1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9#include "cache.h"
10#include "config.h"
11#include "string-list.h"
12#include "lockfile.h"
13#include "delta.h"
14#include "pack.h"
15#include "blob.h"
16#include "commit.h"
17#include "run-command.h"
18#include "tag.h"
19#include "tree.h"
20#include "tree-walk.h"
21#include "refs.h"
22#include "pack-revindex.h"
23#include "sha1-lookup.h"
24#include "bulk-checkin.h"
25#include "streaming.h"
26#include "dir.h"
27#include "list.h"
28#include "mergesort.h"
29#include "quote.h"
30#include "packfile.h"
31
32const unsigned char null_sha1[GIT_MAX_RAWSZ];
33const struct object_id null_oid;
34const struct object_id empty_tree_oid = {
35 EMPTY_TREE_SHA1_BIN_LITERAL
36};
37const struct object_id empty_blob_oid = {
38 EMPTY_BLOB_SHA1_BIN_LITERAL
39};
40
41/*
42 * This is meant to hold a *small* number of objects that you would
43 * want read_sha1_file() to be able to return, but yet you do not want
44 * to write them into the object store (e.g. a browse-only
45 * application).
46 */
47static struct cached_object {
48 unsigned char sha1[20];
49 enum object_type type;
50 void *buf;
51 unsigned long size;
52} *cached_objects;
53static int cached_object_nr, cached_object_alloc;
54
55static struct cached_object empty_tree = {
56 EMPTY_TREE_SHA1_BIN_LITERAL,
57 OBJ_TREE,
58 "",
59 0
60};
61
62static struct cached_object *find_cached_object(const unsigned char *sha1)
63{
64 int i;
65 struct cached_object *co = cached_objects;
66
67 for (i = 0; i < cached_object_nr; i++, co++) {
68 if (!hashcmp(co->sha1, sha1))
69 return co;
70 }
71 if (!hashcmp(sha1, empty_tree.sha1))
72 return &empty_tree;
73 return NULL;
74}
75
76int mkdir_in_gitdir(const char *path)
77{
78 if (mkdir(path, 0777)) {
79 int saved_errno = errno;
80 struct stat st;
81 struct strbuf sb = STRBUF_INIT;
82
83 if (errno != EEXIST)
84 return -1;
85 /*
86 * Are we looking at a path in a symlinked worktree
87 * whose original repository does not yet have it?
88 * e.g. .git/rr-cache pointing at its original
89 * repository in which the user hasn't performed any
90 * conflict resolution yet?
91 */
92 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
93 strbuf_readlink(&sb, path, st.st_size) ||
94 !is_absolute_path(sb.buf) ||
95 mkdir(sb.buf, 0777)) {
96 strbuf_release(&sb);
97 errno = saved_errno;
98 return -1;
99 }
100 strbuf_release(&sb);
101 }
102 return adjust_shared_perm(path);
103}
104
105enum scld_error safe_create_leading_directories(char *path)
106{
107 char *next_component = path + offset_1st_component(path);
108 enum scld_error ret = SCLD_OK;
109
110 while (ret == SCLD_OK && next_component) {
111 struct stat st;
112 char *slash = next_component, slash_character;
113
114 while (*slash && !is_dir_sep(*slash))
115 slash++;
116
117 if (!*slash)
118 break;
119
120 next_component = slash + 1;
121 while (is_dir_sep(*next_component))
122 next_component++;
123 if (!*next_component)
124 break;
125
126 slash_character = *slash;
127 *slash = '\0';
128 if (!stat(path, &st)) {
129 /* path exists */
130 if (!S_ISDIR(st.st_mode)) {
131 errno = ENOTDIR;
132 ret = SCLD_EXISTS;
133 }
134 } else if (mkdir(path, 0777)) {
135 if (errno == EEXIST &&
136 !stat(path, &st) && S_ISDIR(st.st_mode))
137 ; /* somebody created it since we checked */
138 else if (errno == ENOENT)
139 /*
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
147 */
148 ret = SCLD_VANISHED;
149 else
150 ret = SCLD_FAILED;
151 } else if (adjust_shared_perm(path)) {
152 ret = SCLD_PERMS;
153 }
154 *slash = slash_character;
155 }
156 return ret;
157}
158
159enum scld_error safe_create_leading_directories_const(const char *path)
160{
161 int save_errno;
162 /* path points to cache entries, so xstrdup before messing with it */
163 char *buf = xstrdup(path);
164 enum scld_error result = safe_create_leading_directories(buf);
165
166 save_errno = errno;
167 free(buf);
168 errno = save_errno;
169 return result;
170}
171
172int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
173{
174 /*
175 * The number of times we will try to remove empty directories
176 * in the way of path. This is only 1 because if another
177 * process is racily creating directories that conflict with
178 * us, we don't want to fight against them.
179 */
180 int remove_directories_remaining = 1;
181
182 /*
183 * The number of times that we will try to create the
184 * directories containing path. We are willing to attempt this
185 * more than once, because another process could be trying to
186 * clean up empty directories at the same time as we are
187 * trying to create them.
188 */
189 int create_directories_remaining = 3;
190
191 /* A scratch copy of path, filled lazily if we need it: */
192 struct strbuf path_copy = STRBUF_INIT;
193
194 int ret, save_errno;
195
196 /* Sanity check: */
197 assert(*path);
198
199retry_fn:
200 ret = fn(path, cb);
201 save_errno = errno;
202 if (!ret)
203 goto out;
204
205 if (errno == EISDIR && remove_directories_remaining-- > 0) {
206 /*
207 * A directory is in the way. Maybe it is empty; try
208 * to remove it:
209 */
210 if (!path_copy.len)
211 strbuf_addstr(&path_copy, path);
212
213 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
214 goto retry_fn;
215 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
216 /*
217 * Maybe the containing directory didn't exist, or
218 * maybe it was just deleted by a process that is
219 * racing with us to clean up empty directories. Try
220 * to create it:
221 */
222 enum scld_error scld_result;
223
224 if (!path_copy.len)
225 strbuf_addstr(&path_copy, path);
226
227 do {
228 scld_result = safe_create_leading_directories(path_copy.buf);
229 if (scld_result == SCLD_OK)
230 goto retry_fn;
231 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
232 }
233
234out:
235 strbuf_release(&path_copy);
236 errno = save_errno;
237 return ret;
238}
239
240static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
241{
242 int i;
243 for (i = 0; i < 20; i++) {
244 static char hex[] = "0123456789abcdef";
245 unsigned int val = sha1[i];
246 strbuf_addch(buf, hex[val >> 4]);
247 strbuf_addch(buf, hex[val & 0xf]);
248 if (!i)
249 strbuf_addch(buf, '/');
250 }
251}
252
253const char *sha1_file_name(const unsigned char *sha1)
254{
255 static struct strbuf buf = STRBUF_INIT;
256
257 strbuf_reset(&buf);
258 strbuf_addf(&buf, "%s/", get_object_directory());
259
260 fill_sha1_path(&buf, sha1);
261 return buf.buf;
262}
263
264struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
265{
266 strbuf_setlen(&alt->scratch, alt->base_len);
267 return &alt->scratch;
268}
269
270static const char *alt_sha1_path(struct alternate_object_database *alt,
271 const unsigned char *sha1)
272{
273 struct strbuf *buf = alt_scratch_buf(alt);
274 fill_sha1_path(buf, sha1);
275 return buf->buf;
276}
277
278struct alternate_object_database *alt_odb_list;
279static struct alternate_object_database **alt_odb_tail;
280
281/*
282 * Return non-zero iff the path is usable as an alternate object database.
