2735 * the descriptor is not currently open.
2737 if (p->windows) {
2738 struct pack_window *w = p->windows;
2740 if (!w->offset && w->len == p->pack_size)
2741 return 1;
2744 /* Force the pack to open to prove its valid. */
2745 return !open_packed_git(p);
2748 static int fill_pack_entry(const unsigned char *sha1,
2749 struct pack_entry *e,
2750 struct packed_git *p)
2752 off_t offset;
2754 if (p->num_bad_objects) {
2755 unsigned i;
2756 for (i = 0; i < p->num_bad_objects; i++)
2757 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2758 return 0;
2761 offset = find_pack_entry_one(sha1, p);
2762 if (!offset)
2763 return 0;
2766 * We are about to tell the caller where they can locate the
2767 * requested object. We better make sure the packfile is
2768 * still here and can be accessed before supplying that
2769 * answer, as it may have been deleted since the index was
2770 * loaded!
2772 if (!is_pack_valid(p))
2773 return 0;
2774 e->offset = offset;
2775 e->p = p;
2776 hashcpy(e->sha1, sha1);
2777 return 1;
2781 * Iff a pack file contains the object named by sha1, return true and
2782 * store its location to e.
2784 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2786 struct mru_entry *p;
2788 prepare_packed_git();
2789 if (!packed_git)
2790 return 0;
2792 for (p = packed_git_mru->head; p; p = p->next) {
2793 if (fill_pack_entry(sha1, e, p->item)) {
2794 mru_mark(packed_git_mru, p);
2795 return 1;
2798 return 0;
2801 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2802 struct packed_git *packs)
2804 struct packed_git *p;
2806 for (p = packs; p; p = p->next) {
2807 if (find_pack_entry_one(sha1, p))
2808 return p;
2810 return NULL;
2814 static int sha1_loose_object_info(const unsigned char *sha1,
2815 struct object_info *oi,
2816 int flags)
2818 int status = 0;
2819 unsigned long mapsize;
2820 void *map;
2821 git_zstream stream;
2822 char hdr[32];
2823 struct strbuf hdrbuf = STRBUF_INIT;
2825 if (oi->delta_base_sha1)
2826 hashclr(oi->delta_base_sha1);
2829 * If we don't care about type or size, then we don't
2830 * need to look inside the object at all. Note that we
2831 * do not optimize out the stat call, even if the
2832 * caller doesn't care about the disk-size, since our
2833 * return value implicitly indicates whether the
2834 * object even exists.
2836 if (!oi->typep && !oi->typename && !oi->sizep) {
2837 const char *path;
2838 struct stat st;
2839 if (stat_sha1_file(sha1, &st, &path) < 0)
2840 return -1;
2841 if (oi->disk_sizep)
2842 *oi->disk_sizep = st.st_size;
2843 return 0;
2846 map = map_sha1_file(sha1, &mapsize);
2847 if (!map)
2848 return -1;
2849 if (oi->disk_sizep)
2850 *oi->disk_sizep = mapsize;
2851 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2852 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2853 status = error("unable to unpack %s header with --allow-unknown-type",
2854 sha1_to_hex(sha1));
2855 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2856 status = error("unable to unpack %s header",
2857 sha1_to_hex(sha1));
2858 if (status < 0)
2859 ; /* Do nothing */
2860 else if (hdrbuf.len) {
2861 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2862 status = error("unable to parse %s header with --allow-unknown-type",
2863 sha1_to_hex(sha1));
2864 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2865 status = error("unable to parse %s header", sha1_to_hex(sha1));
2866 git_inflate_end(&stream);
2867 munmap(map, mapsize);
2868 if (status && oi->typep)
2869 *oi->typep = status;
2870 strbuf_release(&hdrbuf);
2871 return 0;
2874 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2876 struct cached_object *co;
2877 struct pack_entry e;
2878 int rtype;
2879 enum object_type real_type;
2880 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2882 co = find_cached_object(real);
2883 if (co) {
2884 if (oi->typep)
2885 *(oi->typep) = co->type;
2886 if (oi->sizep)
2887 *(oi->sizep) = co->size;
2888 if (oi->disk_sizep)
2889 *(oi->disk_sizep) = 0;
2890 if (oi->delta_base_sha1)
2891 hashclr(oi->delta_base_sha1);
2892 if (oi->typename)
2893 strbuf_addstr(oi->typename, typename(co->type));
2894 oi->whence = OI_CACHED;
2895 return 0;
2898 if (!find_pack_entry(real, &e)) {
2899 /* Most likely it's a loose object. */
2900 if (!sha1_loose_object_info(real, oi, flags)) {
2901 oi->whence = OI_LOOSE;
2902 return 0;
2905 /* Not a loose object; someone else may have just packed it. */
2906 reprepare_packed_git();
2907 if (!find_pack_entry(real, &e))
2908 return -1;
2912 * packed_object_info() does not follow the delta chain to
2913 * find out the real type, unless it is given oi->typep.
