Documentation/stash: remove mention of git reset --hard
[git.git] / sha1_file.c
blob1173071859dae68f72cc72efb20f816152d3eabc
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 #ifndef O_NOATIME
32 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
33 #define O_NOATIME 01000000
34 #else
35 #define O_NOATIME 0
36 #endif
37 #endif
39 #define SZ_FMT PRIuMAX
40 static inline uintmax_t sz_fmt(size_t s) { return s; }
42 const unsigned char null_sha1[20];
43 const struct object_id null_oid;
44 const struct object_id empty_tree_oid = {
45 EMPTY_TREE_SHA1_BIN_LITERAL
47 const struct object_id empty_blob_oid = {
48 EMPTY_BLOB_SHA1_BIN_LITERAL
52 * This is meant to hold a *small* number of objects that you would
53 * want read_sha1_file() to be able to return, but yet you do not want
54 * to write them into the object store (e.g. a browse-only
55 * application).
57 static struct cached_object {
58 unsigned char sha1[20];
59 enum object_type type;
60 void *buf;
61 unsigned long size;
62 } *cached_objects;
63 static int cached_object_nr, cached_object_alloc;
65 static struct cached_object empty_tree = {
66 EMPTY_TREE_SHA1_BIN_LITERAL,
67 OBJ_TREE,
68 "",
72 static struct cached_object *find_cached_object(const unsigned char *sha1)
74 int i;
75 struct cached_object *co = cached_objects;
77 for (i = 0; i < cached_object_nr; i++, co++) {
78 if (!hashcmp(co->sha1, sha1))
79 return co;
81 if (!hashcmp(sha1, empty_tree.sha1))
82 return &empty_tree;
83 return NULL;
86 int mkdir_in_gitdir(const char *path)
88 if (mkdir(path, 0777)) {
89 int saved_errno = errno;
90 struct stat st;
91 struct strbuf sb = STRBUF_INIT;
93 if (errno != EEXIST)
94 return -1;
96 * Are we looking at a path in a symlinked worktree
97 * whose original repository does not yet have it?
98 * e.g. .git/rr-cache pointing at its original
99 * repository in which the user hasn't performed any
100 * conflict resolution yet?
102 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
103 strbuf_readlink(&sb, path, st.st_size) ||
104 !is_absolute_path(sb.buf) ||
105 mkdir(sb.buf, 0777)) {
106 strbuf_release(&sb);
107 errno = saved_errno;
108 return -1;
110 strbuf_release(&sb);
112 return adjust_shared_perm(path);
115 enum scld_error safe_create_leading_directories(char *path)
117 char *next_component = path + offset_1st_component(path);
118 enum scld_error ret = SCLD_OK;
120 while (ret == SCLD_OK && next_component) {
121 struct stat st;
122 char *slash = next_component, slash_character;
124 while (*slash && !is_dir_sep(*slash))
125 slash++;
127 if (!*slash)
128 break;
130 next_component = slash + 1;
131 while (is_dir_sep(*next_component))
132 next_component++;
133 if (!*next_component)
134 break;
136 slash_character = *slash;
137 *slash = '\0';
138 if (!stat(path, &st)) {
139 /* path exists */
140 if (!S_ISDIR(st.st_mode))
141 ret = SCLD_EXISTS;
142 } else if (mkdir(path, 0777)) {
143 if (errno == EEXIST &&
144 !stat(path, &st) && S_ISDIR(st.st_mode))
145 ; /* somebody created it since we checked */
146 else if (errno == ENOENT)
148 * Either mkdir() failed because
149 * somebody just pruned the containing
150 * directory, or stat() failed because
151 * the file that was in our way was
152 * just removed. Either way, inform
153 * the caller that it might be worth
154 * trying again:
156 ret = SCLD_VANISHED;
157 else
158 ret = SCLD_FAILED;
159 } else if (adjust_shared_perm(path)) {
160 ret = SCLD_PERMS;
162 *slash = slash_character;
164 return ret;
167 enum scld_error safe_create_leading_directories_const(const char *path)
169 /* path points to cache entries, so xstrdup before messing with it */
170 char *buf = xstrdup(path);
171 enum scld_error result = safe_create_leading_directories(buf);
172 free(buf);
173 return result;
176 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
178 int i;
179 for (i = 0; i < 20; i++) {
180 static char hex[] = "0123456789abcdef";
181 unsigned int val = sha1[i];
182 strbuf_addch(buf, hex[val >> 4]);
183 strbuf_addch(buf, hex[val & 0xf]);
184 if (!i)
185 strbuf_addch(buf, '/');
189 const char *sha1_file_name(const unsigned char *sha1)
191 static struct strbuf buf = STRBUF_INIT;
193 strbuf_reset(&buf);
194 strbuf_addf(&buf, "%s/", get_object_directory());
196 fill_sha1_path(&buf, sha1);
197 return buf.buf;
200 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
202 strbuf_setlen(&alt->scratch, alt->base_len);
203 return &alt->scratch;
206 static const char *alt_sha1_path(struct alternate_object_database *alt,
207 const unsigned char *sha1)
209 struct strbuf *buf = alt_scratch_buf(alt);
210 fill_sha1_path(buf, sha1);
211 return buf->buf;
215 * Return the name of the pack or index file with the specified sha1
216 * in its filename. *base and *name are scratch space that must be
217 * provided by the caller. which should be "pack" or "idx".
219 static char *sha1_get_pack_name(const unsigned char *sha1,
220 struct strbuf *buf,
221 const char *which)
223 strbuf_reset(buf);
224 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
225 sha1_to_hex(sha1), which);
226 return buf->buf;
229 char *sha1_pack_name(const unsigned char *sha1)
231 static struct strbuf buf = STRBUF_INIT;
232 return sha1_get_pack_name(sha1, &buf, "pack");
235 char *sha1_pack_index_name(const unsigned char *sha1)
237 static struct strbuf buf = STRBUF_INIT;
238 return sha1_get_pack_name(sha1, &buf, "idx");
241 struct alternate_object_database *alt_odb_list;
242 static struct alternate_object_database **alt_odb_tail;
245 * Return non-zero iff the path is usable as an alternate object database.
247 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
249 struct alternate_object_database *alt;
251 /* Detect cases where alternate disappeared */
252 if (!is_directory(path->buf)) {
253 error("object directory %s does not exist; "
254 "check .git/objects/info/alternates.",
255 path->buf);
256 return 0;
260 * Prevent the common mistake of listing the same
261 * thing twice, or object directory itself.
263 for (alt = alt_odb_list; alt; alt = alt->next) {
264 if (!fspathcmp(path->buf, alt->path))
265 return 0;
267 if (!fspathcmp(path->buf, normalized_objdir))
268 return 0;
270 return 1;
274 * Prepare alternate object database registry.
276 * The variable alt_odb_list points at the list of struct
277 * alternate_object_database. The elements on this list come from
278 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
279 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
280 * whose contents is similar to that environment variable but can be
281 * LF separated. Its base points at a statically allocated buffer that
282 * contains "/the/directory/corresponding/to/.git/objects/...", while
283 * its name points just after the slash at the end of ".git/objects/"
284 * in the example above, and has enough space to hold 40-byte hex
285 * SHA1, an extra slash for the first level indirection, and the
286 * terminating NUL.
288 static int link_alt_odb_entry(const char *entry, const char *relative_base,
289 int depth, const char *normalized_objdir)
291 struct alternate_object_database *ent;
292 struct strbuf pathbuf = STRBUF_INIT;
294 if (!is_absolute_path(entry) && relative_base) {
295 strbuf_addstr(&pathbuf, real_path(relative_base));
296 strbuf_addch(&pathbuf, '/');
298 strbuf_addstr(&pathbuf, entry);
300 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
301 error("unable to normalize alternate object path: %s",
302 pathbuf.buf);
303 strbuf_release(&pathbuf);
304 return -1;
308 * The trailing slash after the directory name is given by
309 * this function at the end. Remove duplicates.
311 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
312 strbuf_setlen(&pathbuf, pathbuf.len - 1);
314 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
315 strbuf_release(&pathbuf);
316 return -1;
319 ent = alloc_alt_odb(pathbuf.buf);
321 /* add the alternate entry */
322 *alt_odb_tail = ent;
323 alt_odb_tail = &(ent->next);
324 ent->next = NULL;
326 /* recursively add alternates */
327 read_info_alternates(pathbuf.buf, depth + 1);
329 strbuf_release(&pathbuf);
330 return 0;
333 static const char *parse_alt_odb_entry(const char *string,
334 int sep,
335 struct strbuf *out)
337 const char *end;
339 strbuf_reset(out);
341 if (*string == '#') {
342 /* comment; consume up to next separator */
343 end = strchrnul(string, sep);
344 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
346 * quoted path; unquote_c_style has copied the
347 * data for us and set "end". Broken quoting (e.g.,
348 * an entry that doesn't end with a quote) falls
349 * back to the unquoted case below.
351 } else {
352 /* normal, unquoted path */
353 end = strchrnul(string, sep);
354 strbuf_add(out, string, end - string);
357 if (*end)
358 end++;
359 return end;
362 static void link_alt_odb_entries(const char *alt, int len, int sep,
363 const char *relative_base, int depth)
365 struct strbuf objdirbuf = STRBUF_INIT;
366 struct strbuf entry = STRBUF_INIT;
368 if (depth > 5) {
369 error("%s: ignoring alternate object stores, nesting too deep.",
370 relative_base);
371 return;
374 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
375 if (strbuf_normalize_path(&objdirbuf) < 0)
376 die("unable to normalize object directory: %s",
377 objdirbuf.buf);
379 while (*alt) {
380 alt = parse_alt_odb_entry(alt, sep, &entry);
381 if (!entry.len)
382 continue;
383 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
385 strbuf_release(&entry);
386 strbuf_release(&objdirbuf);
389 void read_info_alternates(const char * relative_base, int depth)
391 char *map;
392 size_t mapsz;
393 struct stat st;
394 char *path;
395 int fd;
397 path = xstrfmt("%s/info/alternates", relative_base);
398 fd = git_open(path);
399 free(path);
400 if (fd < 0)
401 return;
402 if (fstat(fd, &st) || (st.st_size == 0)) {
403 close(fd);
404 return;
406 mapsz = xsize_t(st.st_size);
407 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
408 close(fd);
410 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
412 munmap(map, mapsz);
415 struct alternate_object_database *alloc_alt_odb(const char *dir)
417 struct alternate_object_database *ent;
419 FLEX_ALLOC_STR(ent, path, dir);
420 strbuf_init(&ent->scratch, 0);
421 strbuf_addf(&ent->scratch, "%s/", dir);
422 ent->base_len = ent->scratch.len;
424 return ent;
427 void add_to_alternates_file(const char *reference)
429 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
430 char *alts = git_pathdup("objects/info/alternates");
431 FILE *in, *out;
433 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
434 out = fdopen_lock_file(lock, "w");
435 if (!out)
436 die_errno("unable to fdopen alternates lockfile");
438 in = fopen(alts, "r");
439 if (in) {
440 struct strbuf line = STRBUF_INIT;
441 int found = 0;
443 while (strbuf_getline(&line, in) != EOF) {
444 if (!strcmp(reference, line.buf)) {
445 found = 1;
446 break;
448 fprintf_or_die(out, "%s\n", line.buf);
451 strbuf_release(&line);
452 fclose(in);
454 if (found) {
455 rollback_lock_file(lock);
456 lock = NULL;
459 else if (errno != ENOENT)
460 die_errno("unable to read alternates file");
462 if (lock) {
463 fprintf_or_die(out, "%s\n", reference);
464 if (commit_lock_file(lock))
465 die_errno("unable to move new alternates file into place");
466 if (alt_odb_tail)
467 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
469 free(alts);
472 void add_to_alternates_memory(const char *reference)
475 * Make sure alternates are initialized, or else our entry may be
476 * overwritten when they are.
478 prepare_alt_odb();
480 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
484 * Compute the exact path an alternate is at and returns it. In case of
485 * error NULL is returned and the human readable error is added to `err`
486 * `path` may be relative and should point to $GITDIR.
487 * `err` must not be null.