283 */
284static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
285{
286 struct alternate_object_database *alt;
287
288 /* Detect cases where alternate disappeared */
289 if (!is_directory(path->buf)) {
290 error("object directory %s does not exist; "
291 "check .git/objects/info/alternates.",
292 path->buf);
293 return 0;
294 }
295
296 /*
297 * Prevent the common mistake of listing the same
298 * thing twice, or object directory itself.
299 */
300 for (alt = alt_odb_list; alt; alt = alt->next) {
301 if (!fspathcmp(path->buf, alt->path))
302 return 0;
303 }
304 if (!fspathcmp(path->buf, normalized_objdir))
305 return 0;
306
307 return 1;
308}
309
310/*
311 * Prepare alternate object database registry.
312 *
313 * The variable alt_odb_list points at the list of struct
314 * alternate_object_database. The elements on this list come from
315 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
316 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
317 * whose contents is similar to that environment variable but can be
318 * LF separated. Its base points at a statically allocated buffer that
319 * contains "/the/directory/corresponding/to/.git/objects/...", while
320 * its name points just after the slash at the end of ".git/objects/"
321 * in the example above, and has enough space to hold 40-byte hex
322 * SHA1, an extra slash for the first level indirection, and the
323 * terminating NUL.
324 */
325static void read_info_alternates(const char * relative_base, int depth);
326static int link_alt_odb_entry(const char *entry, const char *relative_base,
327 int depth, const char *normalized_objdir)
328{
329 struct alternate_object_database *ent;
330 struct strbuf pathbuf = STRBUF_INIT;
331
332 if (!is_absolute_path(entry) && relative_base) {
333 strbuf_realpath(&pathbuf, relative_base, 1);
334 strbuf_addch(&pathbuf, '/');
335 }
336 strbuf_addstr(&pathbuf, entry);
337
338 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
339 error("unable to normalize alternate object path: %s",
340 pathbuf.buf);
341 strbuf_release(&pathbuf);
342 return -1;
343 }
344
345 /*
346 * The trailing slash after the directory name is given by
347 * this function at the end. Remove duplicates.
348 */
349 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
350 strbuf_setlen(&pathbuf, pathbuf.len - 1);
351
352 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
353 strbuf_release(&pathbuf);
354 return -1;
355 }
356
357 ent = alloc_alt_odb(pathbuf.buf);
358
359 /* add the alternate entry */
360 *alt_odb_tail = ent;
361 alt_odb_tail = &(ent->next);
362 ent->next = NULL;
363
364 /* recursively add alternates */
365 read_info_alternates(pathbuf.buf, depth + 1);
366
367 strbuf_release(&pathbuf);
368 return 0;
369}
370
371static const char *parse_alt_odb_entry(const char *string,
372 int sep,
373 struct strbuf *out)
374{
375 const char *end;
376
377 strbuf_reset(out);
378
379 if (*string == '#') {
380 /* comment; consume up to next separator */
381 end = strchrnul(string, sep);
382 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
383 /*
384 * quoted path; unquote_c_style has copied the
385 * data for us and set "end". Broken quoting (e.g.,
386 * an entry that doesn't end with a quote) falls
387 * back to the unquoted case below.
388 */
389 } else {
390 /* normal, unquoted path */
391 end = strchrnul(string, sep);
392 strbuf_add(out, string, end - string);
393 }
394
395 if (*end)
396 end++;
397 return end;
398}
399
400static void link_alt_odb_entries(const char *alt, int sep,
401 const char *relative_base, int depth)
402{
403 struct strbuf objdirbuf = STRBUF_INIT;
404 struct strbuf entry = STRBUF_INIT;
405
406 if (depth > 5) {
407 error("%s: ignoring alternate object stores, nesting too deep.",
408 relative_base);
409 return;
410 }
411
412 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
413 if (strbuf_normalize_path(&objdirbuf) < 0)
414 die("unable to normalize object directory: %s",
415 objdirbuf.buf);
416
417 while (*alt) {
418 alt = parse_alt_odb_entry(alt, sep, &entry);
419 if (!entry.len)
420 continue;
421 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
422 }
423 strbuf_release(&entry);
424 strbuf_release(&objdirbuf);
425}
426
427static void read_info_alternates(const char * relative_base, int depth)
428{
429 char *path;
430 struct strbuf buf = STRBUF_INIT;
431
432 path = xstrfmt("%s/info/alternates", relative_base);
433 if (strbuf_read_file(&buf, path, 1024) < 0) {
434 warn_on_fopen_errors(path);
435 free(path);
436 return;
437 }
438
439 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
440 strbuf_release(&buf);
441 free(path);
442}
443
444struct alternate_object_database *alloc_alt_odb(const char *dir)
445{
446 struct alternate_object_database *ent;
447
448 FLEX_ALLOC_STR(ent, path, dir);
449 strbuf_init(&ent->scratch, 0);
450 strbuf_addf(&ent->scratch, "%s/", dir);
451 ent->base_len = ent->scratch.len;
452
453 return ent;
454}
455
456void add_to_alternates_file(const char *reference)
457{
458 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
459 char *alts = git_pathdup("objects/info/alternates");
460 FILE *in, *out;
461
462 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
463 out = fdopen_lock_file(lock, "w");
464 if (!out)
465 die_errno("unable to fdopen alternates lockfile");
466
467 in = fopen(alts, "r");
468 if (in) {
469 struct strbuf line = STRBUF_INIT;
470 int found = 0;
471
472 while (strbuf_getline(&line, in) != EOF) {
473 if (!strcmp(reference, line.buf)) {
474 found = 1;
475 break;
476 }
477 fprintf_or_die(out, "%s\n", line.buf);
478 }
479
480 strbuf_release(&line);
481 fclose(in);
482
483 if (found) {
484 rollback_lock_file(lock);
485 lock = NULL;
486 }
487 }
488 else if (errno != ENOENT)
489 die_errno("unable to read alternates file");
490
491 if (lock) {
492 fprintf_or_die(out, "%s\n", reference);
493 if (commit_lock_file(lock))
494 die_errno("unable to move new alternates file into place");
495 if (alt_odb_tail)
496 link_alt_odb_entries(reference, '\n', NULL, 0);
497 }
498 free(alts);
499}
500
501void add_to_alternates_memory(const char *reference)
502{
503 /*
504 * Make sure alternates are initialized, or else our entry may be
505 * overwritten when they are.