2915 if (oi->typename && !oi->typep)
2916 oi->typep = &real_type;
2918 rtype = packed_object_info(e.p, e.offset, oi);
2919 if (rtype < 0) {
2920 mark_bad_packed_object(e.p, real);
2921 if (oi->typep == &real_type)
2922 oi->typep = NULL;
2923 return sha1_object_info_extended(real, oi, 0);
2924 } else if (in_delta_base_cache(e.p, e.offset)) {
2925 oi->whence = OI_DBCACHED;
2926 } else {
2927 oi->whence = OI_PACKED;
2928 oi->u.packed.offset = e.offset;
2929 oi->u.packed.pack = e.p;
2930 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2931 rtype == OBJ_OFS_DELTA);
2933 if (oi->typename)
2934 strbuf_addstr(oi->typename, typename(*oi->typep));
2935 if (oi->typep == &real_type)
2936 oi->typep = NULL;
2938 return 0;
2941 /* returns enum object_type or negative */
2942 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2944 enum object_type type;
2945 struct object_info oi = OBJECT_INFO_INIT;
2947 oi.typep = &type;
2948 oi.sizep = sizep;
2949 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2950 return -1;
2951 return type;
2954 static void *read_packed_sha1(const unsigned char *sha1,
2955 enum object_type *type, unsigned long *size)
2957 struct pack_entry e;
2958 void *data;
2960 if (!find_pack_entry(sha1, &e))
2961 return NULL;
2962 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2963 if (!data) {
2965 * We're probably in deep shit, but let's try to fetch
2966 * the required object anyway from another pack or loose.
2967 * This should happen only in the presence of a corrupted
2968 * pack, and is better than failing outright.
2970 error("failed to read object %s at offset %"PRIuMAX" from %s",
2971 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2972 mark_bad_packed_object(e.p, sha1);
2973 data = read_object(sha1, type, size);
2975 return data;
2978 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2979 unsigned char *sha1)
2981 struct cached_object *co;
2983 hash_sha1_file(buf, len, typename(type), sha1);
2984 if (has_sha1_file(sha1) || find_cached_object(sha1))
2985 return 0;
2986 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2987 co = &cached_objects[cached_object_nr++];
2988 co->size = len;
2989 co->type = type;
2990 co->buf = xmalloc(len);
2991 memcpy(co->buf, buf, len);
2992 hashcpy(co->sha1, sha1);
2993 return 0;
2996 static void *read_object(const unsigned char *sha1, enum object_type *type,
2997 unsigned long *size)
2999 unsigned long mapsize;
3000 void *map, *buf;
3001 struct cached_object *co;
3003 co = find_cached_object(sha1);
3004 if (co) {
3005 *type = co->type;
3006 *size = co->size;
3007 return xmemdupz(co->buf, co->size);
3010 buf = read_packed_sha1(sha1, type, size);
3011 if (buf)
3012 return buf;
3013 map = map_sha1_file(sha1, &mapsize);
3014 if (map) {
3015 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
3016 munmap(map, mapsize);
3017 return buf;
3019 reprepare_packed_git();
3020 return read_packed_sha1(sha1, type, size);
3024 * This function dies on corrupt objects; the callers who want to
3025 * deal with them should arrange to call read_object() and give error
3026 * messages themselves.