489 char *compute_alternate_path(const char *path, struct strbuf *err)
491 char *ref_git = NULL;
492 const char *repo, *ref_git_s;
493 int seen_error = 0;
495 ref_git_s = real_path_if_valid(path);
496 if (!ref_git_s) {
497 seen_error = 1;
498 strbuf_addf(err, _("path '%s' does not exist"), path);
499 goto out;
500 } else
502 * Beware: read_gitfile(), real_path() and mkpath()
503 * return static buffer
505 ref_git = xstrdup(ref_git_s);
507 repo = read_gitfile(ref_git);
508 if (!repo)
509 repo = read_gitfile(mkpath("%s/.git", ref_git));
510 if (repo) {
511 free(ref_git);
512 ref_git = xstrdup(repo);
515 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
516 char *ref_git_git = mkpathdup("%s/.git", ref_git);
517 free(ref_git);
518 ref_git = ref_git_git;
519 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
520 struct strbuf sb = STRBUF_INIT;
521 seen_error = 1;
522 if (get_common_dir(&sb, ref_git)) {
523 strbuf_addf(err,
524 _("reference repository '%s' as a linked "
525 "checkout is not supported yet."),
526 path);
527 goto out;
530 strbuf_addf(err, _("reference repository '%s' is not a "
531 "local repository."), path);
532 goto out;
535 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
536 strbuf_addf(err, _("reference repository '%s' is shallow"),
537 path);
538 seen_error = 1;
539 goto out;
542 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
543 strbuf_addf(err,
544 _("reference repository '%s' is grafted"),
545 path);
546 seen_error = 1;
547 goto out;
550 out:
551 if (seen_error) {
552 free(ref_git);
553 ref_git = NULL;
556 return ref_git;
559 int foreach_alt_odb(alt_odb_fn fn, void *cb)
561 struct alternate_object_database *ent;
562 int r = 0;
564 prepare_alt_odb();
565 for (ent = alt_odb_list; ent; ent = ent->next) {
566 r = fn(ent, cb);
567 if (r)
568 break;
570 return r;
573 void prepare_alt_odb(void)
575 const char *alt;
577 if (alt_odb_tail)
578 return;
580 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
581 if (!alt) alt = "";
583 alt_odb_tail = &alt_odb_list;
584 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
586 read_info_alternates(get_object_directory(), 0);
589 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
590 static int freshen_file(const char *fn)
592 struct utimbuf t;
593 t.actime = t.modtime = time(NULL);
594 return !utime(fn, &t);
598 * All of the check_and_freshen functions return 1 if the file exists and was
599 * freshened (if freshening was requested), 0 otherwise. If they return
600 * 0, you should not assume that it is safe to skip a write of the object (it
601 * either does not exist on disk, or has a stale mtime and may be subject to
602 * pruning).
604 static int check_and_freshen_file(const char *fn, int freshen)
606 if (access(fn, F_OK))
607 return 0;
608 if (freshen && !freshen_file(fn))
609 return 0;
610 return 1;
613 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
615 return check_and_freshen_file(sha1_file_name(sha1), freshen);
618 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
620 struct alternate_object_database *alt;
621 prepare_alt_odb();
622 for (alt = alt_odb_list; alt; alt = alt->next) {
623 const char *path = alt_sha1_path(alt, sha1);
624 if (check_and_freshen_file(path, freshen))
625 return 1;
627 return 0;
630 static int check_and_freshen(const unsigned char *sha1, int freshen)
632 return check_and_freshen_local(sha1, freshen) ||
633 check_and_freshen_nonlocal(sha1, freshen);
636 int has_loose_object_nonlocal(const unsigned char *sha1)
638 return check_and_freshen_nonlocal(sha1, 0);
641 static int has_loose_object(const unsigned char *sha1)
643 return check_and_freshen(sha1, 0);
646 static unsigned int pack_used_ctr;
647 static unsigned int pack_mmap_calls;
648 static unsigned int peak_pack_open_windows;
649 static unsigned int pack_open_windows;
650 static unsigned int pack_open_fds;
651 static unsigned int pack_max_fds;
652 static size_t peak_pack_mapped;
653 static size_t pack_mapped;
654 struct packed_git *packed_git;
656 static struct mru packed_git_mru_storage;
657 struct mru *packed_git_mru = &packed_git_mru_storage;
659 void pack_report(void)
661 fprintf(stderr,
662 "pack_report: getpagesize() = %10" SZ_FMT "\n"
663 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
664 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
665 sz_fmt(getpagesize()),
666 sz_fmt(packed_git_window_size),
667 sz_fmt(packed_git_limit));
668 fprintf(stderr,
669 "pack_report: pack_used_ctr = %10u\n"
670 "pack_report: pack_mmap_calls = %10u\n"
671 "pack_report: pack_open_windows = %10u / %10u\n"
672 "pack_report: pack_mapped = "
673 "%10" SZ_FMT " / %10" SZ_FMT "\n",
674 pack_used_ctr,
675 pack_mmap_calls,
676 pack_open_windows, peak_pack_open_windows,
677 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
681 * Open and mmap the index file at path, perform a couple of
682 * consistency checks, then record its information to p. Return 0 on
683 * success.
685 static int check_packed_git_idx(const char *path, struct packed_git *p)
687 void *idx_map;
688 struct pack_idx_header *hdr;
689 size_t idx_size;
690 uint32_t version, nr, i, *index;
691 int fd = git_open(path);
692 struct stat st;
694 if (fd < 0)
695 return -1;
696 if (fstat(fd, &st)) {
697 close(fd);
698 return -1;
700 idx_size = xsize_t(st.st_size);
701 if (idx_size < 4 * 256 + 20 + 20) {
702 close(fd);
703 return error("index file %s is too small", path);
705 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
706 close(fd);
708 hdr = idx_map;
709 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
710 version = ntohl(hdr->idx_version);
711 if (version < 2 || version > 2) {
712 munmap(idx_map, idx_size);
713 return error("index file %s is version %"PRIu32
714 " and is not supported by this binary"
715 " (try upgrading GIT to a newer version)",
716 path, version);
718 } else
719 version = 1;
721 nr = 0;
722 index = idx_map;
723 if (version > 1)
724 index += 2; /* skip index header */
725 for (i = 0; i < 256; i++) {
726 uint32_t n = ntohl(index[i]);
727 if (n < nr) {
728 munmap(idx_map, idx_size);
729 return error("non-monotonic index %s", path);
731 nr = n;
734 if (version == 1) {
736 * Total size:
737 * - 256 index entries 4 bytes each
738 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
739 * - 20-byte SHA1 of the packfile
740 * - 20-byte SHA1 file checksum
742 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
743 munmap(idx_map, idx_size);
744 return error("wrong index v1 file size in %s", path);
746 } else if (version == 2) {
748 * Minimum size:
749 * - 8 bytes of header
750 * - 256 index entries 4 bytes each
751 * - 20-byte sha1 entry * nr
752 * - 4-byte crc entry * nr
753 * - 4-byte offset entry * nr
754 * - 20-byte SHA1 of the packfile
755 * - 20-byte SHA1 file checksum
756 * And after the 4-byte offset table might be a
757 * variable sized table containing 8-byte entries
758 * for offsets larger than 2^31.
760 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
761 unsigned long max_size = min_size;
762 if (nr)
763 max_size += (nr - 1)*8;
764 if (idx_size < min_size || idx_size > max_size) {
765 munmap(idx_map, idx_size);
766 return error("wrong index v2 file size in %s", path);
768 if (idx_size != min_size &&
770 * make sure we can deal with large pack offsets.
771 * 31-bit signed offset won't be enough, neither
772 * 32-bit unsigned one will be.
774 (sizeof(off_t) <= 4)) {
775 munmap(idx_map, idx_size);
776 return error("pack too large for current definition of off_t in %s", path);
780 p->index_version = version;
781 p->index_data = idx_map;
782 p->index_size = idx_size;
783 p->num_objects = nr;
784 return 0;
787 int open_pack_index(struct packed_git *p)
789 char *idx_name;
790 size_t len;
791 int ret;
793 if (p->index_data)
794 return 0;
796 if (!strip_suffix(p->pack_name, ".pack", &len))
797 die("BUG: pack_name does not end in .pack");
798 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
799 ret = check_packed_git_idx(idx_name, p);
800 free(idx_name);
801 return ret;
804 static void scan_windows(struct packed_git *p,
805 struct packed_git **lru_p,
806 struct pack_window **lru_w,
807 struct pack_window **lru_l)
809 struct pack_window *w, *w_l;
811 for (w_l = NULL, w = p->windows; w; w = w->next) {
812 if (!w->inuse_cnt) {
813 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
814 *lru_p = p;
815 *lru_w = w;
816 *lru_l = w_l;
819 w_l = w;
823 static int unuse_one_window(struct packed_git *current)
825 struct packed_git *p, *lru_p = NULL;
826 struct pack_window *lru_w = NULL, *lru_l = NULL;
828 if (current)
829 scan_windows(current, &lru_p, &lru_w, &lru_l);
830 for (p = packed_git; p; p = p->next)
831 scan_windows(p, &lru_p, &lru_w, &lru_l);
832 if (lru_p) {
833 munmap(lru_w->base, lru_w->len);
834 pack_mapped -= lru_w->len;
835 if (lru_l)
836 lru_l->next = lru_w->next;
837 else
838 lru_p->windows = lru_w->next;
839 free(lru_w);
840 pack_open_windows--;
841 return 1;
843 return 0;
846 void release_pack_memory(size_t need)
848 size_t cur = pack_mapped;
849 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
850 ; /* nothing */
853 static void mmap_limit_check(size_t length)
855 static size_t limit = 0;
856 if (!limit) {
857 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
858 if (!limit)
859 limit = SIZE_MAX;
861 if (length > limit)
862 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
863 (uintmax_t)length, (uintmax_t)limit);
866 void *xmmap_gently(void *start, size_t length,
867 int prot, int flags, int fd, off_t offset)
869 void *ret;
871 mmap_limit_check(length);
872 ret = mmap(start, length, prot, flags, fd, offset);
873 if (ret == MAP_FAILED) {
874 if (!length)
875 return NULL;
876 release_pack_memory(length);
877 ret = mmap(start, length, prot, flags, fd, offset);
879 return ret;
882 void *xmmap(void *start, size_t length,
883 int prot, int flags, int fd, off_t offset)
885 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
886 if (ret == MAP_FAILED)
887 die_errno("mmap failed");
888 return ret;
891 void close_pack_windows(struct packed_git *p)
893 while (p->windows) {
894 struct pack_window *w = p->windows;
896 if (w->inuse_cnt)
897 die("pack '%s' still has open windows to it",
898 p->pack_name);
899 munmap(w->base, w->len);
900 pack_mapped -= w->len;
901 pack_open_windows--;
902 p->windows = w->next;
903 free(w);
907 static int close_pack_fd(struct packed_git *p)
909 if (p->pack_fd < 0)
910 return 0;
912 close(p->pack_fd);
913 pack_open_fds--;
914 p->pack_fd = -1;
916 return 1;
919 static void close_pack(struct packed_git *p)
921 close_pack_windows(p);
922 close_pack_fd(p);
923 close_pack_index(p);
926 void close_all_packs(void)
928 struct packed_git *p;
930 for (p = packed_git; p; p = p->next)
931 if (p->do_not_close)
932 die("BUG: want to close pack marked 'do-not-close'");
933 else
934 close_pack(p);
939 * The LRU pack is the one with the oldest MRU window, preferring packs
940 * with no used windows, or the oldest mtime if it has no windows allocated.
942 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
944 struct pack_window *w, *this_mru_w;
945 int has_windows_inuse = 0;
948 * Reject this pack if it has windows and the previously selected
949 * one does not. If this pack does not have windows, reject
950 * it if the pack file is newer than the previously selected one.
952 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
953 return;
955 for (w = this_mru_w = p->windows; w; w = w->next) {
957 * Reject this pack if any of its windows are in use,
958 * but the previously selected pack did not have any
959 * inuse windows. Otherwise, record that this pack
960 * has windows in use.
962 if (w->inuse_cnt) {
963 if (*accept_windows_inuse)
964 has_windows_inuse = 1;
965 else
966 return;
969 if (w->last_used > this_mru_w->last_used)
970 this_mru_w = w;
973 * Reject this pack if it has windows that have been
974 * used more recently than the previously selected pack.
975 * If the previously selected pack had windows inuse and
976 * we have not encountered a window in this pack that is
977 * inuse, skip this check since we prefer a pack with no
978 * inuse windows to one that has inuse windows.
980 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
981 this_mru_w->last_used > (*mru_w)->last_used)
982 return;
986 * Select this pack.