506 */
507 prepare_alt_odb();
508
509 link_alt_odb_entries(reference, '\n', NULL, 0);
510}
511
512/*
513 * Compute the exact path an alternate is at and returns it. In case of
514 * error NULL is returned and the human readable error is added to `err`
515 * `path` may be relative and should point to $GITDIR.
516 * `err` must not be null.
517 */
518char *compute_alternate_path(const char *path, struct strbuf *err)
519{
520 char *ref_git = NULL;
521 const char *repo, *ref_git_s;
522 int seen_error = 0;
523
524 ref_git_s = real_path_if_valid(path);
525 if (!ref_git_s) {
526 seen_error = 1;
527 strbuf_addf(err, _("path '%s' does not exist"), path);
528 goto out;
529 } else
530 /*
531 * Beware: read_gitfile(), real_path() and mkpath()
532 * return static buffer
533 */
534 ref_git = xstrdup(ref_git_s);
535
536 repo = read_gitfile(ref_git);
537 if (!repo)
538 repo = read_gitfile(mkpath("%s/.git", ref_git));
539 if (repo) {
540 free(ref_git);
541 ref_git = xstrdup(repo);
542 }
543
544 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
545 char *ref_git_git = mkpathdup("%s/.git", ref_git);
546 free(ref_git);
547 ref_git = ref_git_git;
548 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
549 struct strbuf sb = STRBUF_INIT;
550 seen_error = 1;
551 if (get_common_dir(&sb, ref_git)) {
552 strbuf_addf(err,
553 _("reference repository '%s' as a linked "
554 "checkout is not supported yet."),
555 path);
556 goto out;
557 }
558
559 strbuf_addf(err, _("reference repository '%s' is not a "
560 "local repository."), path);
561 goto out;
562 }
563
564 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
565 strbuf_addf(err, _("reference repository '%s' is shallow"),
566 path);
567 seen_error = 1;
568 goto out;
569 }
570
571 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
572 strbuf_addf(err,
573 _("reference repository '%s' is grafted"),
574 path);
575 seen_error = 1;
576 goto out;
577 }
578
579out:
580 if (seen_error) {
581 FREE_AND_NULL(ref_git);
582 }
583
584 return ref_git;
585}
586
587int foreach_alt_odb(alt_odb_fn fn, void *cb)
588{
589 struct alternate_object_database *ent;
590 int r = 0;
591
592 prepare_alt_odb();
593 for (ent = alt_odb_list; ent; ent = ent->next) {
594 r = fn(ent, cb);
595 if (r)
596 break;
597 }
598 return r;
599}
600
601void prepare_alt_odb(void)
602{
603 const char *alt;
604
605 if (alt_odb_tail)
606 return;
607
608 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
609 if (!alt) alt = "";
610
611 alt_odb_tail = &alt_odb_list;
612 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
613
614 read_info_alternates(get_object_directory(), 0);
615}
616
617/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
618static int freshen_file(const char *fn)
619{
620 struct utimbuf t;
621 t.actime = t.modtime = time(NULL);
622 return !utime(fn, &t);
623}
624
625/*
626 * All of the check_and_freshen functions return 1 if the file exists and was
627 * freshened (if freshening was requested), 0 otherwise. If they return
628 * 0, you should not assume that it is safe to skip a write of the object (it
629 * either does not exist on disk, or has a stale mtime and may be subject to
630 * pruning).
631 */
632int check_and_freshen_file(const char *fn, int freshen)
633{
634 if (access(fn, F_OK))
635 return 0;
636 if (freshen && !freshen_file(fn))
637 return 0;
638 return 1;
639}
640
641static int check_and_freshen_local(const unsigned char *sha1, int freshen)
642{
643 return check_and_freshen_file(sha1_file_name(sha1), freshen);
644}
645
646static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
647{
648 struct alternate_object_database *alt;
649 prepare_alt_odb();
650 for (alt = alt_odb_list; alt; alt = alt->next) {
651 const char *path = alt_sha1_path(alt, sha1);
652 if (check_and_freshen_file(path, freshen))
653 return 1;
654 }
655 return 0;
656}
657
658static int check_and_freshen(const unsigned char *sha1, int freshen)
659{
660 return check_and_freshen_local(sha1, freshen) ||
661 check_and_freshen_nonlocal(sha1, freshen);
662}
663
664int has_loose_object_nonlocal(const unsigned char *sha1)
665{
666 return check_and_freshen_nonlocal(sha1, 0);
667}
668
669static int has_loose_object(const unsigned char *sha1)
670{
671 return check_and_freshen(sha1, 0);
672}
673
674static void mmap_limit_check(size_t length)
675{
676 static size_t limit = 0;
677 if (!limit) {
678 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
679 if (!limit)
680 limit = SIZE_MAX;
681 }
682 if (length > limit)
683 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
684 (uintmax_t)length, (uintmax_t)limit);
685}
686
687void *xmmap_gently(void *start, size_t length,
688 int prot, int flags, int fd, off_t offset)
689{
690 void *ret;
691
692 mmap_limit_check(length);
693 ret = mmap(start, length, prot, flags, fd, offset);
694 if (ret == MAP_FAILED) {
695 if (!length)
696 return NULL;
697 release_pack_memory(length);
698 ret = mmap(start, length, prot, flags, fd, offset);
699 }
700 return ret;
701}
702
703void *xmmap(void *start, size_t length,
704 int prot, int flags, int fd, off_t offset)
705{
706 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
707 if (ret == MAP_FAILED)
708 die_errno("mmap failed");
709 return ret;
710}
711
712/*
713 * With an in-core object data in "map", rehash it to make sure the
714 * object name actually matches "sha1" to detect object corruption.
715 * With "map" == NULL, try reading the object named with "sha1" using
716 * the streaming interface and rehash it to do the same.
717 */
718int check_sha1_signature(const unsigned char *sha1, void *map,
719 unsigned long size, const char *type)
720{
721 unsigned char real_sha1[20];
722 enum object_type obj_type;
723 struct git_istream *st;
724 git_SHA_CTX c;
725 char hdr[32];
726 int hdrlen;
727
728 if (map) {
729 hash_sha1_file(map, size, type, real_sha1);
730 return hashcmp(sha1, real_sha1) ? -1 : 0;
731 }
732
733 st = open_istream(sha1, &obj_type, &size, NULL);
734 if (!st)
735 return -1;
736
737 /* Generate the header */
738 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
739
740 /* Sha1.. */
741 git_SHA1_Init(&c);
742 git_SHA1_Update(&c, hdr, hdrlen);
743 for (;;) {
744 char buf[1024 * 16];
745 ssize_t readlen = read_istream(st, buf, sizeof(buf));
746
747 if (readlen < 0) {
748 close_istream(st);
749 return -1;
750 }
751 if (!readlen)
752 break;
753 git_SHA1_Update(&c, buf, readlen);
754 }
755 git_SHA1_Final(real_sha1, &c);
756 close_istream(st);
757 return hashcmp(sha1, real_sha1) ? -1 : 0;
758}
759
760int git_open_cloexec(const char *name, int flags)
761{
762 int fd;
763 static int o_cloexec = O_CLOEXEC;
764
765 fd = open(name, flags | o_cloexec);
766 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
767 /* Try again w/o O_CLOEXEC: the kernel might not support it */
768 o_cloexec &= ~O_CLOEXEC;
769 fd = open(name, flags | o_cloexec);
770 }
771
772#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
773 {
774 static int fd_cloexec = FD_CLOEXEC;
775
776 if (!o_cloexec && 0 <= fd && fd_cloexec) {
777 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
778 int flags = fcntl(fd, F_GETFD);
779 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
780 fd_cloexec = 0;
781 }
782 }
783#endif
784 return fd;
785}
786
787/*
788 * Find "sha1" as a loose object in the local repository or in an alternate.