3028 void *read_sha1_file_extended(const unsigned char *sha1,
3029 enum object_type *type,
3030 unsigned long *size,
3031 unsigned flag)
3033 void *data;
3034 const struct packed_git *p;
3035 const char *path;
3036 struct stat st;
3037 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3039 errno = 0;
3040 data = read_object(repl, type, size);
3041 if (data)
3042 return data;
3044 if (errno && errno != ENOENT)
3045 die_errno("failed to read object %s", sha1_to_hex(sha1));
3047 /* die if we replaced an object with one that does not exist */
3048 if (repl != sha1)
3049 die("replacement %s not found for %s",
3050 sha1_to_hex(repl), sha1_to_hex(sha1));
3052 if (!stat_sha1_file(repl, &st, &path))
3053 die("loose object %s (stored in %s) is corrupt",
3054 sha1_to_hex(repl), path);
3056 if ((p = has_packed_and_bad(repl)) != NULL)
3057 die("packed object %s (stored in %s) is corrupt",
3058 sha1_to_hex(repl), p->pack_name);
3060 return NULL;
3063 void *read_object_with_reference(const unsigned char *sha1,
3064 const char *required_type_name,
3065 unsigned long *size,
3066 unsigned char *actual_sha1_return)
3068 enum object_type type, required_type;
3069 void *buffer;
3070 unsigned long isize;
3071 unsigned char actual_sha1[20];
3073 required_type = type_from_string(required_type_name);
3074 hashcpy(actual_sha1, sha1);
3075 while (1) {
3076 int ref_length = -1;
3077 const char *ref_type = NULL;
3079 buffer = read_sha1_file(actual_sha1, &type, &isize);
3080 if (!buffer)
3081 return NULL;
3082 if (type == required_type) {
3083 *size = isize;
3084 if (actual_sha1_return)
3085 hashcpy(actual_sha1_return, actual_sha1);
3086 return buffer;
3088 /* Handle references */
3089 else if (type == OBJ_COMMIT)
3090 ref_type = "tree ";
3091 else if (type == OBJ_TAG)
3092 ref_type = "object ";
3093 else {
3094 free(buffer);
3095 return NULL;
3097 ref_length = strlen(ref_type);
3099 if (ref_length + 40 > isize ||
3100 memcmp(buffer, ref_type, ref_length) ||
3101 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3102 free(buffer);
3103 return NULL;
3105 free(buffer);
3106 /* Now we have the ID of the referred-to object in
3107 * actual_sha1. Check again. */
3111 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3112 const char *type, unsigned char *sha1,
3113 char *hdr, int *hdrlen)
3115 git_SHA_CTX c;
3117 /* Generate the header */
3118 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3120 /* Sha1.. */
3121 git_SHA1_Init(&c);
3122 git_SHA1_Update(&c, hdr, *hdrlen);
3123 git_SHA1_Update(&c, buf, len);
3124 git_SHA1_Final(sha1, &c);
3128 * Move the just written object into its final resting place.
3130 int finalize_object_file(const char *tmpfile, const char *filename)
3132 int ret = 0;
3134 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3135 goto try_rename;
3136 else if (link(tmpfile, filename))
3137 ret = errno;
3140 * Coda hack - coda doesn't like cross-directory links,
3141 * so we fall back to a rename, which will mean that it
3142 * won't be able to check collisions, but that's not a
3143 * big deal.
3145 * The same holds for FAT formatted media.
3147 * When this succeeds, we just return. We have nothing
3148 * left to unlink.