988 *mru_w = this_mru_w;
989 *lru_p = p;
990 *accept_windows_inuse = has_windows_inuse;
993 static int close_one_pack(void)
995 struct packed_git *p, *lru_p = NULL;
996 struct pack_window *mru_w = NULL;
997 int accept_windows_inuse = 1;
999 for (p = packed_git; p; p = p->next) {
1000 if (p->pack_fd == -1)
1001 continue;
1002 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1005 if (lru_p)
1006 return close_pack_fd(lru_p);
1008 return 0;
1011 void unuse_pack(struct pack_window **w_cursor)
1013 struct pack_window *w = *w_cursor;
1014 if (w) {
1015 w->inuse_cnt--;
1016 *w_cursor = NULL;
1020 void close_pack_index(struct packed_git *p)
1022 if (p->index_data) {
1023 munmap((void *)p->index_data, p->index_size);
1024 p->index_data = NULL;
1028 static unsigned int get_max_fd_limit(void)
1030 #ifdef RLIMIT_NOFILE
1032 struct rlimit lim;
1034 if (!getrlimit(RLIMIT_NOFILE, &lim))
1035 return lim.rlim_cur;
1037 #endif
1039 #ifdef _SC_OPEN_MAX
1041 long open_max = sysconf(_SC_OPEN_MAX);
1042 if (0 < open_max)
1043 return open_max;
1045 * Otherwise, we got -1 for one of the two
1046 * reasons:
1048 * (1) sysconf() did not understand _SC_OPEN_MAX
1049 * and signaled an error with -1; or
1050 * (2) sysconf() said there is no limit.
1052 * We _could_ clear errno before calling sysconf() to
1053 * tell these two cases apart and return a huge number
1054 * in the latter case to let the caller cap it to a
1055 * value that is not so selfish, but letting the
1056 * fallback OPEN_MAX codepath take care of these cases
1057 * is a lot simpler.
1060 #endif
1062 #ifdef OPEN_MAX
1063 return OPEN_MAX;
1064 #else
1065 return 1; /* see the caller ;-) */
1066 #endif
1070 * Do not call this directly as this leaks p->pack_fd on error return;
1071 * call open_packed_git() instead.
1073 static int open_packed_git_1(struct packed_git *p)
1075 struct stat st;
1076 struct pack_header hdr;
1077 unsigned char sha1[20];
1078 unsigned char *idx_sha1;
1079 long fd_flag;
1081 if (!p->index_data && open_pack_index(p))
1082 return error("packfile %s index unavailable", p->pack_name);
1084 if (!pack_max_fds) {
1085 unsigned int max_fds = get_max_fd_limit();
1087 /* Save 3 for stdin/stdout/stderr, 22 for work */
1088 if (25 < max_fds)
1089 pack_max_fds = max_fds - 25;
1090 else
1091 pack_max_fds = 1;
1094 while (pack_max_fds <= pack_open_fds && close_one_pack())
1095 ; /* nothing */
1097 p->pack_fd = git_open(p->pack_name);
1098 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1099 return -1;
1100 pack_open_fds++;
1102 /* If we created the struct before we had the pack we lack size. */
1103 if (!p->pack_size) {
1104 if (!S_ISREG(st.st_mode))
1105 return error("packfile %s not a regular file", p->pack_name);
1106 p->pack_size = st.st_size;
1107 } else if (p->pack_size != st.st_size)
1108 return error("packfile %s size changed", p->pack_name);
1110 /* We leave these file descriptors open with sliding mmap;
1111 * there is no point keeping them open across exec(), though.
1113 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1114 if (fd_flag < 0)
1115 return error("cannot determine file descriptor flags");
1116 fd_flag |= FD_CLOEXEC;
1117 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1118 return error("cannot set FD_CLOEXEC");
1120 /* Verify we recognize this pack file format. */
1121 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1122 return error("file %s is far too short to be a packfile", p->pack_name);
1123 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1124 return error("file %s is not a GIT packfile", p->pack_name);
1125 if (!pack_version_ok(hdr.hdr_version))
1126 return error("packfile %s is version %"PRIu32" and not"
1127 " supported (try upgrading GIT to a newer version)",
1128 p->pack_name, ntohl(hdr.hdr_version));
1130 /* Verify the pack matches its index. */
1131 if (p->num_objects != ntohl(hdr.hdr_entries))
1132 return error("packfile %s claims to have %"PRIu32" objects"
1133 " while index indicates %"PRIu32" objects",
1134 p->pack_name, ntohl(hdr.hdr_entries),
1135 p->num_objects);
1136 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1137 return error("end of packfile %s is unavailable", p->pack_name);
1138 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1139 return error("packfile %s signature is unavailable", p->pack_name);
1140 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1141 if (hashcmp(sha1, idx_sha1))
1142 return error("packfile %s does not match index", p->pack_name);
1143 return 0;
1146 static int open_packed_git(struct packed_git *p)
1148 if (!open_packed_git_1(p))
1149 return 0;
1150 close_pack_fd(p);
1151 return -1;
1154 static int in_window(struct pack_window *win, off_t offset)
1156 /* We must promise at least 20 bytes (one hash) after the
1157 * offset is available from this window, otherwise the offset
1158 * is not actually in this window and a different window (which
1159 * has that one hash excess) must be used. This is to support
1160 * the object header and delta base parsing routines below.
1162 off_t win_off = win->offset;
1163 return win_off <= offset
1164 && (offset + 20) <= (win_off + win->len);
1167 unsigned char *use_pack(struct packed_git *p,
1168 struct pack_window **w_cursor,
1169 off_t offset,
1170 unsigned long *left)
1172 struct pack_window *win = *w_cursor;
1174 /* Since packfiles end in a hash of their content and it's
1175 * pointless to ask for an offset into the middle of that
1176 * hash, and the in_window function above wouldn't match
1177 * don't allow an offset too close to the end of the file.
1179 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1180 die("packfile %s cannot be accessed", p->pack_name);
1181 if (offset > (p->pack_size - 20))
1182 die("offset beyond end of packfile (truncated pack?)");
1183 if (offset < 0)
1184 die(_("offset before end of packfile (broken .idx?)"));
1186 if (!win || !in_window(win, offset)) {
1187 if (win)
1188 win->inuse_cnt--;
1189 for (win = p->windows; win; win = win->next) {
1190 if (in_window(win, offset))
1191 break;
1193 if (!win) {
1194 size_t window_align = packed_git_window_size / 2;
1195 off_t len;
1197 if (p->pack_fd == -1 && open_packed_git(p))
1198 die("packfile %s cannot be accessed", p->pack_name);
1200 win = xcalloc(1, sizeof(*win));
1201 win->offset = (offset / window_align) * window_align;
1202 len = p->pack_size - win->offset;
1203 if (len > packed_git_window_size)
1204 len = packed_git_window_size;
1205 win->len = (size_t)len;
1206 pack_mapped += win->len;
1207 while (packed_git_limit < pack_mapped
1208 && unuse_one_window(p))
1209 ; /* nothing */
1210 win->base = xmmap(NULL, win->len,
1211 PROT_READ, MAP_PRIVATE,
1212 p->pack_fd, win->offset);
1213 if (win->base == MAP_FAILED)
1214 die_errno("packfile %s cannot be mapped",
1215 p->pack_name);
1216 if (!win->offset && win->len == p->pack_size
1217 && !p->do_not_close)
1218 close_pack_fd(p);
1219 pack_mmap_calls++;
1220 pack_open_windows++;
1221 if (pack_mapped > peak_pack_mapped)
1222 peak_pack_mapped = pack_mapped;
1223 if (pack_open_windows > peak_pack_open_windows)
1224 peak_pack_open_windows = pack_open_windows;
1225 win->next = p->windows;
1226 p->windows = win;
1229 if (win != *w_cursor) {
1230 win->last_used = pack_used_ctr++;
1231 win->inuse_cnt++;
1232 *w_cursor = win;
1234 offset -= win->offset;
1235 if (left)
1236 *left = win->len - xsize_t(offset);
1237 return win->base + offset;
1240 static struct packed_git *alloc_packed_git(int extra)
1242 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1243 memset(p, 0, sizeof(*p));
1244 p->pack_fd = -1;
1245 return p;
1248 static void try_to_free_pack_memory(size_t size)
1250 release_pack_memory(size);
1253 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1255 static int have_set_try_to_free_routine;
1256 struct stat st;
1257 size_t alloc;
1258 struct packed_git *p;
1260 if (!have_set_try_to_free_routine) {
1261 have_set_try_to_free_routine = 1;
1262 set_try_to_free_routine(try_to_free_pack_memory);
1266 * Make sure a corresponding .pack file exists and that
1267 * the index looks sane.
1269 if (!strip_suffix_mem(path, &path_len, ".idx"))
1270 return NULL;
1273 * ".pack" is long enough to hold any suffix we're adding (and
1274 * the use xsnprintf double-checks that)
1276 alloc = st_add3(path_len, strlen(".pack"), 1);
1277 p = alloc_packed_git(alloc);
1278 memcpy(p->pack_name, path, path_len);
1280 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1281 if (!access(p->pack_name, F_OK))
1282 p->pack_keep = 1;
1284 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1285 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1286 free(p);
1287 return NULL;
1290 /* ok, it looks sane as far as we can check without
1291 * actually mapping the pack file.
1293 p->pack_size = st.st_size;
1294 p->pack_local = local;
1295 p->mtime = st.st_mtime;
1296 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1297 hashclr(p->sha1);
1298 return p;
1301 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1303 const char *path = sha1_pack_name(sha1);
1304 size_t alloc = st_add(strlen(path), 1);
1305 struct packed_git *p = alloc_packed_git(alloc);
1307 memcpy(p->pack_name, path, alloc); /* includes NUL */
1308 hashcpy(p->sha1, sha1);
1309 if (check_packed_git_idx(idx_path, p)) {
1310 free(p);
1311 return NULL;
1314 return p;
1317 void install_packed_git(struct packed_git *pack)
1319 if (pack->pack_fd != -1)
1320 pack_open_fds++;
1322 pack->next = packed_git;
1323 packed_git = pack;
1326 void (*report_garbage)(unsigned seen_bits, const char *path);
1328 static void report_helper(const struct string_list *list,
1329 int seen_bits, int first, int last)
1331 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1332 return;
1334 for (; first < last; first++)
1335 report_garbage(seen_bits, list->items[first].string);
1338 static void report_pack_garbage(struct string_list *list)
1340 int i, baselen = -1, first = 0, seen_bits = 0;
1342 if (!report_garbage)
1343 return;
1345 string_list_sort(list);
1347 for (i = 0; i < list->nr; i++) {
1348 const char *path = list->items[i].string;
1349 if (baselen != -1 &&
1350 strncmp(path, list->items[first].string, baselen)) {
1351 report_helper(list, seen_bits, first, i);
1352 baselen = -1;
1353 seen_bits = 0;
1355 if (baselen == -1) {
1356 const char *dot = strrchr(path, '.');
1357 if (!dot) {
1358 report_garbage(PACKDIR_FILE_GARBAGE, path);
1359 continue;
1361 baselen = dot - path + 1;
1362 first = i;
1364 if (!strcmp(path + baselen, "pack"))
1365 seen_bits |= 1;
1366 else if (!strcmp(path + baselen, "idx"))
1367 seen_bits |= 2;
1369 report_helper(list, seen_bits, first, list->nr);
1372 static void prepare_packed_git_one(char *objdir, int local)
1374 struct strbuf path = STRBUF_INIT;
1375 size_t dirnamelen;
1376 DIR *dir;
1377 struct dirent *de;
1378 struct string_list garbage = STRING_LIST_INIT_DUP;
1380 strbuf_addstr(&path, objdir);
1381 strbuf_addstr(&path, "/pack");
1382 dir = opendir(path.buf);
1383 if (!dir) {
1384 if (errno != ENOENT)
1385 error_errno("unable to open object pack directory: %s",
1386 path.buf);
1387 strbuf_release(&path);
1388 return;
1390 strbuf_addch(&path, '/');
1391 dirnamelen = path.len;
1392 while ((de = readdir(dir)) != NULL) {
1393 struct packed_git *p;
1394 size_t base_len;
1396 if (is_dot_or_dotdot(de->d_name))
1397 continue;
1399 strbuf_setlen(&path, dirnamelen);
1400 strbuf_addstr(&path, de->d_name);
1402 base_len = path.len;
1403 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1404 /* Don't reopen a pack we already have. */
1405 for (p = packed_git; p; p = p->next) {
1406 size_t len;
1407 if (strip_suffix(p->pack_name, ".pack", &len) &&
1408 len == base_len &&
1409 !memcmp(p->pack_name, path.buf, len))
1410 break;
1412 if (p == NULL &&
1414 * See if it really is a valid .idx file with
1415 * corresponding .pack file that we can map.