789 * Returns 0 on success, negative on failure.
790 *
791 * The "path" out-parameter will give the path of the object we found (if any).
792 * Note that it may point to static storage and is only valid until another
793 * call to sha1_file_name(), etc.
794 */
795static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
796 const char **path)
797{
798 struct alternate_object_database *alt;
799
800 *path = sha1_file_name(sha1);
801 if (!lstat(*path, st))
802 return 0;
803
804 prepare_alt_odb();
805 errno = ENOENT;
806 for (alt = alt_odb_list; alt; alt = alt->next) {
807 *path = alt_sha1_path(alt, sha1);
808 if (!lstat(*path, st))
809 return 0;
810 }
811
812 return -1;
813}
814
815/*
816 * Like stat_sha1_file(), but actually open the object and return the
817 * descriptor. See the caveats on the "path" parameter above.
818 */
819static int open_sha1_file(const unsigned char *sha1, const char **path)
820{
821 int fd;
822 struct alternate_object_database *alt;
823 int most_interesting_errno;
824
825 *path = sha1_file_name(sha1);
826 fd = git_open(*path);
827 if (fd >= 0)
828 return fd;
829 most_interesting_errno = errno;
830
831 prepare_alt_odb();
832 for (alt = alt_odb_list; alt; alt = alt->next) {
833 *path = alt_sha1_path(alt, sha1);
834 fd = git_open(*path);
835 if (fd >= 0)
836 return fd;
837 if (most_interesting_errno == ENOENT)
838 most_interesting_errno = errno;
839 }
840 errno = most_interesting_errno;
841 return -1;
842}
843
844/*
845 * Map the loose object at "path" if it is not NULL, or the path found by
846 * searching for a loose object named "sha1".
847 */
848static void *map_sha1_file_1(const char *path,
849 const unsigned char *sha1,
850 unsigned long *size)
851{
852 void *map;
853 int fd;
854
855 if (path)
856 fd = git_open(path);
857 else
858 fd = open_sha1_file(sha1, &path);
859 map = NULL;
860 if (fd >= 0) {
861 struct stat st;
862
863 if (!fstat(fd, &st)) {
864 *size = xsize_t(st.st_size);
865 if (!*size) {
866 /* mmap() is forbidden on empty files */
867 error("object file %s is empty", path);
868 return NULL;
869 }
870 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
871 }
872 close(fd);
873 }
874 return map;
875}
876
877void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
878{
879 return map_sha1_file_1(NULL, sha1, size);
880}
881
882static int unpack_sha1_short_header(git_zstream *stream,
883 unsigned char *map, unsigned long mapsize,
884 void *buffer, unsigned long bufsiz)
885{
886 /* Get the data stream */
887 memset(stream, 0, sizeof(*stream));
888 stream->next_in = map;
889 stream->avail_in = mapsize;
890 stream->next_out = buffer;
891 stream->avail_out = bufsiz;
892
893 git_inflate_init(stream);
894 return git_inflate(stream, 0);
895}
896
897int unpack_sha1_header(git_zstream *stream,
898 unsigned char *map, unsigned long mapsize,
899 void *buffer, unsigned long bufsiz)
900{
901 int status = unpack_sha1_short_header(stream, map, mapsize,
902 buffer, bufsiz);
903
904 if (status < Z_OK)
905 return status;
906
907 /* Make sure we have the terminating NUL */
908 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
909 return -1;
910 return 0;
911}
912
913static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
914 unsigned long mapsize, void *buffer,
915 unsigned long bufsiz, struct strbuf *header)
916{
917 int status;
918
919 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
920 if (status < Z_OK)
921 return -1;
922
923 /*
924 * Check if entire header is unpacked in the first iteration.
925 */
926 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
927 return 0;
928
929 /*
930 * buffer[0..bufsiz] was not large enough. Copy the partial
931 * result out to header, and then append the result of further
932 * reading the stream.
933 */
934 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
935 stream->next_out = buffer;
936 stream->avail_out = bufsiz;
937
938 do {
939 status = git_inflate(stream, 0);
940 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
941 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
942 return 0;
943 stream->next_out = buffer;
944 stream->avail_out = bufsiz;
945 } while (status != Z_STREAM_END);
946 return -1;
947}
948
949static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
950{
951 int bytes = strlen(buffer) + 1;
952 unsigned char *buf = xmallocz(size);
953 unsigned long n;
954 int status = Z_OK;
955
956 n = stream->total_out - bytes;
957 if (n > size)
958 n = size;
959 memcpy(buf, (char *) buffer + bytes, n);
960 bytes = n;
961 if (bytes <= size) {
962 /*
963 * The above condition must be (bytes <= size), not
964 * (bytes < size). In other words, even though we
965 * expect no more output and set avail_out to zero,
966 * the input zlib stream may have bytes that express
967 * "this concludes the stream", and we *do* want to
968 * eat that input.
969 *
970 * Otherwise we would not be able to test that we
971 * consumed all the input to reach the expected size;
972 * we also want to check that zlib tells us that all
973 * went well with status == Z_STREAM_END at the end.
974 */
975 stream->next_out = buf + bytes;
976 stream->avail_out = size - bytes;
977 while (status == Z_OK)
978 status = git_inflate(stream, Z_FINISH);
979 }
980 if (status == Z_STREAM_END && !stream->avail_in) {
981 git_inflate_end(stream);
982 return buf;
983 }
984
985 if (status < 0)
986 error("corrupt loose object '%s'", sha1_to_hex(sha1));
987 else if (stream->avail_in)
988 error("garbage at end of loose object '%s'",
989 sha1_to_hex(sha1));
990 free(buf);
991 return NULL;
992}
993
994/*
995 * We used to just use "sscanf()", but that's actually way
996 * too permissive for what we want to check. So do an anal
997 * object header parse by hand.
998 */
999static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1000 unsigned int flags)
1001{
1002 const char *type_buf = hdr;
1003 unsigned long size;
1004 int type, type_len = 0;
1005
1006 /*
1007 * The type can be of any size but is followed by
1008 * a space.