3150 if (ret && ret != EEXIST) {
3151 try_rename:
3152 if (!rename(tmpfile, filename))
3153 goto out;
3154 ret = errno;
3156 unlink_or_warn(tmpfile);
3157 if (ret) {
3158 if (ret != EEXIST) {
3159 return error_errno("unable to write sha1 filename %s", filename);
3161 /* FIXME!!! Collision check here ? */
3164 out:
3165 if (adjust_shared_perm(filename))
3166 return error("unable to set permission to '%s'", filename);
3167 return 0;
3170 static int write_buffer(int fd, const void *buf, size_t len)
3172 if (write_in_full(fd, buf, len) < 0)
3173 return error_errno("file write error");
3174 return 0;
3177 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3178 unsigned char *sha1)
3180 char hdr[32];
3181 int hdrlen = sizeof(hdr);
3182 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3183 return 0;
3186 /* Finalize a file on disk, and close it. */
3187 static void close_sha1_file(int fd)
3189 if (fsync_object_files)
3190 fsync_or_die(fd, "sha1 file");
3191 if (close(fd) != 0)
3192 die_errno("error when closing sha1 file");
3195 /* Size of directory component, including the ending '/' */
3196 static inline int directory_size(const char *filename)
3198 const char *s = strrchr(filename, '/');
3199 if (!s)
3200 return 0;
3201 return s - filename + 1;
3205 * This creates a temporary file in the same directory as the final
3206 * 'filename'
3208 * We want to avoid cross-directory filename renames, because those
3209 * can have problems on various filesystems (FAT, NFS, Coda).
3211 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3213 int fd, dirlen = directory_size(filename);
3215 strbuf_reset(tmp);
3216 strbuf_add(tmp, filename, dirlen);
3217 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3218 fd = git_mkstemp_mode(tmp->buf, 0444);
3219 if (fd < 0 && dirlen && errno == ENOENT) {
3221 * Make sure the directory exists; note that the contents
3222 * of the buffer are undefined after mkstemp returns an
3223 * error, so we have to rewrite the whole buffer from
3224 * scratch.
3226 strbuf_reset(tmp);
3227 strbuf_add(tmp, filename, dirlen - 1);
3228 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3229 return -1;
3230 if (adjust_shared_perm(tmp->buf))
3231 return -1;
3233 /* Try again */
3234 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3235 fd = git_mkstemp_mode(tmp->buf, 0444);
3237 return fd;
3240 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3241 const void *buf, unsigned long len, time_t mtime)
3243 int fd, ret;
3244 unsigned char compressed[4096];
3245 git_zstream stream;
3246 git_SHA_CTX c;
3247 unsigned char parano_sha1[20];
3248 static struct strbuf tmp_file = STRBUF_INIT;
3249 const char *filename = sha1_file_name(sha1);
3251 fd = create_tmpfile(&tmp_file, filename);
3252 if (fd < 0) {
3253 if (errno == EACCES)
3254 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3255 else
3256 return error_errno("unable to create temporary file");
3259 /* Set it up */
3260 git_deflate_init(&stream, zlib_compression_level);
3261 stream.next_out = compressed;
3262 stream.avail_out = sizeof(compressed);
3263 git_SHA1_Init(&c);
3265 /* First header.. */
3266 stream.next_in = (unsigned char *)hdr;
3267 stream.avail_in = hdrlen;
3268 while (git_deflate(&stream, 0) == Z_OK)
3269 ; /* nothing */
3270 git_SHA1_Update(&c, hdr, hdrlen);
3272 /* Then the data itself.. */
3273 stream.next_in = (void *)buf;
3274 stream.avail_in = len;
3275 do {
3276 unsigned char *in0 = stream.next_in;
3277 ret = git_deflate(&stream, Z_FINISH);
3278 git_SHA1_Update(&c, in0, stream.next_in - in0);
3279 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3280 die("unable to write sha1 file");
3281 stream.next_out = compressed;
3282 stream.avail_out = sizeof(compressed);
3283 } while (ret == Z_OK);
3285 if (ret != Z_STREAM_END)
3286 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3287 ret = git_deflate_end_gently(&stream);
3288 if (ret != Z_OK)
3289 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3290 git_SHA1_Final(parano_sha1, &c);
3291 if (hashcmp(sha1, parano_sha1) != 0)
3292 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3294 close_sha1_file(fd);
3296 if (mtime) {
3297 struct utimbuf utb;
3298 utb.actime = mtime;
3299 utb.modtime = mtime;
3300 if (utime(tmp_file.buf, &utb) < 0)
3301 warning_errno("failed utime() on %s", tmp_file.buf);
3304 return finalize_object_file(tmp_file.buf, filename);
3307 static int freshen_loose_object(const unsigned char *sha1)
3309 return check_and_freshen(sha1, 1);
3312 static int freshen_packed_object(const unsigned char *sha1)
3314 struct pack_entry e;
3315 if (!find_pack_entry(sha1, &e))
3316 return 0;
3317 if (e.p->freshened)
3318 return 1;
3319 if (!freshen_file(e.p->pack_name))
3320 return 0;
3321 e.p->freshened = 1;
3322 return 1;
3325 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3327 char hdr[32];
3328 int hdrlen = sizeof(hdr);
3330 /* Normally if we have it in the pack then we do not bother writing
3331 * it out into .git/objects/??/?{38} file.