1417 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1418 install_packed_git(p);
1421 if (!report_garbage)
1422 continue;
1424 if (ends_with(de->d_name, ".idx") ||
1425 ends_with(de->d_name, ".pack") ||
1426 ends_with(de->d_name, ".bitmap") ||
1427 ends_with(de->d_name, ".keep"))
1428 string_list_append(&garbage, path.buf);
1429 else
1430 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1432 closedir(dir);
1433 report_pack_garbage(&garbage);
1434 string_list_clear(&garbage, 0);
1435 strbuf_release(&path);
1438 static int approximate_object_count_valid;
1441 * Give a fast, rough count of the number of objects in the repository. This
1442 * ignores loose objects completely. If you have a lot of them, then either
1443 * you should repack because your performance will be awful, or they are
1444 * all unreachable objects about to be pruned, in which case they're not really
1445 * interesting as a measure of repo size in the first place.
1447 unsigned long approximate_object_count(void)
1449 static unsigned long count;
1450 if (!approximate_object_count_valid) {
1451 struct packed_git *p;
1453 prepare_packed_git();
1454 count = 0;
1455 for (p = packed_git; p; p = p->next) {
1456 if (open_pack_index(p))
1457 continue;
1458 count += p->num_objects;
1461 return count;
1464 static void *get_next_packed_git(const void *p)
1466 return ((const struct packed_git *)p)->next;
1469 static void set_next_packed_git(void *p, void *next)
1471 ((struct packed_git *)p)->next = next;
1474 static int sort_pack(const void *a_, const void *b_)
1476 const struct packed_git *a = a_;
1477 const struct packed_git *b = b_;
1478 int st;
1481 * Local packs tend to contain objects specific to our
1482 * variant of the project than remote ones. In addition,
1483 * remote ones could be on a network mounted filesystem.
1484 * Favor local ones for these reasons.
1486 st = a->pack_local - b->pack_local;
1487 if (st)
1488 return -st;
1491 * Younger packs tend to contain more recent objects,
1492 * and more recent objects tend to get accessed more
1493 * often.
1495 if (a->mtime < b->mtime)
1496 return 1;
1497 else if (a->mtime == b->mtime)
1498 return 0;
1499 return -1;
1502 static void rearrange_packed_git(void)
1504 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1505 set_next_packed_git, sort_pack);
1508 static void prepare_packed_git_mru(void)
1510 struct packed_git *p;
1512 mru_clear(packed_git_mru);
1513 for (p = packed_git; p; p = p->next)
1514 mru_append(packed_git_mru, p);
1517 static int prepare_packed_git_run_once = 0;
1518 void prepare_packed_git(void)
1520 struct alternate_object_database *alt;
1522 if (prepare_packed_git_run_once)
1523 return;
1524 prepare_packed_git_one(get_object_directory(), 1);
1525 prepare_alt_odb();
1526 for (alt = alt_odb_list; alt; alt = alt->next)
1527 prepare_packed_git_one(alt->path, 0);
1528 rearrange_packed_git();
1529 prepare_packed_git_mru();
1530 prepare_packed_git_run_once = 1;
1533 void reprepare_packed_git(void)
1535 approximate_object_count_valid = 0;
1536 prepare_packed_git_run_once = 0;
1537 prepare_packed_git();
1540 static void mark_bad_packed_object(struct packed_git *p,
1541 const unsigned char *sha1)
1543 unsigned i;
1544 for (i = 0; i < p->num_bad_objects; i++)
1545 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1546 return;
1547 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1548 st_mult(GIT_SHA1_RAWSZ,
1549 st_add(p->num_bad_objects, 1)));
1550 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1551 p->num_bad_objects++;
1554 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1556 struct packed_git *p;
1557 unsigned i;
1559 for (p = packed_git; p; p = p->next)
1560 for (i = 0; i < p->num_bad_objects; i++)
1561 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1562 return p;
1563 return NULL;
1567 * With an in-core object data in "map", rehash it to make sure the
1568 * object name actually matches "sha1" to detect object corruption.
1569 * With "map" == NULL, try reading the object named with "sha1" using
1570 * the streaming interface and rehash it to do the same.
1572 int check_sha1_signature(const unsigned char *sha1, void *map,
1573 unsigned long size, const char *type)
1575 unsigned char real_sha1[20];
1576 enum object_type obj_type;
1577 struct git_istream *st;
1578 git_SHA_CTX c;
1579 char hdr[32];
1580 int hdrlen;
1582 if (map) {
1583 hash_sha1_file(map, size, type, real_sha1);
1584 return hashcmp(sha1, real_sha1) ? -1 : 0;
1587 st = open_istream(sha1, &obj_type, &size, NULL);
1588 if (!st)
1589 return -1;
1591 /* Generate the header */
1592 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1594 /* Sha1.. */
1595 git_SHA1_Init(&c);
1596 git_SHA1_Update(&c, hdr, hdrlen);
1597 for (;;) {
1598 char buf[1024 * 16];
1599 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1601 if (readlen < 0) {
1602 close_istream(st);
1603 return -1;
1605 if (!readlen)
1606 break;
1607 git_SHA1_Update(&c, buf, readlen);
1609 git_SHA1_Final(real_sha1, &c);
1610 close_istream(st);
1611 return hashcmp(sha1, real_sha1) ? -1 : 0;
1614 int git_open(const char *name)
1616 static int sha1_file_open_flag = O_NOATIME | O_CLOEXEC;
1618 for (;;) {
1619 int fd;
1621 errno = 0;
1622 fd = open(name, O_RDONLY | sha1_file_open_flag);
1623 if (fd >= 0)
1624 return fd;
1626 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1627 if ((sha1_file_open_flag & O_CLOEXEC) && errno == EINVAL) {
1628 sha1_file_open_flag &= ~O_CLOEXEC;
1629 continue;
1632 /* Might the failure be due to O_NOATIME? */
1633 if (errno != ENOENT && (sha1_file_open_flag & O_NOATIME)) {
1634 sha1_file_open_flag &= ~O_NOATIME;
1635 continue;
1637 return -1;
1641 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1643 struct alternate_object_database *alt;
1645 if (!lstat(sha1_file_name(sha1), st))
1646 return 0;
1648 prepare_alt_odb();
1649 errno = ENOENT;
1650 for (alt = alt_odb_list; alt; alt = alt->next) {
1651 const char *path = alt_sha1_path(alt, sha1);
1652 if (!lstat(path, st))
1653 return 0;
1656 return -1;
1659 static int open_sha1_file(const unsigned char *sha1)
1661 int fd;
1662 struct alternate_object_database *alt;
1663 int most_interesting_errno;
1665 fd = git_open(sha1_file_name(sha1));
1666 if (fd >= 0)
1667 return fd;
1668 most_interesting_errno = errno;
1670 prepare_alt_odb();
1671 for (alt = alt_odb_list; alt; alt = alt->next) {
1672 const char *path = alt_sha1_path(alt, sha1);
1673 fd = git_open(path);
1674 if (fd >= 0)
1675 return fd;
1676 if (most_interesting_errno == ENOENT)
1677 most_interesting_errno = errno;
1679 errno = most_interesting_errno;
1680 return -1;
1683 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1685 void *map;
1686 int fd;
1688 fd = open_sha1_file(sha1);
1689 map = NULL;
1690 if (fd >= 0) {
1691 struct stat st;
1693 if (!fstat(fd, &st)) {
1694 *size = xsize_t(st.st_size);
1695 if (!*size) {
1696 /* mmap() is forbidden on empty files */
1697 error("object file %s is empty", sha1_file_name(sha1));
1698 return NULL;
1700 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1702 close(fd);
1704 return map;
1707 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1708 unsigned long len, enum object_type *type, unsigned long *sizep)
1710 unsigned shift;
1711 unsigned long size, c;
1712 unsigned long used = 0;
1714 c = buf[used++];
1715 *type = (c >> 4) & 7;
1716 size = c & 15;
1717 shift = 4;
1718 while (c & 0x80) {
1719 if (len <= used || bitsizeof(long) <= shift) {
1720 error("bad object header");
1721 size = used = 0;
1722 break;
1724 c = buf[used++];
1725 size += (c & 0x7f) << shift;
1726 shift += 7;
1728 *sizep = size;
1729 return used;
1732 static int unpack_sha1_short_header(git_zstream *stream,
1733 unsigned char *map, unsigned long mapsize,
1734 void *buffer, unsigned long bufsiz)
1736 /* Get the data stream */
1737 memset(stream, 0, sizeof(*stream));
1738 stream->next_in = map;
1739 stream->avail_in = mapsize;
1740 stream->next_out = buffer;
1741 stream->avail_out = bufsiz;
1743 git_inflate_init(stream);
1744 return git_inflate(stream, 0);
1747 int unpack_sha1_header(git_zstream *stream,
1748 unsigned char *map, unsigned long mapsize,
1749 void *buffer, unsigned long bufsiz)
1751 int status = unpack_sha1_short_header(stream, map, mapsize,
1752 buffer, bufsiz);
1754 if (status < Z_OK)
1755 return status;
1757 /* Make sure we have the terminating NUL */
1758 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1759 return -1;
1760 return 0;
1763 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1764 unsigned long mapsize, void *buffer,
1765 unsigned long bufsiz, struct strbuf *header)
1767 int status;
1769 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1770 if (status < Z_OK)
1771 return -1;
1774 * Check if entire header is unpacked in the first iteration.
1776 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1777 return 0;
1780 * buffer[0..bufsiz] was not large enough. Copy the partial
1781 * result out to header, and then append the result of further
1782 * reading the stream.
1784 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1785 stream->next_out = buffer;
1786 stream->avail_out = bufsiz;
1788 do {
1789 status = git_inflate(stream, 0);
1790 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1791 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1792 return 0;
1793 stream->next_out = buffer;
1794 stream->avail_out = bufsiz;
1795 } while (status != Z_STREAM_END);
1796 return -1;
1799 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1801 int bytes = strlen(buffer) + 1;
1802 unsigned char *buf = xmallocz(size);
1803 unsigned long n;
1804 int status = Z_OK;
1806 n = stream->total_out - bytes;
1807 if (n > size)
1808 n = size;
1809 memcpy(buf, (char *) buffer + bytes, n);
1810 bytes = n;
1811 if (bytes <= size) {
1813 * The above condition must be (bytes <= size), not
1814 * (bytes < size). In other words, even though we
1815 * expect no more output and set avail_out to zero,
1816 * the input zlib stream may have bytes that express
1817 * "this concludes the stream", and we *do* want to
1818 * eat that input.
1820 * Otherwise we would not be able to test that we
1821 * consumed all the input to reach the expected size;
1822 * we also want to check that zlib tells us that all
1823 * went well with status == Z_STREAM_END at the end.
1825 stream->next_out = buf + bytes;
1826 stream->avail_out = size - bytes;
1827 while (status == Z_OK)
1828 status = git_inflate(stream, Z_FINISH);
1830 if (status == Z_STREAM_END && !stream->avail_in) {
1831 git_inflate_end(stream);
1832 return buf;
1835 if (status < 0)
1836 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1837 else if (stream->avail_in)
1838 error("garbage at end of loose object '%s'",
1839 sha1_to_hex(sha1));
1840 free(buf);
1841 return NULL;
1845 * We used to just use "sscanf()", but that's actually way
1846 * too permissive for what we want to check. So do an anal
1847 * object header parse by hand.
1849 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1850 unsigned int flags)
1852 const char *type_buf = hdr;
1853 unsigned long size;
1854 int type, type_len = 0;
1857 * The type can be of any size but is followed by
1858 * a space.
1860 for (;;) {
1861 char c = *hdr++;
1862 if (!c)
1863 return -1;
1864 if (c == ' ')
1865 break;
1866 type_len++;
1869 type = type_from_string_gently(type_buf, type_len, 1);
1870 if (oi->typename)
1871 strbuf_add(oi->typename, type_buf, type_len);
1873 * Set type to 0 if its an unknown object and
1874 * we're obtaining the type using '--allow-unknown-type'
1875 * option.
1877 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1878 type = 0;
1879 else if (type < 0)
1880 die("invalid object type");
1881 if (oi->typep)
1882 *oi->typep = type;
1885 * The length must follow immediately, and be in canonical
1886 * decimal format (ie "010" is not valid).
1888 size = *hdr++ - '0';
1889 if (size > 9)
1890 return -1;
1891 if (size) {
1892 for (;;) {
1893 unsigned long c = *hdr - '0';
1894 if (c > 9)
1895 break;
1896 hdr++;
1897 size = size * 10 + c;
1901 if (oi->sizep)
1902 *oi->sizep = size;
1905 * The length must be followed by a zero byte
1907 return *hdr ? -1 : type;
1910 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1912 struct object_info oi = OBJECT_INFO_INIT;
1914 oi.sizep = sizep;
1915 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1918 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1920 int ret;
1921 git_zstream stream;
1922 char hdr[8192];
1924 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1925 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1926 return NULL;
1928 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1931 unsigned long get_size_from_delta(struct packed_git *p,
1932 struct pack_window **w_curs,
1933 off_t curpos)
1935 const unsigned char *data;
1936 unsigned char delta_head[20], *in;
1937 git_zstream stream;
1938 int st;
1940 memset(&stream, 0, sizeof(stream));
1941 stream.next_out = delta_head;
1942 stream.avail_out = sizeof(delta_head);
1944 git_inflate_init(&stream);
1945 do {
1946 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1947 stream.next_in = in;
1948 st = git_inflate(&stream, Z_FINISH);
1949 curpos += stream.next_in - in;
1950 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1951 stream.total_out < sizeof(delta_head));
1952 git_inflate_end(&stream);
1953 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1954 error("delta data unpack-initial failed");
1955 return 0;
1958 /* Examine the initial part of the delta to figure out
1959 * the result size.