1009 */
1010 for (;;) {
1011 char c = *hdr++;
1012 if (!c)
1013 return -1;
1014 if (c == ' ')
1015 break;
1016 type_len++;
1017 }
1018
1019 type = type_from_string_gently(type_buf, type_len, 1);
1020 if (oi->typename)
1021 strbuf_add(oi->typename, type_buf, type_len);
1022 /*
1023 * Set type to 0 if its an unknown object and
1024 * we're obtaining the type using '--allow-unknown-type'
1025 * option.
1026 */
1027 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1028 type = 0;
1029 else if (type < 0)
1030 die("invalid object type");
1031 if (oi->typep)
1032 *oi->typep = type;
1033
1034 /*
1035 * The length must follow immediately, and be in canonical
1036 * decimal format (ie "010" is not valid).
1037 */
1038 size = *hdr++ - '0';
1039 if (size > 9)
1040 return -1;
1041 if (size) {
1042 for (;;) {
1043 unsigned long c = *hdr - '0';
1044 if (c > 9)
1045 break;
1046 hdr++;
1047 size = size * 10 + c;
1048 }
1049 }
1050
1051 if (oi->sizep)
1052 *oi->sizep = size;
1053
1054 /*
1055 * The length must be followed by a zero byte
1056 */
1057 return *hdr ? -1 : type;
1058}
1059
1060int parse_sha1_header(const char *hdr, unsigned long *sizep)
1061{
1062 struct object_info oi = OBJECT_INFO_INIT;
1063
1064 oi.sizep = sizep;
1065 return parse_sha1_header_extended(hdr, &oi, 0);
1066}
1067
1068static int sha1_loose_object_info(const unsigned char *sha1,
1069 struct object_info *oi,
1070 int flags)
1071{
1072 int status = 0;
1073 unsigned long mapsize;
1074 void *map;
1075 git_zstream stream;
1076 char hdr[32];
1077 struct strbuf hdrbuf = STRBUF_INIT;
1078 unsigned long size_scratch;
1079
1080 if (oi->delta_base_sha1)
1081 hashclr(oi->delta_base_sha1);
1082
1083 /*
1084 * If we don't care about type or size, then we don't
1085 * need to look inside the object at all. Note that we
1086 * do not optimize out the stat call, even if the
1087 * caller doesn't care about the disk-size, since our
1088 * return value implicitly indicates whether the
1089 * object even exists.
1090 */
1091 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1092 const char *path;
1093 struct stat st;
1094 if (stat_sha1_file(sha1, &st, &path) < 0)
1095 return -1;
1096 if (oi->disk_sizep)
1097 *oi->disk_sizep = st.st_size;
1098 return 0;
1099 }
1100
1101 map = map_sha1_file(sha1, &mapsize);
1102 if (!map)
1103 return -1;
1104
1105 if (!oi->sizep)
1106 oi->sizep = &size_scratch;
1107
1108 if (oi->disk_sizep)
1109 *oi->disk_sizep = mapsize;
1110 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1111 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1112 status = error("unable to unpack %s header with --allow-unknown-type",
1113 sha1_to_hex(sha1));
1114 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1115 status = error("unable to unpack %s header",
1116 sha1_to_hex(sha1));
1117 if (status < 0)
1118 ; /* Do nothing */
1119 else if (hdrbuf.len) {
1120 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1121 status = error("unable to parse %s header with --allow-unknown-type",
1122 sha1_to_hex(sha1));
1123 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1124 status = error("unable to parse %s header", sha1_to_hex(sha1));
1125
1126 if (status >= 0 && oi->contentp)
1127 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1128 *oi->sizep, sha1);
1129 else
1130 git_inflate_end(&stream);
1131
1132 munmap(map, mapsize);
1133 if (status && oi->typep)
1134 *oi->typep = status;
1135 if (oi->sizep == &size_scratch)
1136 oi->sizep = NULL;
1137 strbuf_release(&hdrbuf);
1138 oi->whence = OI_LOOSE;
1139 return (status < 0) ? status : 0;
1140}
1141
1142int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1143{
1144 static struct object_info blank_oi = OBJECT_INFO_INIT;
1145 struct pack_entry e;
1146 int rtype;
1147 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1148 lookup_replace_object(sha1) :
1149 sha1;
1150
1151 if (!oi)
1152 oi = &blank_oi;
1153
1154 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1155 struct cached_object *co = find_cached_object(real);
1156 if (co) {
1157 if (oi->typep)
1158 *(oi->typep) = co->type;
1159 if (oi->sizep)
1160 *(oi->sizep) = co->size;
1161 if (oi->disk_sizep)
1162 *(oi->disk_sizep) = 0;
1163 if (oi->delta_base_sha1)
1164 hashclr(oi->delta_base_sha1);
1165 if (oi->typename)
1166 strbuf_addstr(oi->typename, typename(co->type));
1167 if (oi->contentp)
1168 *oi->contentp = xmemdupz(co->buf, co->size);
1169 oi->whence = OI_CACHED;
1170 return 0;
1171 }
1172 }
1173
1174 if (!find_pack_entry(real, &e)) {
1175 /* Most likely it's a loose object. */
1176 if (!sha1_loose_object_info(real, oi, flags))
1177 return 0;
1178
1179 /* Not a loose object; someone else may have just packed it. */
1180 if (flags & OBJECT_INFO_QUICK) {
1181 return -1;
1182 } else {
1183 reprepare_packed_git();
1184 if (!find_pack_entry(real, &e))
1185 return -1;
1186 }
1187 }
1188
1189 if (oi == &blank_oi)
1190 /*
1191 * We know that the caller doesn't actually need the
1192 * information below, so return early.
1193 */
1194 return 0;
1195
1196 rtype = packed_object_info(e.p, e.offset, oi);
1197 if (rtype < 0) {
1198 mark_bad_packed_object(e.p, real);
1199 return sha1_object_info_extended(real, oi, 0);
1200 } else if (oi->whence == OI_PACKED) {
1201 oi->u.packed.offset = e.offset;
1202 oi->u.packed.pack = e.p;
1203 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1204 rtype == OBJ_OFS_DELTA);
1205 }
1206
1207 return 0;
1208}
1209
1210/* returns enum object_type or negative */
1211int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1212{
1213 enum object_type type;
1214 struct object_info oi = OBJECT_INFO_INIT;
1215
1216 oi.typep = &type;
1217 oi.sizep = sizep;
1218 if (sha1_object_info_extended(sha1, &oi,
1219 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1220 return -1;
1221 return type;
1222}
1223
1224static void *read_object(const unsigned char *sha1, enum object_type *type,
1225 unsigned long *size)
1226{
1227 struct object_info oi = OBJECT_INFO_INIT;
1228 void *content;
1229 oi.typep = type;
1230 oi.sizep = size;
1231 oi.contentp = &content;
1232
1233 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1234 return NULL;
1235 return content;
1236}
1237
1238int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1239 unsigned char *sha1)
1240{
1241 struct cached_object *co;
1242
1243 hash_sha1_file(buf, len, typename(type), sha1);
1244 if (has_sha1_file(sha1) || find_cached_object(sha1))
1245 return 0;
1246 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1247 co = &cached_objects[cached_object_nr++];
1248 co->size = len;
1249 co->type = type;
1250 co->buf = xmalloc(len);
1251 memcpy(co->buf, buf, len);
1252 hashcpy(co->sha1, sha1);
1253 return 0;
1254}
1255
1256/*
1257 * This function dies on corrupt objects; the callers who want to
1258 * deal with them should arrange to call read_object() and give error
1259 * messages themselves.