3333 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3334 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3335 return 0;
3336 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3339 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3340 unsigned char *sha1, unsigned flags)
3342 char *header;
3343 int hdrlen, status = 0;
3345 /* type string, SP, %lu of the length plus NUL must fit this */
3346 hdrlen = strlen(type) + 32;
3347 header = xmalloc(hdrlen);
3348 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3350 if (!(flags & HASH_WRITE_OBJECT))
3351 goto cleanup;
3352 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3353 goto cleanup;
3354 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3356 cleanup:
3357 free(header);
3358 return status;
3361 int force_object_loose(const unsigned char *sha1, time_t mtime)
3363 void *buf;
3364 unsigned long len;
3365 enum object_type type;
3366 char hdr[32];
3367 int hdrlen;
3368 int ret;
3370 if (has_loose_object(sha1))
3371 return 0;
3372 buf = read_packed_sha1(sha1, &type, &len);
3373 if (!buf)
3374 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3375 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3376 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3377 free(buf);
3379 return ret;
3382 int has_pack_index(const unsigned char *sha1)
3384 struct stat st;
3385 if (stat(sha1_pack_index_name(sha1), &st))
3386 return 0;
3387 return 1;
3390 int has_sha1_pack(const unsigned char *sha1)
3392 struct pack_entry e;
3393 return find_pack_entry(sha1, &e);
3396 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3398 struct pack_entry e;
3400 if (find_pack_entry(sha1, &e))
3401 return 1;
3402 if (has_loose_object(sha1))
3403 return 1;
3404 if (flags & HAS_SHA1_QUICK)
3405 return 0;
3406 reprepare_packed_git();
3407 return find_pack_entry(sha1, &e);
3410 int has_object_file(const struct object_id *oid)
3412 return has_sha1_file(oid->hash);
3415 int has_object_file_with_flags(const struct object_id *oid, int flags)
3417 return has_sha1_file_with_flags(oid->hash, flags);
3420 static void check_tree(const void *buf, size_t size)
3422 struct tree_desc desc;
3423 struct name_entry entry;
3425 init_tree_desc(&desc, buf, size);
3426 while (tree_entry(&desc, &entry))
3427 /* do nothing
3428 * tree_entry() will die() on malformed entries */
3432 static void check_commit(const void *buf, size_t size)
3434 struct commit c;
3435 memset(&c, 0, sizeof(c));
3436 if (parse_commit_buffer(&c, buf, size))
3437 die("corrupt commit");
3440 static void check_tag(const void *buf, size_t size)
3442 struct tag t;
3443 memset(&t, 0, sizeof(t));
3444 if (parse_tag_buffer(&t, buf, size))
3445 die("corrupt tag");
3448 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3449 enum object_type type,
3450 const char *path, unsigned flags)
3452 int ret, re_allocated = 0;
3453 int write_object = flags & HASH_WRITE_OBJECT;
3455 if (!type)
3456 type = OBJ_BLOB;
3459 * Convert blobs to git internal format
3461 if ((type == OBJ_BLOB) && path) {
3462 struct strbuf nbuf = STRBUF_INIT;
3463 if (convert_to_git(path, buf, size, &nbuf,
3464 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3465 buf = strbuf_detach(&nbuf, &size);
3466 re_allocated = 1;
3469 if (flags & HASH_FORMAT_CHECK) {
3470 if (type == OBJ_TREE)
3471 check_tree(buf, size);
3472 if (type == OBJ_COMMIT)
3473 check_commit(buf, size);
3474 if (type == OBJ_TAG)
3475 check_tag(buf, size);
3478 if (write_object)
3479 ret = write_sha1_file(buf, size, typename(type), sha1);
3480 else
3481 ret = hash_sha1_file(buf, size, typename(type), sha1);
3482 if (re_allocated)
3483 free(buf);
3484 return ret;
3487 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3488 const char *path, unsigned flags)
3490 int ret;
3491 const int write_object = flags & HASH_WRITE_OBJECT;