1961 data = delta_head;
1963 /* ignore base size */
1964 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1966 /* Read the result size */
1967 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1970 static off_t get_delta_base(struct packed_git *p,
1971 struct pack_window **w_curs,
1972 off_t *curpos,
1973 enum object_type type,
1974 off_t delta_obj_offset)
1976 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1977 off_t base_offset;
1979 /* use_pack() assured us we have [base_info, base_info + 20)
1980 * as a range that we can look at without walking off the
1981 * end of the mapped window. Its actually the hash size
1982 * that is assured. An OFS_DELTA longer than the hash size
1983 * is stupid, as then a REF_DELTA would be smaller to store.
1985 if (type == OBJ_OFS_DELTA) {
1986 unsigned used = 0;
1987 unsigned char c = base_info[used++];
1988 base_offset = c & 127;
1989 while (c & 128) {
1990 base_offset += 1;
1991 if (!base_offset || MSB(base_offset, 7))
1992 return 0; /* overflow */
1993 c = base_info[used++];
1994 base_offset = (base_offset << 7) + (c & 127);
1996 base_offset = delta_obj_offset - base_offset;
1997 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1998 return 0; /* out of bound */
1999 *curpos += used;
2000 } else if (type == OBJ_REF_DELTA) {
2001 /* The base entry _must_ be in the same pack */
2002 base_offset = find_pack_entry_one(base_info, p);
2003 *curpos += 20;
2004 } else
2005 die("I am totally screwed");
2006 return base_offset;
2010 * Like get_delta_base above, but we return the sha1 instead of the pack
2011 * offset. This means it is cheaper for REF deltas (we do not have to do
2012 * the final object lookup), but more expensive for OFS deltas (we
2013 * have to load the revidx to convert the offset back into a sha1).
2015 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2016 struct pack_window **w_curs,
2017 off_t curpos,
2018 enum object_type type,
2019 off_t delta_obj_offset)
2021 if (type == OBJ_REF_DELTA) {
2022 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2023 return base;
2024 } else if (type == OBJ_OFS_DELTA) {
2025 struct revindex_entry *revidx;
2026 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2027 type, delta_obj_offset);
2029 if (!base_offset)
2030 return NULL;
2032 revidx = find_pack_revindex(p, base_offset);
2033 if (!revidx)
2034 return NULL;
2036 return nth_packed_object_sha1(p, revidx->nr);
2037 } else
2038 return NULL;
2041 int unpack_object_header(struct packed_git *p,
2042 struct pack_window **w_curs,
2043 off_t *curpos,
2044 unsigned long *sizep)
2046 unsigned char *base;
2047 unsigned long left;
2048 unsigned long used;
2049 enum object_type type;
2051 /* use_pack() assures us we have [base, base + 20) available
2052 * as a range that we can look at. (Its actually the hash
2053 * size that is assured.) With our object header encoding
2054 * the maximum deflated object size is 2^137, which is just
2055 * insane, so we know won't exceed what we have been given.
2057 base = use_pack(p, w_curs, *curpos, &left);
2058 used = unpack_object_header_buffer(base, left, &type, sizep);
2059 if (!used) {
2060 type = OBJ_BAD;
2061 } else
2062 *curpos += used;
2064 return type;
2067 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2069 int type;
2070 struct revindex_entry *revidx;
2071 const unsigned char *sha1;
2072 revidx = find_pack_revindex(p, obj_offset);
2073 if (!revidx)
2074 return OBJ_BAD;
2075 sha1 = nth_packed_object_sha1(p, revidx->nr);
2076 mark_bad_packed_object(p, sha1);
2077 type = sha1_object_info(sha1, NULL);
2078 if (type <= OBJ_NONE)
2079 return OBJ_BAD;
2080 return type;
2083 #define POI_STACK_PREALLOC 64
2085 static enum object_type packed_to_object_type(struct packed_git *p,
2086 off_t obj_offset,
2087 enum object_type type,
2088 struct pack_window **w_curs,
2089 off_t curpos)
2091 off_t small_poi_stack[POI_STACK_PREALLOC];
2092 off_t *poi_stack = small_poi_stack;
2093 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2095 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2096 off_t base_offset;
2097 unsigned long size;
2098 /* Push the object we're going to leave behind */
2099 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2100 poi_stack_alloc = alloc_nr(poi_stack_nr);
2101 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2102 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2103 } else {
2104 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2106 poi_stack[poi_stack_nr++] = obj_offset;
2107 /* If parsing the base offset fails, just unwind */
2108 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2109 if (!base_offset)
2110 goto unwind;
2111 curpos = obj_offset = base_offset;
2112 type = unpack_object_header(p, w_curs, &curpos, &size);
2113 if (type <= OBJ_NONE) {
2114 /* If getting the base itself fails, we first
2115 * retry the base, otherwise unwind */
2116 type = retry_bad_packed_offset(p, base_offset);
2117 if (type > OBJ_NONE)
2118 goto out;
2119 goto unwind;
2123 switch (type) {
2124 case OBJ_BAD:
2125 case OBJ_COMMIT:
2126 case OBJ_TREE:
2127 case OBJ_BLOB:
2128 case OBJ_TAG:
2129 break;
2130 default:
2131 error("unknown object type %i at offset %"PRIuMAX" in %s",
2132 type, (uintmax_t)obj_offset, p->pack_name);
2133 type = OBJ_BAD;
2136 out:
2137 if (poi_stack != small_poi_stack)
2138 free(poi_stack);
2139 return type;
2141 unwind:
2142 while (poi_stack_nr) {
2143 obj_offset = poi_stack[--poi_stack_nr];
2144 type = retry_bad_packed_offset(p, obj_offset);
2145 if (type > OBJ_NONE)
2146 goto out;
2148 type = OBJ_BAD;
2149 goto out;
2152 int packed_object_info(struct packed_git *p, off_t obj_offset,
2153 struct object_info *oi)
2155 struct pack_window *w_curs = NULL;
2156 unsigned long size;
2157 off_t curpos = obj_offset;
2158 enum object_type type;
2161 * We always get the representation type, but only convert it to
2162 * a "real" type later if the caller is interested.
2164 type = unpack_object_header(p, &w_curs, &curpos, &size);
2166 if (oi->sizep) {
2167 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2168 off_t tmp_pos = curpos;
2169 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2170 type, obj_offset);
2171 if (!base_offset) {
2172 type = OBJ_BAD;
2173 goto out;
2175 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2176 if (*oi->sizep == 0) {
2177 type = OBJ_BAD;
2178 goto out;
2180 } else {
2181 *oi->sizep = size;
2185 if (oi->disk_sizep) {
2186 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2187 *oi->disk_sizep = revidx[1].offset - obj_offset;
2190 if (oi->typep) {
2191 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2192 if (*oi->typep < 0) {
2193 type = OBJ_BAD;
2194 goto out;
2198 if (oi->delta_base_sha1) {
2199 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2200 const unsigned char *base;
2202 base = get_delta_base_sha1(p, &w_curs, curpos,
2203 type, obj_offset);
2204 if (!base) {
2205 type = OBJ_BAD;
2206 goto out;
2209 hashcpy(oi->delta_base_sha1, base);
2210 } else
2211 hashclr(oi->delta_base_sha1);
2214 out:
2215 unuse_pack(&w_curs);
2216 return type;
2219 static void *unpack_compressed_entry(struct packed_git *p,
2220 struct pack_window **w_curs,
2221 off_t curpos,
2222 unsigned long size)
2224 int st;
2225 git_zstream stream;
2226 unsigned char *buffer, *in;
2228 buffer = xmallocz_gently(size);
2229 if (!buffer)
2230 return NULL;
2231 memset(&stream, 0, sizeof(stream));
2232 stream.next_out = buffer;
2233 stream.avail_out = size + 1;
2235 git_inflate_init(&stream);
2236 do {
2237 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2238 stream.next_in = in;
2239 st = git_inflate(&stream, Z_FINISH);
2240 if (!stream.avail_out)
2241 break; /* the payload is larger than it should be */
2242 curpos += stream.next_in - in;
2243 } while (st == Z_OK || st == Z_BUF_ERROR);
2244 git_inflate_end(&stream);
2245 if ((st != Z_STREAM_END) || stream.total_out != size) {
2246 free(buffer);
2247 return NULL;
2250 return buffer;
2253 static struct hashmap delta_base_cache;
2254 static size_t delta_base_cached;
2256 static LIST_HEAD(delta_base_cache_lru);
2258 struct delta_base_cache_key {
2259 struct packed_git *p;
2260 off_t base_offset;
2263 struct delta_base_cache_entry {
2264 struct hashmap hash;
2265 struct delta_base_cache_key key;
2266 struct list_head lru;
2267 void *data;
2268 unsigned long size;
2269 enum object_type type;
2272 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2274 unsigned int hash;
2276 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2277 hash += (hash >> 8) + (hash >> 16);
2278 return hash;
2281 static struct delta_base_cache_entry *
2282 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2284 struct hashmap_entry entry;
2285 struct delta_base_cache_key key;
2287 if (!delta_base_cache.cmpfn)
2288 return NULL;
2290 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2291 key.p = p;
2292 key.base_offset = base_offset;
2293 return hashmap_get(&delta_base_cache, &entry, &key);
2296 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2297 const struct delta_base_cache_key *b)
2299 return a->p == b->p && a->base_offset == b->base_offset;
2302 static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2303 const void *vkey)
2305 const struct delta_base_cache_entry *a = va, *b = vb;
2306 const struct delta_base_cache_key *key = vkey;
2307 if (key)
2308 return !delta_base_cache_key_eq(&a->key, key);
2309 else
2310 return !delta_base_cache_key_eq(&a->key, &b->key);
2313 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2315 return !!get_delta_base_cache_entry(p, base_offset);
2319 * Remove the entry from the cache, but do _not_ free the associated
2320 * entry data. The caller takes ownership of the "data" buffer, and
2321 * should copy out any fields it wants before detaching.