1260 */
1261void *read_sha1_file_extended(const unsigned char *sha1,
1262 enum object_type *type,
1263 unsigned long *size,
1264 int lookup_replace)
1265{
1266 void *data;
1267 const struct packed_git *p;
1268 const char *path;
1269 struct stat st;
1270 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1271 : sha1;
1272
1273 errno = 0;
1274 data = read_object(repl, type, size);
1275 if (data)
1276 return data;
1277
1278 if (errno && errno != ENOENT)
1279 die_errno("failed to read object %s", sha1_to_hex(sha1));
1280
1281 /* die if we replaced an object with one that does not exist */
1282 if (repl != sha1)
1283 die("replacement %s not found for %s",
1284 sha1_to_hex(repl), sha1_to_hex(sha1));
1285
1286 if (!stat_sha1_file(repl, &st, &path))
1287 die("loose object %s (stored in %s) is corrupt",
1288 sha1_to_hex(repl), path);
1289
1290 if ((p = has_packed_and_bad(repl)) != NULL)
1291 die("packed object %s (stored in %s) is corrupt",
1292 sha1_to_hex(repl), p->pack_name);
1293
1294 return NULL;
1295}
1296
1297void *read_object_with_reference(const unsigned char *sha1,
1298 const char *required_type_name,
1299 unsigned long *size,
1300 unsigned char *actual_sha1_return)
1301{
1302 enum object_type type, required_type;
1303 void *buffer;
1304 unsigned long isize;
1305 unsigned char actual_sha1[20];
1306
1307 required_type = type_from_string(required_type_name);
1308 hashcpy(actual_sha1, sha1);
1309 while (1) {
1310 int ref_length = -1;
1311 const char *ref_type = NULL;
1312
1313 buffer = read_sha1_file(actual_sha1, &type, &isize);
1314 if (!buffer)
1315 return NULL;
1316 if (type == required_type) {
1317 *size = isize;
1318 if (actual_sha1_return)
1319 hashcpy(actual_sha1_return, actual_sha1);
1320 return buffer;
1321 }
1322 /* Handle references */
1323 else if (type == OBJ_COMMIT)
1324 ref_type = "tree ";
1325 else if (type == OBJ_TAG)
1326 ref_type = "object ";
1327 else {
1328 free(buffer);
1329 return NULL;
1330 }
1331 ref_length = strlen(ref_type);
1332
1333 if (ref_length + 40 > isize ||
1334 memcmp(buffer, ref_type, ref_length) ||
1335 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1336 free(buffer);
1337 return NULL;
1338 }
1339 free(buffer);
1340 /* Now we have the ID of the referred-to object in
1341 * actual_sha1. Check again. */
1342 }
1343}
1344
1345static void write_sha1_file_prepare(const void *buf, unsigned long len,
1346 const char *type, unsigned char *sha1,
1347 char *hdr, int *hdrlen)
1348{
1349 git_SHA_CTX c;
1350
1351 /* Generate the header */
1352 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1353
1354 /* Sha1.. */
1355 git_SHA1_Init(&c);
1356 git_SHA1_Update(&c, hdr, *hdrlen);
1357 git_SHA1_Update(&c, buf, len);
1358 git_SHA1_Final(sha1, &c);
1359}
1360
1361/*
1362 * Move the just written object into its final resting place.
1363 */
1364int finalize_object_file(const char *tmpfile, const char *filename)
1365{
1366 int ret = 0;
1367
1368 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1369 goto try_rename;
1370 else if (link(tmpfile, filename))
1371 ret = errno;
1372
1373 /*
1374 * Coda hack - coda doesn't like cross-directory links,
1375 * so we fall back to a rename, which will mean that it
1376 * won't be able to check collisions, but that's not a
1377 * big deal.
1378 *
1379 * The same holds for FAT formatted media.
1380 *
1381 * When this succeeds, we just return. We have nothing
1382 * left to unlink.
1383 */
1384 if (ret && ret != EEXIST) {
1385 try_rename:
1386 if (!rename(tmpfile, filename))
1387 goto out;
1388 ret = errno;
1389 }
1390 unlink_or_warn(tmpfile);
1391 if (ret) {
1392 if (ret != EEXIST) {
1393 return error_errno("unable to write sha1 filename %s", filename);
1394 }
1395 /* FIXME!!! Collision check here ? */
1396 }
1397
1398out:
1399 if (adjust_shared_perm(filename))
1400 return error("unable to set permission to '%s'", filename);
1401 return 0;
1402}
1403
1404static int write_buffer(int fd, const void *buf, size_t len)
1405{
1406 if (write_in_full(fd, buf, len) < 0)
1407 return error_errno("file write error");
1408 return 0;
1409}
1410
1411int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1412 unsigned char *sha1)
1413{
1414 char hdr[32];
1415 int hdrlen = sizeof(hdr);
1416 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1417 return 0;
1418}
1419
1420/* Finalize a file on disk, and close it. */
1421static void close_sha1_file(int fd)
1422{
1423 if (fsync_object_files)
1424 fsync_or_die(fd, "sha1 file");
1425 if (close(fd) != 0)
1426 die_errno("error when closing sha1 file");
1427}
1428
1429/* Size of directory component, including the ending '/' */
1430static inline int directory_size(const char *filename)
1431{
1432 const char *s = strrchr(filename, '/');
1433 if (!s)
1434 return 0;
1435 return s - filename + 1;
1436}
1437
1438/*
1439 * This creates a temporary file in the same directory as the final
1440 * 'filename'
1441 *
1442 * We want to avoid cross-directory filename renames, because those
1443 * can have problems on various filesystems (FAT, NFS, Coda).
1444 */
1445static int create_tmpfile(struct strbuf *tmp, const char *filename)
1446{
1447 int fd, dirlen = directory_size(filename);
1448
1449 strbuf_reset(tmp);
1450 strbuf_add(tmp, filename, dirlen);
1451 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1452 fd = git_mkstemp_mode(tmp->buf, 0444);
1453 if (fd < 0 && dirlen && errno == ENOENT) {
1454 /*
1455 * Make sure the directory exists; note that the contents
1456 * of the buffer are undefined after mkstemp returns an
1457 * error, so we have to rewrite the whole buffer from
1458 * scratch.