3492 struct strbuf sbuf = STRBUF_INIT;
3494 assert(path);
3495 assert(would_convert_to_git_filter_fd(path));
3497 convert_to_git_filter_fd(path, fd, &sbuf,
3498 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3500 if (write_object)
3501 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3502 sha1);
3503 else
3504 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3505 sha1);
3506 strbuf_release(&sbuf);
3507 return ret;
3510 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3511 const char *path, unsigned flags)
3513 struct strbuf sbuf = STRBUF_INIT;
3514 int ret;
3516 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3517 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3518 else
3519 ret = -1;
3520 strbuf_release(&sbuf);
3521 return ret;
3524 #define SMALL_FILE_SIZE (32*1024)
3526 static int index_core(unsigned char *sha1, int fd, size_t size,
3527 enum object_type type, const char *path,
3528 unsigned flags)
3530 int ret;
3532 if (!size) {
3533 ret = index_mem(sha1, "", size, type, path, flags);
3534 } else if (size <= SMALL_FILE_SIZE) {
3535 char *buf = xmalloc(size);
3536 if (size == read_in_full(fd, buf, size))
3537 ret = index_mem(sha1, buf, size, type, path, flags);
3538 else
3539 ret = error_errno("short read");
3540 free(buf);
3541 } else {
3542 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3543 ret = index_mem(sha1, buf, size, type, path, flags);
3544 munmap(buf, size);
3546 return ret;
3550 * This creates one packfile per large blob unless bulk-checkin
3551 * machinery is "plugged".
3553 * This also bypasses the usual "convert-to-git" dance, and that is on
3554 * purpose. We could write a streaming version of the converting
3555 * functions and insert that before feeding the data to fast-import
3556 * (or equivalent in-core API described above). However, that is
3557 * somewhat complicated, as we do not know the size of the filter
3558 * result, which we need to know beforehand when writing a git object.
3559 * Since the primary motivation for trying to stream from the working
3560 * tree file and to avoid mmaping it in core is to deal with large
3561 * binary blobs, they generally do not want to get any conversion, and
3562 * callers should avoid this code path when filters are requested.
3564 static int index_stream(unsigned char *sha1, int fd, size_t size,
3565 enum object_type type, const char *path,
3566 unsigned flags)
3568 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3571 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3572 enum object_type type, const char *path, unsigned flags)
3574 int ret;
3577 * Call xsize_t() only when needed to avoid potentially unnecessary
3578 * die() for large files.
3580 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3581 ret = index_stream_convert_blob(sha1, fd, path, flags);
3582 else if (!S_ISREG(st->st_mode))
3583 ret = index_pipe(sha1, fd, type, path, flags);
3584 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3585 (path && would_convert_to_git(path)))
3586 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3587 flags);
3588 else
3589 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3590 flags);
3591 close(fd);
3592 return ret;
3595 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3597 int fd;
3598 struct strbuf sb = STRBUF_INIT;
3600 switch (st->st_mode & S_IFMT) {
3601 case S_IFREG:
3602 fd = open(path, O_RDONLY);
3603 if (fd < 0)
3604 return error_errno("open(\"%s\")", path);
3605 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3606 return error("%s: failed to insert into database",
3607 path);
3608 break;
3609 case S_IFLNK:
3610 if (strbuf_readlink(&sb, path, st->st_size))
3611 return error_errno("readlink(\"%s\")", path);
3612 if (!(flags & HASH_WRITE_OBJECT))
3613 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3614 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3615 return error("%s: failed to insert into database",