2323 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2325 hashmap_remove(&delta_base_cache, ent, &ent->key);
2326 list_del(&ent->lru);
2327 delta_base_cached -= ent->size;
2328 free(ent);
2331 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2332 unsigned long *base_size, enum object_type *type)
2334 struct delta_base_cache_entry *ent;
2336 ent = get_delta_base_cache_entry(p, base_offset);
2337 if (!ent)
2338 return unpack_entry(p, base_offset, type, base_size);
2340 *type = ent->type;
2341 *base_size = ent->size;
2342 return xmemdupz(ent->data, ent->size);
2345 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2347 free(ent->data);
2348 detach_delta_base_cache_entry(ent);
2351 void clear_delta_base_cache(void)
2353 struct hashmap_iter iter;
2354 struct delta_base_cache_entry *entry;
2355 for (entry = hashmap_iter_first(&delta_base_cache, &iter);
2356 entry;
2357 entry = hashmap_iter_next(&iter)) {
2358 release_delta_base_cache(entry);
2362 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2363 void *base, unsigned long base_size, enum object_type type)
2365 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2366 struct list_head *lru, *tmp;
2368 delta_base_cached += base_size;
2370 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2371 struct delta_base_cache_entry *f =
2372 list_entry(lru, struct delta_base_cache_entry, lru);
2373 if (delta_base_cached <= delta_base_cache_limit)
2374 break;
2375 release_delta_base_cache(f);
2378 ent->key.p = p;
2379 ent->key.base_offset = base_offset;
2380 ent->type = type;
2381 ent->data = base;
2382 ent->size = base_size;
2383 list_add_tail(&ent->lru, &delta_base_cache_lru);
2385 if (!delta_base_cache.cmpfn)
2386 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2387 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2388 hashmap_add(&delta_base_cache, ent);
2391 static void *read_object(const unsigned char *sha1, enum object_type *type,
2392 unsigned long *size);
2394 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2396 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2397 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2398 p->pack_name, (uintmax_t)obj_offset);
2401 int do_check_packed_object_crc;
2403 #define UNPACK_ENTRY_STACK_PREALLOC 64
2404 struct unpack_entry_stack_ent {
2405 off_t obj_offset;
2406 off_t curpos;
2407 unsigned long size;
2410 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2411 enum object_type *final_type, unsigned long *final_size)
2413 struct pack_window *w_curs = NULL;
2414 off_t curpos = obj_offset;
2415 void *data = NULL;
2416 unsigned long size;
2417 enum object_type type;
2418 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2419 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2420 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2421 int base_from_cache = 0;
2423 write_pack_access_log(p, obj_offset);
2425 /* PHASE 1: drill down to the innermost base object */
2426 for (;;) {
2427 off_t base_offset;
2428 int i;
2429 struct delta_base_cache_entry *ent;
2431 ent = get_delta_base_cache_entry(p, curpos);
2432 if (ent) {
2433 type = ent->type;
2434 data = ent->data;
2435 size = ent->size;
2436 detach_delta_base_cache_entry(ent);
2437 base_from_cache = 1;
2438 break;
2441 if (do_check_packed_object_crc && p->index_version > 1) {
2442 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2443 off_t len = revidx[1].offset - obj_offset;
2444 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2445 const unsigned char *sha1 =
2446 nth_packed_object_sha1(p, revidx->nr);
2447 error("bad packed object CRC for %s",
2448 sha1_to_hex(sha1));
2449 mark_bad_packed_object(p, sha1);
2450 unuse_pack(&w_curs);
2451 return NULL;
2455 type = unpack_object_header(p, &w_curs, &curpos, &size);
2456 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2457 break;
2459 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2460 if (!base_offset) {
2461 error("failed to validate delta base reference "
2462 "at offset %"PRIuMAX" from %s",
2463 (uintmax_t)curpos, p->pack_name);
2464 /* bail to phase 2, in hopes of recovery */
2465 data = NULL;
2466 break;
2469 /* push object, proceed to base */
2470 if (delta_stack_nr >= delta_stack_alloc
2471 && delta_stack == small_delta_stack) {
2472 delta_stack_alloc = alloc_nr(delta_stack_nr);
2473 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2474 memcpy(delta_stack, small_delta_stack,
2475 sizeof(*delta_stack)*delta_stack_nr);
2476 } else {
2477 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2479 i = delta_stack_nr++;
2480 delta_stack[i].obj_offset = obj_offset;
2481 delta_stack[i].curpos = curpos;
2482 delta_stack[i].size = size;
2484 curpos = obj_offset = base_offset;
2487 /* PHASE 2: handle the base */
2488 switch (type) {
2489 case OBJ_OFS_DELTA:
2490 case OBJ_REF_DELTA:
2491 if (data)
2492 die("BUG: unpack_entry: left loop at a valid delta");
2493 break;
2494 case OBJ_COMMIT:
2495 case OBJ_TREE:
2496 case OBJ_BLOB:
2497 case OBJ_TAG:
2498 if (!base_from_cache)
2499 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2500 break;
2501 default:
2502 data = NULL;
2503 error("unknown object type %i at offset %"PRIuMAX" in %s",
2504 type, (uintmax_t)obj_offset, p->pack_name);
2507 /* PHASE 3: apply deltas in order */
2509 /* invariants:
2510 * 'data' holds the base data, or NULL if there was corruption
2512 while (delta_stack_nr) {
2513 void *delta_data;
2514 void *base = data;
2515 unsigned long delta_size, base_size = size;
2516 int i;
2518 data = NULL;
2520 if (base)
2521 add_delta_base_cache(p, obj_offset, base, base_size, type);
2523 if (!base) {
2525 * We're probably in deep shit, but let's try to fetch
2526 * the required base anyway from another pack or loose.
2527 * This is costly but should happen only in the presence
2528 * of a corrupted pack, and is better than failing outright.
2530 struct revindex_entry *revidx;
2531 const unsigned char *base_sha1;
2532 revidx = find_pack_revindex(p, obj_offset);
2533 if (revidx) {
2534 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2535 error("failed to read delta base object %s"
2536 " at offset %"PRIuMAX" from %s",
2537 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2538 p->pack_name);
2539 mark_bad_packed_object(p, base_sha1);
2540 base = read_object(base_sha1, &type, &base_size);
2544 i = --delta_stack_nr;
2545 obj_offset = delta_stack[i].obj_offset;
2546 curpos = delta_stack[i].curpos;
2547 delta_size = delta_stack[i].size;
2549 if (!base)
2550 continue;
2552 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2554 if (!delta_data) {
2555 error("failed to unpack compressed delta "
2556 "at offset %"PRIuMAX" from %s",
2557 (uintmax_t)curpos, p->pack_name);
2558 data = NULL;
2559 continue;
2562 data = patch_delta(base, base_size,
2563 delta_data, delta_size,
2564 &size);
2567 * We could not apply the delta; warn the user, but keep going.
2568 * Our failure will be noticed either in the next iteration of
2569 * the loop, or if this is the final delta, in the caller when
2570 * we return NULL. Those code paths will take care of making
2571 * a more explicit warning and retrying with another copy of
2572 * the object.
2574 if (!data)
2575 error("failed to apply delta");
2577 free(delta_data);
2580 *final_type = type;
2581 *final_size = size;
2583 unuse_pack(&w_curs);
2585 if (delta_stack != small_delta_stack)
2586 free(delta_stack);
2588 return data;
2591 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2592 uint32_t n)
2594 const unsigned char *index = p->index_data;
2595 if (!index) {
2596 if (open_pack_index(p))
2597 return NULL;
2598 index = p->index_data;
2600 if (n >= p->num_objects)
2601 return NULL;
2602 index += 4 * 256;
2603 if (p->index_version == 1) {
2604 return index + 24 * n + 4;
2605 } else {
2606 index += 8;
2607 return index + 20 * n;
2611 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2613 const unsigned char *ptr = vptr;
2614 const unsigned char *start = p->index_data;
2615 const unsigned char *end = start + p->index_size;
2616 if (ptr < start)
2617 die(_("offset before start of pack index for %s (corrupt index?)"),
2618 p->pack_name);
2619 /* No need to check for underflow; .idx files must be at least 8 bytes */
2620 if (ptr >= end - 8)
2621 die(_("offset beyond end of pack index for %s (truncated index?)"),
2622 p->pack_name);
2625 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2627 const unsigned char *index = p->index_data;
2628 index += 4 * 256;
2629 if (p->index_version == 1) {
2630 return ntohl(*((uint32_t *)(index + 24 * n)));
2631 } else {
2632 uint32_t off;
2633 index += 8 + p->num_objects * (20 + 4);
2634 off = ntohl(*((uint32_t *)(index + 4 * n)));
2635 if (!(off & 0x80000000))
2636 return off;
2637 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2638 check_pack_index_ptr(p, index);
2639 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2640 ntohl(*((uint32_t *)(index + 4)));
2644 off_t find_pack_entry_one(const unsigned char *sha1,
2645 struct packed_git *p)
2647 const uint32_t *level1_ofs = p->index_data;
2648 const unsigned char *index = p->index_data;
2649 unsigned hi, lo, stride;
2650 static int use_lookup = -1;
2651 static int debug_lookup = -1;
2653 if (debug_lookup < 0)
2654 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2656 if (!index) {
2657 if (open_pack_index(p))
2658 return 0;
2659 level1_ofs = p->index_data;
2660 index = p->index_data;
2662 if (p->index_version > 1) {
2663 level1_ofs += 2;
2664 index += 8;
2666 index += 4 * 256;
2667 hi = ntohl(level1_ofs[*sha1]);
2668 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2669 if (p->index_version > 1) {
2670 stride = 20;
2671 } else {
2672 stride = 24;
2673 index += 4;
2676 if (debug_lookup)
2677 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2678 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2680 if (use_lookup < 0)
2681 use_lookup = !!getenv("GIT_USE_LOOKUP");
2682 if (use_lookup) {
2683 int pos = sha1_entry_pos(index, stride, 0,
2684 lo, hi, p->num_objects, sha1);
2685 if (pos < 0)
2686 return 0;
2687 return nth_packed_object_offset(p, pos);
2690 do {
2691 unsigned mi = (lo + hi) / 2;
2692 int cmp = hashcmp(index + mi * stride, sha1);
2694 if (debug_lookup)
2695 printf("lo %u hi %u rg %u mi %u\n",
2696 lo, hi, hi - lo, mi);
2697 if (!cmp)
2698 return nth_packed_object_offset(p, mi);
2699 if (cmp > 0)
2700 hi = mi;
2701 else
2702 lo = mi+1;
2703 } while (lo < hi);
2704 return 0;
2707 int is_pack_valid(struct packed_git *p)
2709 /* An already open pack is known to be valid. */
2710 if (p->pack_fd != -1)
2711 return 1;
2713 /* If the pack has one window completely covering the
2714 * file size, the pack is known to be valid even if
2715 * the descriptor is not currently open.
2717 if (p->windows) {
2718 struct pack_window *w = p->windows;
2720 if (!w->offset && w->len == p->pack_size)
2721 return 1;
2724 /* Force the pack to open to prove its valid. */
2725 return !open_packed_git(p);
2728 static int fill_pack_entry(const unsigned char *sha1,
2729 struct pack_entry *e,
2730 struct packed_git *p)
2732 off_t offset;
2734 if (p->num_bad_objects) {
2735 unsigned i;
2736 for (i = 0; i < p->num_bad_objects; i++)
2737 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2738 return 0;
2741 offset = find_pack_entry_one(sha1, p);
2742 if (!offset)
2743 return 0;
2746 * We are about to tell the caller where they can locate the
2747 * requested object. We better make sure the packfile is
2748 * still here and can be accessed before supplying that
2749 * answer, as it may have been deleted since the index was
2750 * loaded!
2752 if (!is_pack_valid(p))
2753 return 0;
2754 e->offset = offset;
2755 e->p = p;
2756 hashcpy(e->sha1, sha1);
2757 return 1;
2761 * Iff a pack file contains the object named by sha1, return true and
2762 * store its location to e.
2764 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2766 struct mru_entry *p;
2768 prepare_packed_git();
2769 if (!packed_git)
2770 return 0;
2772 for (p = packed_git_mru->head; p; p = p->next) {
2773 if (fill_pack_entry(sha1, e, p->item)) {
2774 mru_mark(packed_git_mru, p);
2775 return 1;
2778 return 0;
2781 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2782 struct packed_git *packs)
2784 struct packed_git *p;
2786 for (p = packs; p; p = p->next) {
2787 if (find_pack_entry_one(sha1, p))
2788 return p;
2790 return NULL;
2794 static int sha1_loose_object_info(const unsigned char *sha1,
2795 struct object_info *oi,
2796 int flags)
2798 int status = 0;
2799 unsigned long mapsize;
2800 void *map;
2801 git_zstream stream;
2802 char hdr[32];
2803 struct strbuf hdrbuf = STRBUF_INIT;
2805 if (oi->delta_base_sha1)
2806 hashclr(oi->delta_base_sha1);
2809 * If we don't care about type or size, then we don't
2810 * need to look inside the object at all. Note that we
2811 * do not optimize out the stat call, even if the
2812 * caller doesn't care about the disk-size, since our
2813 * return value implicitly indicates whether the
2814 * object even exists.
2816 if (!oi->typep && !oi->typename && !oi->sizep) {
2817 struct stat st;
2818 if (stat_sha1_file(sha1, &st) < 0)
2819 return -1;
2820 if (oi->disk_sizep)
2821 *oi->disk_sizep = st.st_size;
2822 return 0;
2825 map = map_sha1_file(sha1, &mapsize);
2826 if (!map)
2827 return -1;
2828 if (oi->disk_sizep)
2829 *oi->disk_sizep = mapsize;
2830 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2831 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2832 status = error("unable to unpack %s header with --allow-unknown-type",
2833 sha1_to_hex(sha1));
2834 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2835 status = error("unable to unpack %s header",
2836 sha1_to_hex(sha1));
2837 if (status < 0)
2838 ; /* Do nothing */
2839 else if (hdrbuf.len) {
2840 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2841 status = error("unable to parse %s header with --allow-unknown-type",
2842 sha1_to_hex(sha1));
2843 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2844 status = error("unable to parse %s header", sha1_to_hex(sha1));
2845 git_inflate_end(&stream);
2846 munmap(map, mapsize);
2847 if (status && oi->typep)
2848 *oi->typep = status;
2849 strbuf_release(&hdrbuf);
2850 return 0;
2853 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2855 struct cached_object *co;
2856 struct pack_entry e;
2857 int rtype;
2858 enum object_type real_type;
2859 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2861 co = find_cached_object(real);
2862 if (co) {
2863 if (oi->typep)
2864 *(oi->typep) = co->type;
2865 if (oi->sizep)
2866 *(oi->sizep) = co->size;
2867 if (oi->disk_sizep)
2868 *(oi->disk_sizep) = 0;
2869 if (oi->delta_base_sha1)
2870 hashclr(oi->delta_base_sha1);
2871 if (oi->typename)
2872 strbuf_addstr(oi->typename, typename(co->type));
2873 oi->whence = OI_CACHED;
2874 return 0;
2877 if (!find_pack_entry(real, &e)) {
2878 /* Most likely it's a loose object. */
2879 if (!sha1_loose_object_info(real, oi, flags)) {
2880 oi->whence = OI_LOOSE;
2881 return 0;
2884 /* Not a loose object; someone else may have just packed it. */
2885 reprepare_packed_git();
2886 if (!find_pack_entry(real, &e))
2887 return -1;
2891 * packed_object_info() does not follow the delta chain to
2892 * find out the real type, unless it is given oi->typep.