1459 */
1460 strbuf_reset(tmp);
1461 strbuf_add(tmp, filename, dirlen - 1);
1462 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1463 return -1;
1464 if (adjust_shared_perm(tmp->buf))
1465 return -1;
1466
1467 /* Try again */
1468 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1469 fd = git_mkstemp_mode(tmp->buf, 0444);
1470 }
1471 return fd;
1472}
1473
1474static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1475 const void *buf, unsigned long len, time_t mtime)
1476{
1477 int fd, ret;
1478 unsigned char compressed[4096];
1479 git_zstream stream;
1480 git_SHA_CTX c;
1481 unsigned char parano_sha1[20];
1482 static struct strbuf tmp_file = STRBUF_INIT;
1483 const char *filename = sha1_file_name(sha1);
1484
1485 fd = create_tmpfile(&tmp_file, filename);
1486 if (fd < 0) {
1487 if (errno == EACCES)
1488 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1489 else
1490 return error_errno("unable to create temporary file");
1491 }
1492
1493 /* Set it up */
1494 git_deflate_init(&stream, zlib_compression_level);
1495 stream.next_out = compressed;
1496 stream.avail_out = sizeof(compressed);
1497 git_SHA1_Init(&c);
1498
1499 /* First header.. */
1500 stream.next_in = (unsigned char *)hdr;
1501 stream.avail_in = hdrlen;
1502 while (git_deflate(&stream, 0) == Z_OK)
1503 ; /* nothing */
1504 git_SHA1_Update(&c, hdr, hdrlen);
1505
1506 /* Then the data itself.. */
1507 stream.next_in = (void *)buf;
1508 stream.avail_in = len;
1509 do {
1510 unsigned char *in0 = stream.next_in;
1511 ret = git_deflate(&stream, Z_FINISH);
1512 git_SHA1_Update(&c, in0, stream.next_in - in0);
1513 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1514 die("unable to write sha1 file");
1515 stream.next_out = compressed;
1516 stream.avail_out = sizeof(compressed);
1517 } while (ret == Z_OK);
1518
1519 if (ret != Z_STREAM_END)
1520 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1521 ret = git_deflate_end_gently(&stream);
1522 if (ret != Z_OK)
1523 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1524 git_SHA1_Final(parano_sha1, &c);
1525 if (hashcmp(sha1, parano_sha1) != 0)
1526 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1527
1528 close_sha1_file(fd);
1529
1530 if (mtime) {
1531 struct utimbuf utb;
1532 utb.actime = mtime;
1533 utb.modtime = mtime;
1534 if (utime(tmp_file.buf, &utb) < 0)
1535 warning_errno("failed utime() on %s", tmp_file.buf);
1536 }
1537
1538 return finalize_object_file(tmp_file.buf, filename);
1539}
1540
1541static int freshen_loose_object(const unsigned char *sha1)
1542{
1543 return check_and_freshen(sha1, 1);
1544}
1545
1546static int freshen_packed_object(const unsigned char *sha1)
1547{
1548 struct pack_entry e;
1549 if (!find_pack_entry(sha1, &e))
1550 return 0;
1551 if (e.p->freshened)
1552 return 1;
1553 if (!freshen_file(e.p->pack_name))
1554 return 0;
1555 e.p->freshened = 1;
1556 return 1;
1557}
1558
1559int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1560{
1561 char hdr[32];
1562 int hdrlen = sizeof(hdr);
1563
1564 /* Normally if we have it in the pack then we do not bother writing
1565 * it out into .git/objects/??/?{38} file.
1566 */
1567 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1568 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1569 return 0;
1570 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1571}
1572
1573int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1574 struct object_id *oid, unsigned flags)
1575{
1576 char *header;
1577 int hdrlen, status = 0;
1578
1579 /* type string, SP, %lu of the length plus NUL must fit this */
1580 hdrlen = strlen(type) + 32;
1581 header = xmalloc(hdrlen);
1582 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1583
1584 if (!(flags & HASH_WRITE_OBJECT))
1585 goto cleanup;
1586 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1587 goto cleanup;
1588 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1589
1590cleanup:
1591 free(header);
1592 return status;
1593}
1594
1595int force_object_loose(const unsigned char *sha1, time_t mtime)
1596{
1597 void *buf;
1598 unsigned long len;
1599 enum object_type type;
1600 char hdr[32];
1601 int hdrlen;
1602 int ret;
1603
1604 if (has_loose_object(sha1))
1605 return 0;
1606 buf = read_object(sha1, &type, &len);
1607 if (!buf)
1608 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1609 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1610 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1611 free(buf);
1612
1613 return ret;
1614}
1615
1616int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1617{
1618 if (!startup_info->have_repository)
1619 return 0;
1620 return sha1_object_info_extended(sha1, NULL,
1621 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1622}
1623
1624int has_object_file(const struct object_id *oid)
1625{
1626 return has_sha1_file(oid->hash);
1627}
1628
1629int has_object_file_with_flags(const struct object_id *oid, int flags)
1630{
1631 return has_sha1_file_with_flags(oid->hash, flags);
1632}
1633
1634static void check_tree(const void *buf, size_t size)
1635{
1636 struct tree_desc desc;
1637 struct name_entry entry;
1638
1639 init_tree_desc(&desc, buf, size);
1640 while (tree_entry(&desc, &entry))
1641 /* do nothing
1642 * tree_entry() will die() on malformed entries */
1643 ;
1644}
1645
1646static void check_commit(const void *buf, size_t size)
1647{
1648 struct commit c;
1649 memset(&c, 0, sizeof(c));
1650 if (parse_commit_buffer(&c, buf, size))
1651 die("corrupt commit");
1652}
1653
1654static void check_tag(const void *buf, size_t size)
1655{
1656 struct tag t;
1657 memset(&t, 0, sizeof(t));
1658 if (parse_tag_buffer(&t, buf, size))
1659 die("corrupt tag");
1660}
1661
1662static int index_mem(unsigned char *sha1, void *buf, size_t size,
1663 enum object_type type,
1664 const char *path, unsigned flags)
1665{
1666 int ret, re_allocated = 0;
1667 int write_object = flags & HASH_WRITE_OBJECT;
1668
1669 if (!type)
1670 type = OBJ_BLOB;
1671
1672 /*
1673 * Convert blobs to git internal format
1674 */
1675 if ((type == OBJ_BLOB) && path) {
1676 struct strbuf nbuf = STRBUF_INIT;
1677 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1678 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1679 buf = strbuf_detach(&nbuf, &size);
1680 re_allocated = 1;
1681 }
1682 }
1683 if (flags & HASH_FORMAT_CHECK) {
1684 if (type == OBJ_TREE)
1685 check_tree(buf, size);
1686 if (type == OBJ_COMMIT)
1687 check_commit(buf, size);
1688 if (type == OBJ_TAG)
1689 check_tag(buf, size);
1690 }
1691
1692 if (write_object)
1693 ret = write_sha1_file(buf, size, typename(type), sha1);
1694 else
1695 ret = hash_sha1_file(buf, size, typename(type), sha1);
1696 if (re_allocated)
1697 free(buf);
1698 return ret;
1699}
1700
1701static int index_stream_convert_blob(unsigned char *sha1, int fd,
1702 const char *path, unsigned flags)
1703{
1704 int ret;
1705 const int write_object = flags & HASH_WRITE_OBJECT;
1706 struct strbuf sbuf = STRBUF_INIT;
1707
1708 assert(path);
1709 assert(would_convert_to_git_filter_fd(path));
1710
1711 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1712 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1713
1714 if (write_object)
1715 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1716 sha1);
1717 else
1718 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1719 sha1);
1720 strbuf_release(&sbuf);
1721 return ret;
1722}
1723
1724static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1725 const char *path, unsigned flags)
1726{
1727 struct strbuf sbuf = STRBUF_INIT;
1728 int ret;
1729
1730 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1731 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1732 else
1733 ret = -1;
1734 strbuf_release(&sbuf);
1735 return ret;
1736}
1737
1738#define SMALL_FILE_SIZE (32*1024)
1739
1740static int index_core(unsigned char *sha1, int fd, size_t size,
1741 enum object_type type, const char *path,
1742 unsigned flags)
1743{
1744 int ret;
1745
1746 if (!size) {
1747 ret = index_mem(sha1, "", size, type, path, flags);
1748 } else if (size <= SMALL_FILE_SIZE) {
1749 char *buf = xmalloc(size);
1750 if (size == read_in_full(fd, buf, size))
1751 ret = index_mem(sha1, buf, size, type, path, flags);
1752 else
1753 ret = error_errno("short read");
1754 free(buf);
1755 } else {
1756 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1757 ret = index_mem(sha1, buf, size, type, path, flags);
1758 munmap(buf, size);
1759 }
1760 return ret;
1761}
1762
1763/*
1764 * This creates one packfile per large blob unless bulk-checkin
1765 * machinery is "plugged".