3616 path);
3617 strbuf_release(&sb);
3618 break;
3619 case S_IFDIR:
3620 return resolve_gitlink_ref(path, "HEAD", sha1);
3621 default:
3622 return error("%s: unsupported file type", path);
3624 return 0;
3627 int read_pack_header(int fd, struct pack_header *header)
3629 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3630 /* "eof before pack header was fully read" */
3631 return PH_ERROR_EOF;
3633 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3634 /* "protocol error (pack signature mismatch detected)" */
3635 return PH_ERROR_PACK_SIGNATURE;
3636 if (!pack_version_ok(header->hdr_version))
3637 /* "protocol error (pack version unsupported)" */
3638 return PH_ERROR_PROTOCOL;
3639 return 0;
3642 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3644 enum object_type type = sha1_object_info(sha1, NULL);
3645 if (type < 0)
3646 die("%s is not a valid object", sha1_to_hex(sha1));
3647 if (type != expect)
3648 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3649 typename(expect));
3652 static int for_each_file_in_obj_subdir(int subdir_nr,
3653 struct strbuf *path,
3654 each_loose_object_fn obj_cb,
3655 each_loose_cruft_fn cruft_cb,
3656 each_loose_subdir_fn subdir_cb,
3657 void *data)
3659 size_t baselen = path->len;
3660 DIR *dir = opendir(path->buf);
3661 struct dirent *de;
3662 int r = 0;
3664 if (!dir) {
3665 if (errno == ENOENT)
3666 return 0;
3667 return error_errno("unable to open %s", path->buf);
3670 while ((de = readdir(dir))) {
3671 if (is_dot_or_dotdot(de->d_name))
3672 continue;
3674 strbuf_setlen(path, baselen);
3675 strbuf_addf(path, "/%s", de->d_name);
3677 if (strlen(de->d_name) == 38) {
3678 char hex[41];
3679 unsigned char sha1[20];
3681 snprintf(hex, sizeof(hex), "%02x%s",
3682 subdir_nr, de->d_name);
3683 if (!get_sha1_hex(hex, sha1)) {
3684 if (obj_cb) {
3685 r = obj_cb(sha1, path->buf, data);
3686 if (r)
3687 break;
3689 continue;
3693 if (cruft_cb) {
3694 r = cruft_cb(de->d_name, path->buf, data);
3695 if (r)
3696 break;
3699 closedir(dir);
3701 strbuf_setlen(path, baselen);
3702 if (!r && subdir_cb)
3703 r = subdir_cb(subdir_nr, path->buf, data);
3705 return r;
3708 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3709 each_loose_object_fn obj_cb,
3710 each_loose_cruft_fn cruft_cb,
3711 each_loose_subdir_fn subdir_cb,
3712 void *data)
3714 size_t baselen = path->len;
3715 int r = 0;
3716 int i;
3718 for (i = 0; i < 256; i++) {
3719 strbuf_addf(path, "/%02x", i);
3720 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3721 subdir_cb, data);
3722 strbuf_setlen(path, baselen);
3723 if (r)
3724 break;
3727 return r;
3730 int for_each_loose_file_in_objdir(const char *path,
3731 each_loose_object_fn obj_cb,
3732 each_loose_cruft_fn cruft_cb,
3733 each_loose_subdir_fn subdir_cb,
3734 void *data)
3736 struct strbuf buf = STRBUF_INIT;
3737 int r;
3739 strbuf_addstr(&buf, path);
3740 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3741 subdir_cb, data);
3742 strbuf_release(&buf);
3744 return r;
3747 struct loose_alt_odb_data {
3748 each_loose_object_fn *cb;
3749 void *data;
3752 static int loose_from_alt_odb(struct alternate_object_database *alt,
3753 void *vdata)
3755 struct loose_alt_odb_data *data = vdata;
3756 struct strbuf buf = STRBUF_INIT;
3757 int r;
3759 strbuf_addstr(&buf, alt->path);
3760 r = for_each_loose_file_in_objdir_buf(&buf,
3761 data->cb, NULL, NULL,
3762 data->data);
3763 strbuf_release(&buf);
3764 return r;
3767 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3769 struct loose_alt_odb_data alt;
3770 int r;
3772 r = for_each_loose_file_in_objdir(get_object_directory(),
3773 cb, NULL, NULL, data);
3774 if (r)
3775 return r;
3777 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3778 return 0;
3780 alt.cb = cb;