2894 if (oi->typename && !oi->typep)
2895 oi->typep = &real_type;
2897 rtype = packed_object_info(e.p, e.offset, oi);
2898 if (rtype < 0) {
2899 mark_bad_packed_object(e.p, real);
2900 if (oi->typep == &real_type)
2901 oi->typep = NULL;
2902 return sha1_object_info_extended(real, oi, 0);
2903 } else if (in_delta_base_cache(e.p, e.offset)) {
2904 oi->whence = OI_DBCACHED;
2905 } else {
2906 oi->whence = OI_PACKED;
2907 oi->u.packed.offset = e.offset;
2908 oi->u.packed.pack = e.p;
2909 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2910 rtype == OBJ_OFS_DELTA);
2912 if (oi->typename)
2913 strbuf_addstr(oi->typename, typename(*oi->typep));
2914 if (oi->typep == &real_type)
2915 oi->typep = NULL;
2917 return 0;
2920 /* returns enum object_type or negative */
2921 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2923 enum object_type type;
2924 struct object_info oi = OBJECT_INFO_INIT;
2926 oi.typep = &type;
2927 oi.sizep = sizep;
2928 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2929 return -1;
2930 return type;
2933 static void *read_packed_sha1(const unsigned char *sha1,
2934 enum object_type *type, unsigned long *size)
2936 struct pack_entry e;
2937 void *data;
2939 if (!find_pack_entry(sha1, &e))
2940 return NULL;
2941 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2942 if (!data) {
2944 * We're probably in deep shit, but let's try to fetch
2945 * the required object anyway from another pack or loose.
2946 * This should happen only in the presence of a corrupted
2947 * pack, and is better than failing outright.
2949 error("failed to read object %s at offset %"PRIuMAX" from %s",
2950 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2951 mark_bad_packed_object(e.p, sha1);
2952 data = read_object(sha1, type, size);
2954 return data;
2957 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2958 unsigned char *sha1)
2960 struct cached_object *co;
2962 hash_sha1_file(buf, len, typename(type), sha1);
2963 if (has_sha1_file(sha1) || find_cached_object(sha1))
2964 return 0;
2965 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2966 co = &cached_objects[cached_object_nr++];
2967 co->size = len;
2968 co->type = type;
2969 co->buf = xmalloc(len);
2970 memcpy(co->buf, buf, len);
2971 hashcpy(co->sha1, sha1);
2972 return 0;
2975 static void *read_object(const unsigned char *sha1, enum object_type *type,
2976 unsigned long *size)
2978 unsigned long mapsize;
2979 void *map, *buf;
2980 struct cached_object *co;
2982 co = find_cached_object(sha1);
2983 if (co) {
2984 *type = co->type;
2985 *size = co->size;
2986 return xmemdupz(co->buf, co->size);
2989 buf = read_packed_sha1(sha1, type, size);
2990 if (buf)
2991 return buf;
2992 map = map_sha1_file(sha1, &mapsize);
2993 if (map) {
2994 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2995 munmap(map, mapsize);
2996 return buf;
2998 reprepare_packed_git();
2999 return read_packed_sha1(sha1, type, size);
3003 * This function dies on corrupt objects; the callers who want to
3004 * deal with them should arrange to call read_object() and give error
3005 * messages themselves.
3007 void *read_sha1_file_extended(const unsigned char *sha1,
3008 enum object_type *type,
3009 unsigned long *size,
3010 unsigned flag)
3012 void *data;
3013 const struct packed_git *p;
3014 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3016 errno = 0;
3017 data = read_object(repl, type, size);
3018 if (data)
3019 return data;
3021 if (errno && errno != ENOENT)
3022 die_errno("failed to read object %s", sha1_to_hex(sha1));
3024 /* die if we replaced an object with one that does not exist */
3025 if (repl != sha1)
3026 die("replacement %s not found for %s",
3027 sha1_to_hex(repl), sha1_to_hex(sha1));
3029 if (has_loose_object(repl)) {
3030 const char *path = sha1_file_name(sha1);
3032 die("loose object %s (stored in %s) is corrupt",
3033 sha1_to_hex(repl), path);
3036 if ((p = has_packed_and_bad(repl)) != NULL)
3037 die("packed object %s (stored in %s) is corrupt",
3038 sha1_to_hex(repl), p->pack_name);
3040 return NULL;
3043 void *read_object_with_reference(const unsigned char *sha1,
3044 const char *required_type_name,
3045 unsigned long *size,
3046 unsigned char *actual_sha1_return)
3048 enum object_type type, required_type;
3049 void *buffer;
3050 unsigned long isize;
3051 unsigned char actual_sha1[20];
3053 required_type = type_from_string(required_type_name);
3054 hashcpy(actual_sha1, sha1);
3055 while (1) {
3056 int ref_length = -1;
3057 const char *ref_type = NULL;
3059 buffer = read_sha1_file(actual_sha1, &type, &isize);
3060 if (!buffer)
3061 return NULL;
3062 if (type == required_type) {
3063 *size = isize;
3064 if (actual_sha1_return)
3065 hashcpy(actual_sha1_return, actual_sha1);
3066 return buffer;
3068 /* Handle references */
3069 else if (type == OBJ_COMMIT)
3070 ref_type = "tree ";
3071 else if (type == OBJ_TAG)
3072 ref_type = "object ";
3073 else {
3074 free(buffer);
3075 return NULL;
3077 ref_length = strlen(ref_type);
3079 if (ref_length + 40 > isize ||
3080 memcmp(buffer, ref_type, ref_length) ||
3081 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3082 free(buffer);
3083 return NULL;
3085 free(buffer);
3086 /* Now we have the ID of the referred-to object in
3087 * actual_sha1. Check again. */
3091 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3092 const char *type, unsigned char *sha1,
3093 char *hdr, int *hdrlen)
3095 git_SHA_CTX c;
3097 /* Generate the header */
3098 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3100 /* Sha1.. */
3101 git_SHA1_Init(&c);
3102 git_SHA1_Update(&c, hdr, *hdrlen);
3103 git_SHA1_Update(&c, buf, len);
3104 git_SHA1_Final(sha1, &c);
3108 * Move the just written object into its final resting place.
3110 int finalize_object_file(const char *tmpfile, const char *filename)
3112 int ret = 0;
3114 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3115 goto try_rename;
3116 else if (link(tmpfile, filename))
3117 ret = errno;
3120 * Coda hack - coda doesn't like cross-directory links,
3121 * so we fall back to a rename, which will mean that it
3122 * won't be able to check collisions, but that's not a
3123 * big deal.
3125 * The same holds for FAT formatted media.
3127 * When this succeeds, we just return. We have nothing
3128 * left to unlink.
3130 if (ret && ret != EEXIST) {
3131 try_rename:
3132 if (!rename(tmpfile, filename))
3133 goto out;
3134 ret = errno;
3136 unlink_or_warn(tmpfile);
3137 if (ret) {
3138 if (ret != EEXIST) {
3139 return error_errno("unable to write sha1 filename %s", filename);
3141 /* FIXME!!! Collision check here ? */
3144 out:
3145 if (adjust_shared_perm(filename))
3146 return error("unable to set permission to '%s'", filename);
3147 return 0;
3150 static int write_buffer(int fd, const void *buf, size_t len)
3152 if (write_in_full(fd, buf, len) < 0)
3153 return error_errno("file write error");
3154 return 0;
3157 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3158 unsigned char *sha1)
3160 char hdr[32];
3161 int hdrlen = sizeof(hdr);
3162 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3163 return 0;
3166 /* Finalize a file on disk, and close it. */
3167 static void close_sha1_file(int fd)
3169 if (fsync_object_files)
3170 fsync_or_die(fd, "sha1 file");
3171 if (close(fd) != 0)
3172 die_errno("error when closing sha1 file");
3175 /* Size of directory component, including the ending '/' */
3176 static inline int directory_size(const char *filename)
3178 const char *s = strrchr(filename, '/');
3179 if (!s)
3180 return 0;
3181 return s - filename + 1;
3185 * This creates a temporary file in the same directory as the final
3186 * 'filename'
3188 * We want to avoid cross-directory filename renames, because those
3189 * can have problems on various filesystems (FAT, NFS, Coda).
3191 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3193 int fd, dirlen = directory_size(filename);
3195 strbuf_reset(tmp);
3196 strbuf_add(tmp, filename, dirlen);
3197 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3198 fd = git_mkstemp_mode(tmp->buf, 0444);
3199 if (fd < 0 && dirlen && errno == ENOENT) {
3201 * Make sure the directory exists; note that the contents
3202 * of the buffer are undefined after mkstemp returns an
3203 * error, so we have to rewrite the whole buffer from
3204 * scratch.
3206 strbuf_reset(tmp);
3207 strbuf_add(tmp, filename, dirlen - 1);
3208 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3209 return -1;
3210 if (adjust_shared_perm(tmp->buf))
3211 return -1;
3213 /* Try again */
3214 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3215 fd = git_mkstemp_mode(tmp->buf, 0444);
3217 return fd;
3220 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3221 const void *buf, unsigned long len, time_t mtime)
3223 int fd, ret;
3224 unsigned char compressed[4096];
3225 git_zstream stream;
3226 git_SHA_CTX c;
3227 unsigned char parano_sha1[20];
3228 static struct strbuf tmp_file = STRBUF_INIT;
3229 const char *filename = sha1_file_name(sha1);
3231 fd = create_tmpfile(&tmp_file, filename);
3232 if (fd < 0) {
3233 if (errno == EACCES)
3234 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3235 else
3236 return error_errno("unable to create temporary file");
3239 /* Set it up */
3240 git_deflate_init(&stream, zlib_compression_level);
3241 stream.next_out = compressed;
3242 stream.avail_out = sizeof(compressed);
3243 git_SHA1_Init(&c);
3245 /* First header.. */
3246 stream.next_in = (unsigned char *)hdr;
3247 stream.avail_in = hdrlen;
3248 while (git_deflate(&stream, 0) == Z_OK)
3249 ; /* nothing */
3250 git_SHA1_Update(&c, hdr, hdrlen);
3252 /* Then the data itself.. */
3253 stream.next_in = (void *)buf;
3254 stream.avail_in = len;
3255 do {
3256 unsigned char *in0 = stream.next_in;
3257 ret = git_deflate(&stream, Z_FINISH);
3258 git_SHA1_Update(&c, in0, stream.next_in - in0);
3259 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3260 die("unable to write sha1 file");
3261 stream.next_out = compressed;
3262 stream.avail_out = sizeof(compressed);
3263 } while (ret == Z_OK);
3265 if (ret != Z_STREAM_END)
3266 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3267 ret = git_deflate_end_gently(&stream);
3268 if (ret != Z_OK)
3269 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3270 git_SHA1_Final(parano_sha1, &c);
3271 if (hashcmp(sha1, parano_sha1) != 0)
3272 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3274 close_sha1_file(fd);
3276 if (mtime) {
3277 struct utimbuf utb;
3278 utb.actime = mtime;
3279 utb.modtime = mtime;
3280 if (utime(tmp_file.buf, &utb) < 0)
3281 warning_errno("failed utime() on %s", tmp_file.buf);
3284 return finalize_object_file(tmp_file.buf, filename);
3287 static int freshen_loose_object(const unsigned char *sha1)
3289 return check_and_freshen(sha1, 1);
3292 static int freshen_packed_object(const unsigned char *sha1)
3294 struct pack_entry e;
3295 if (!find_pack_entry(sha1, &e))
3296 return 0;
3297 if (e.p->freshened)
3298 return 1;
3299 if (!freshen_file(e.p->pack_name))
3300 return 0;
3301 e.p->freshened = 1;
3302 return 1;
3305 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3307 char hdr[32];
3308 int hdrlen = sizeof(hdr);
3310 /* Normally if we have it in the pack then we do not bother writing
3311 * it out into .git/objects/??/?{38} file.
3313 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3314 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3315 return 0;
3316 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3319 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3320 unsigned char *sha1, unsigned flags)
3322 char *header;
3323 int hdrlen, status = 0;
3325 /* type string, SP, %lu of the length plus NUL must fit this */
3326 hdrlen = strlen(type) + 32;
3327 header = xmalloc(hdrlen);
3328 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3330 if (!(flags & HASH_WRITE_OBJECT))
3331 goto cleanup;
3332 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3333 goto cleanup;
3334 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3336 cleanup:
3337 free(header);
3338 return status;
3341 int force_object_loose(const unsigned char *sha1, time_t mtime)
3343 void *buf;
3344 unsigned long len;
3345 enum object_type type;
3346 char hdr[32];
3347 int hdrlen;
3348 int ret;
3350 if (has_loose_object(sha1))
3351 return 0;
3352 buf = read_packed_sha1(sha1, &type, &len);
3353 if (!buf)
3354 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3355 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3356 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3357 free(buf);
3359 return ret;
3362 int has_pack_index(const unsigned char *sha1)
3364 struct stat st;
3365 if (stat(sha1_pack_index_name(sha1), &st))
3366 return 0;
3367 return 1;
3370 int has_sha1_pack(const unsigned char *sha1)
3372 struct pack_entry e;
3373 return find_pack_entry(sha1, &e);
3376 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3378 struct pack_entry e;
3380 if (find_pack_entry(sha1, &e))
3381 return 1;
3382 if (has_loose_object(sha1))
3383 return 1;
3384 if (flags & HAS_SHA1_QUICK)
3385 return 0;
3386 reprepare_packed_git();
3387 return find_pack_entry(sha1, &e);
3390 int has_object_file(const struct object_id *oid)
3392 return has_sha1_file(oid->hash);
3395 int has_object_file_with_flags(const struct object_id *oid, int flags)
3397 return has_sha1_file_with_flags(oid->hash, flags);
3400 static void check_tree(const void *buf, size_t size)
3402 struct tree_desc desc;
3403 struct name_entry entry;
3405 init_tree_desc(&desc, buf, size);
3406 while (tree_entry(&desc, &entry))
3407 /* do nothing
3408 * tree_entry() will die() on malformed entries */
3412 static void check_commit(const void *buf, size_t size)
3414 struct commit c;
3415 memset(&c, 0, sizeof(c));
3416 if (parse_commit_buffer(&c, buf, size))
3417 die("corrupt commit");
3420 static void check_tag(const void *buf, size_t size)
3422 struct tag t;
3423 memset(&t, 0, sizeof(t));
3424 if (parse_tag_buffer(&t, buf, size))
3425 die("corrupt tag");
3428 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3429 enum object_type type,
3430 const char *path, unsigned flags)
3432 int ret, re_allocated = 0;
3433 int write_object = flags & HASH_WRITE_OBJECT;
3435 if (!type)
3436 type = OBJ_BLOB;
3439 * Convert blobs to git internal format
3441 if ((type == OBJ_BLOB) && path) {
3442 struct strbuf nbuf = STRBUF_INIT;
3443 if (convert_to_git(path, buf, size, &nbuf,
3444 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3445 buf = strbuf_detach(&nbuf, &size);
3446 re_allocated = 1;
3449 if (flags & HASH_FORMAT_CHECK) {
3450 if (type == OBJ_TREE)
3451 check_tree(buf, size);
3452 if (type == OBJ_COMMIT)
3453 check_commit(buf, size);
3454 if (type == OBJ_TAG)
3455 check_tag(buf, size);
3458 if (write_object)
3459 ret = write_sha1_file(buf, size, typename(type), sha1);
3460 else
3461 ret = hash_sha1_file(buf, size, typename(type), sha1);
3462 if (re_allocated)
3463 free(buf);
3464 return ret;
3467 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3468 const char *path, unsigned flags)
3470 int ret;
3471 const int write_object = flags & HASH_WRITE_OBJECT;
3472 struct strbuf sbuf = STRBUF_INIT;
3474 assert(path);
3475 assert(would_convert_to_git_filter_fd(path));
3477 convert_to_git_filter_fd(path, fd, &sbuf,
3478 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3480 if (write_object)
3481 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3482 sha1);
3483 else
3484 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3485 sha1);
3486 strbuf_release(&sbuf);
3487 return ret;
3490 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3491 const char *path, unsigned flags)
3493 struct strbuf sbuf = STRBUF_INIT;
3494 int ret;
3496 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3497 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3498 else
3499 ret = -1;
3500 strbuf_release(&sbuf);
3501 return ret;
3504 #define SMALL_FILE_SIZE (32*1024)
3506 static int index_core(unsigned char *sha1, int fd, size_t size,
3507 enum object_type type, const char *path,
3508 unsigned flags)
3510 int ret;
3512 if (!size) {
3513 ret = index_mem(sha1, "", size, type, path, flags);
3514 } else if (size <= SMALL_FILE_SIZE) {
3515 char *buf = xmalloc(size);
3516 if (size == read_in_full(fd, buf, size))
3517 ret = index_mem(sha1, buf, size, type, path, flags);
3518 else
3519 ret = error_errno("short read");
3520 free(buf);
3521 } else {
3522 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3523 ret = index_mem(sha1, buf, size, type, path, flags);
3524 munmap(buf, size);
3526 return ret;
3530 * This creates one packfile per large blob unless bulk-checkin
3531 * machinery is "plugged".
3533 * This also bypasses the usual "convert-to-git" dance, and that is on
3534 * purpose. We could write a streaming version of the converting
3535 * functions and insert that before feeding the data to fast-import
3536 * (or equivalent in-core API described above). However, that is
3537 * somewhat complicated, as we do not know the size of the filter
3538 * result, which we need to know beforehand when writing a git object.
3539 * Since the primary motivation for trying to stream from the working
3540 * tree file and to avoid mmaping it in core is to deal with large
3541 * binary blobs, they generally do not want to get any conversion, and
3542 * callers should avoid this code path when filters are requested.
3544 static int index_stream(unsigned char *sha1, int fd, size_t size,
3545 enum object_type type, const char *path,
3546 unsigned flags)
3548 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3551 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3552 enum object_type type, const char *path, unsigned flags)
3554 int ret;
3557 * Call xsize_t() only when needed to avoid potentially unnecessary
3558 * die() for large files.
3560 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3561 ret = index_stream_convert_blob(sha1, fd, path, flags);
3562 else if (!S_ISREG(st->st_mode))
3563 ret = index_pipe(sha1, fd, type, path, flags);
3564 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3565 (path && would_convert_to_git(path)))
3566 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3567 flags);
3568 else
3569 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3570 flags);
3571 close(fd);
3572 return ret;
3575 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3577 int fd;
3578 struct strbuf sb = STRBUF_INIT;
3580 switch (st->st_mode & S_IFMT) {
3581 case S_IFREG:
3582 fd = open(path, O_RDONLY);
3583 if (fd < 0)
3584 return error_errno("open(\"%s\")", path);
3585 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3586 return error("%s: failed to insert into database",
3587 path);
3588 break;
3589 case S_IFLNK:
3590 if (strbuf_readlink(&sb, path, st->st_size))
3591 return error_errno("readlink(\"%s\")", path);
3592 if (!(flags & HASH_WRITE_OBJECT))
3593 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3594 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3595 return error("%s: failed to insert into database",
3596 path);
3597 strbuf_release(&sb);
3598 break;
3599 case S_IFDIR:
3600 return resolve_gitlink_ref(path, "HEAD", sha1);
3601 default:
3602 return error("%s: unsupported file type", path);
3604 return 0;
3607 int read_pack_header(int fd, struct pack_header *header)
3609 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3610 /* "eof before pack header was fully read" */
3611 return PH_ERROR_EOF;
3613 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3614 /* "protocol error (pack signature mismatch detected)" */
3615 return PH_ERROR_PACK_SIGNATURE;
3616 if (!pack_version_ok(header->hdr_version))
3617 /* "protocol error (pack version unsupported)" */
3618 return PH_ERROR_PROTOCOL;
3619 return 0;
3622 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3624 enum object_type type = sha1_object_info(sha1, NULL);
3625 if (type < 0)
3626 die("%s is not a valid object", sha1_to_hex(sha1));
3627 if (type != expect)
3628 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3629 typename(expect));
3632 static int for_each_file_in_obj_subdir(int subdir_nr,
3633 struct strbuf *path,
3634 each_loose_object_fn obj_cb,
3635 each_loose_cruft_fn cruft_cb,
3636 each_loose_subdir_fn subdir_cb,
3637 void *data)
3639 size_t baselen = path->len;
3640 DIR *dir = opendir(path->buf);
3641 struct dirent *de;
3642 int r = 0;
3644 if (!dir) {
3645 if (errno == ENOENT)
3646 return 0;
3647 return error_errno("unable to open %s", path->buf);
3650 while ((de = readdir(dir))) {
3651 if (is_dot_or_dotdot(de->d_name))
3652 continue;
3654 strbuf_setlen(path, baselen);
3655 strbuf_addf(path, "/%s", de->d_name);
3657 if (strlen(de->d_name) == 38) {
3658 char hex[41];
3659 unsigned char sha1[20];
3661 snprintf(hex, sizeof(hex), "%02x%s",
3662 subdir_nr, de->d_name);
3663 if (!get_sha1_hex(hex, sha1)) {
3664 if (obj_cb) {
3665 r = obj_cb(sha1, path->buf, data);
3666 if (r)
3667 break;
3669 continue;
3673 if (cruft_cb) {
3674 r = cruft_cb(de->d_name, path->buf, data);
3675 if (r)
3676 break;
3679 closedir(dir);
3681 strbuf_setlen(path, baselen);
3682 if (!r && subdir_cb)
3683 r = subdir_cb(subdir_nr, path->buf, data);
3685 return r;
3688 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3689 each_loose_object_fn obj_cb,
3690 each_loose_cruft_fn cruft_cb,
3691 each_loose_subdir_fn subdir_cb,
3692 void *data)
3694 size_t baselen = path->len;
3695 int r = 0;
3696 int i;
3698 for (i = 0; i < 256; i++) {
3699 strbuf_addf(path, "/%02x", i);
3700 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3701 subdir_cb, data);
3702 strbuf_setlen(path, baselen);
3703 if (r)
3704 break;
3707 return r;
3710 int for_each_loose_file_in_objdir(const char *path,
3711 each_loose_object_fn obj_cb,
3712 each_loose_cruft_fn cruft_cb,
3713 each_loose_subdir_fn subdir_cb,
3714 void *data)
3716 struct strbuf buf = STRBUF_INIT;
3717 int r;
3719 strbuf_addstr(&buf, path);
3720 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3721 subdir_cb, data);
3722 strbuf_release(&buf);
3724 return r;
3727 struct loose_alt_odb_data {
3728 each_loose_object_fn *cb;
3729 void *data;
3732 static int loose_from_alt_odb(struct alternate_object_database *alt,
3733 void *vdata)
3735 struct loose_alt_odb_data *data = vdata;
3736 struct strbuf buf = STRBUF_INIT;
3737 int r;
3739 strbuf_addstr(&buf, alt->path);
3740 r = for_each_loose_file_in_objdir_buf(&buf,
3741 data->cb, NULL, NULL,
3742 data->data);
3743 strbuf_release(&buf);
3744 return r;
3747 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3749 struct loose_alt_odb_data alt;
3750 int r;
3752 r = for_each_loose_file_in_objdir(get_object_directory(),
3753 cb, NULL, NULL, data);
3754 if (r)
3755 return r;
3757 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3758 return 0;
3760 alt.cb = cb;
3761 alt.data = data;
3762 return foreach_alt_odb(loose_from_alt_odb, &alt);
3765 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3767 uint32_t i;
3768 int r = 0;
3770 for (i = 0; i < p->num_objects; i++) {
3771 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3773 if (!sha1)
3774 return error("unable to get sha1 of object %u in %s",
3775 i, p->pack_name);
3777 r = cb(sha1, p, i, data);
3778 if (r)
3779 break;
3781 return r;
3784 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3786 struct packed_git *p;
3787 int r = 0;
3788 int pack_errors = 0;
3790 prepare_packed_git();
3791 for (p = packed_git; p; p = p->next) {
3792 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3793 continue;
3794 if (open_pack_index(p)) {
3795 pack_errors = 1;
3796 continue;
3798 r = for_each_object_in_pack(p, cb, data);
3799 if (r)
3800 break;
3802 return r ? r : pack_errors;