3781 alt.data = data;
3782 return foreach_alt_odb(loose_from_alt_odb, &alt);
3785 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3787 uint32_t i;
3788 int r = 0;
3790 for (i = 0; i < p->num_objects; i++) {
3791 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3793 if (!sha1)
3794 return error("unable to get sha1 of object %u in %s",
3795 i, p->pack_name);
3797 r = cb(sha1, p, i, data);
3798 if (r)
3799 break;
3801 return r;
3804 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3806 struct packed_git *p;
3807 int r = 0;
3808 int pack_errors = 0;
3810 prepare_packed_git();
3811 for (p = packed_git; p; p = p->next) {
3812 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3813 continue;
3814 if (open_pack_index(p)) {
3815 pack_errors = 1;
3816 continue;
3818 r = for_each_object_in_pack(p, cb, data);
3819 if (r)
3820 break;
3822 return r ? r : pack_errors;
3825 static int check_stream_sha1(git_zstream *stream,
3826 const char *hdr,
3827 unsigned long size,
3828 const char *path,
3829 const unsigned char *expected_sha1)
3831 git_SHA_CTX c;
3832 unsigned char real_sha1[GIT_SHA1_RAWSZ];
3833 unsigned char buf[4096];
3834 unsigned long total_read;
3835 int status = Z_OK;
3837 git_SHA1_Init(&c);
3838 git_SHA1_Update(&c, hdr, stream->total_out);
3841 * We already read some bytes into hdr, but the ones up to the NUL
3842 * do not count against the object's content size.
3844 total_read = stream->total_out - strlen(hdr) - 1;
3847 * This size comparison must be "<=" to read the final zlib packets;
3848 * see the comment in unpack_sha1_rest for details.
3850 while (total_read <= size &&
3851 (status == Z_OK || status == Z_BUF_ERROR)) {
3852 stream->next_out = buf;
3853 stream->avail_out = sizeof(buf);
3854 if (size - total_read < stream->avail_out)
3855 stream->avail_out = size - total_read;
3856 status = git_inflate(stream, Z_FINISH);
3857 git_SHA1_Update(&c, buf, stream->next_out - buf);
3858 total_read += stream->next_out - buf;
3860 git_inflate_end(stream);
3862 if (status != Z_STREAM_END) {
3863 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3864 return -1;
3866 if (stream->avail_in) {
3867 error("garbage at end of loose object '%s'",
3868 sha1_to_hex(expected_sha1));
3869 return -1;
3872 git_SHA1_Final(real_sha1, &c);
3873 if (hashcmp(expected_sha1, real_sha1)) {
3874 error("sha1 mismatch for %s (expected %s)", path,
3875 sha1_to_hex(expected_sha1));
3876 return -1;
3879 return 0;
3882 int read_loose_object(const char *path,
3883 const unsigned char *expected_sha1,
3884 enum object_type *type,
3885 unsigned long *size,
3886 void **contents)
3888 int ret = -1;
3889 int fd = -1;
3890 void *map = NULL;
3891 unsigned long mapsize;
3892 git_zstream stream;
3893 char hdr[32];
3895 *contents = NULL;
3897 map = map_sha1_file_1(path, NULL, &mapsize);
3898 if (!map) {
3899 error_errno("unable to mmap %s", path);
3900 goto out;
3903 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3904 error("unable to unpack header of %s", path);
3905 goto out;
3908 *type = parse_sha1_header(hdr, size);
3909 if (*type < 0) {
3910 error("unable to parse header of %s", path);
3911 git_inflate_end(&stream);
3912 goto out;
3915 if (*type == OBJ_BLOB) {
3916 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3917 goto out;
3918 } else {
3919 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3920 if (!*contents) {
3921 error("unable to unpack contents of %s", path);
3922 git_inflate_end(&stream);
3923 goto out;
3925 if (check_sha1_signature(expected_sha1, *contents,
3926 *size, typename(*type))) {
3927 error("sha1 mismatch for %s (expected %s)", path,
3928 sha1_to_hex(expected_sha1));
3929 free(*contents);
3930 goto out;
3934 ret = 0; /* everything checks out */
3936 out:
3937 if (map)
3938 munmap(map, mapsize);
3939 if (fd >= 0)
3940 close(fd);
3941 return ret;