1766 *
1767 * This also bypasses the usual "convert-to-git" dance, and that is on
1768 * purpose. We could write a streaming version of the converting
1769 * functions and insert that before feeding the data to fast-import
1770 * (or equivalent in-core API described above). However, that is
1771 * somewhat complicated, as we do not know the size of the filter
1772 * result, which we need to know beforehand when writing a git object.
1773 * Since the primary motivation for trying to stream from the working
1774 * tree file and to avoid mmaping it in core is to deal with large
1775 * binary blobs, they generally do not want to get any conversion, and
1776 * callers should avoid this code path when filters are requested.
1777 */
1778static int index_stream(struct object_id *oid, int fd, size_t size,
1779 enum object_type type, const char *path,
1780 unsigned flags)
1781{
1782 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1783}
1784
1785int index_fd(struct object_id *oid, int fd, struct stat *st,
1786 enum object_type type, const char *path, unsigned flags)
1787{
1788 int ret;
1789
1790 /*
1791 * Call xsize_t() only when needed to avoid potentially unnecessary
1792 * die() for large files.
1793 */
1794 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1795 ret = index_stream_convert_blob(oid->hash, fd, path, flags);
1796 else if (!S_ISREG(st->st_mode))
1797 ret = index_pipe(oid->hash, fd, type, path, flags);
1798 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1799 (path && would_convert_to_git(&the_index, path)))
1800 ret = index_core(oid->hash, fd, xsize_t(st->st_size), type, path,
1801 flags);
1802 else
1803 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1804 flags);
1805 close(fd);
1806 return ret;
1807}
1808
1809int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1810{
1811 int fd;
1812 struct strbuf sb = STRBUF_INIT;
1813 int rc = 0;
1814
1815 switch (st->st_mode & S_IFMT) {
1816 case S_IFREG:
1817 fd = open(path, O_RDONLY);
1818 if (fd < 0)
1819 return error_errno("open(\"%s\")", path);
1820 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1821 return error("%s: failed to insert into database",
1822 path);
1823 break;
1824 case S_IFLNK:
1825 if (strbuf_readlink(&sb, path, st->st_size))
1826 return error_errno("readlink(\"%s\")", path);
1827 if (!(flags & HASH_WRITE_OBJECT))
1828 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1829 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1830 rc = error("%s: failed to insert into database", path);
1831 strbuf_release(&sb);
1832 break;
1833 case S_IFDIR:
1834 return resolve_gitlink_ref(path, "HEAD", oid->hash);
1835 default:
1836 return error("%s: unsupported file type", path);
1837 }
1838 return rc;
1839}
1840
1841int read_pack_header(int fd, struct pack_header *header)
1842{
1843 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1844 /* "eof before pack header was fully read" */
1845 return PH_ERROR_EOF;
1846
1847 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1848 /* "protocol error (pack signature mismatch detected)" */
1849 return PH_ERROR_PACK_SIGNATURE;
1850 if (!pack_version_ok(header->hdr_version))
1851 /* "protocol error (pack version unsupported)" */
1852 return PH_ERROR_PROTOCOL;
1853 return 0;
1854}
1855
1856void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1857{
1858 enum object_type type = sha1_object_info(sha1, NULL);
1859 if (type < 0)
1860 die("%s is not a valid object", sha1_to_hex(sha1));
1861 if (type != expect)
1862 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1863 typename(expect));
1864}
1865
1866int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1867 struct strbuf *path,
1868 each_loose_object_fn obj_cb,
1869 each_loose_cruft_fn cruft_cb,
1870 each_loose_subdir_fn subdir_cb,
1871 void *data)
1872{
1873 size_t origlen, baselen;
1874 DIR *dir;
1875 struct dirent *de;
1876 int r = 0;
1877
1878 if (subdir_nr > 0xff)
1879 BUG("invalid loose object subdirectory: %x", subdir_nr);
1880
1881 origlen = path->len;
1882 strbuf_complete(path, '/');
1883 strbuf_addf(path, "%02x", subdir_nr);
1884 baselen = path->len;
1885
1886 dir = opendir(path->buf);
1887 if (!dir) {
1888 if (errno != ENOENT)
1889 r = error_errno("unable to open %s", path->buf);
1890 strbuf_setlen(path, origlen);
1891 return r;
1892 }
1893
1894 while ((de = readdir(dir))) {
1895 if (is_dot_or_dotdot(de->d_name))
1896 continue;
1897
1898 strbuf_setlen(path, baselen);
1899 strbuf_addf(path, "/%s", de->d_name);
1900
1901 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1902 char hex[GIT_MAX_HEXSZ+1];
1903 struct object_id oid;
1904
1905 xsnprintf(hex, sizeof(hex), "%02x%s",
1906 subdir_nr, de->d_name);
1907 if (!get_oid_hex(hex, &oid)) {
1908 if (obj_cb) {
1909 r = obj_cb(&oid, path->buf, data);
1910 if (r)
1911 break;
1912 }
1913 continue;
1914 }
1915 }
1916
1917 if (cruft_cb) {
1918 r = cruft_cb(de->d_name, path->buf, data);
1919 if (r)
1920 break;
1921 }
1922 }
1923 closedir(dir);
1924
1925 strbuf_setlen(path, baselen);
1926 if (!r && subdir_cb)
1927 r = subdir_cb(subdir_nr, path->buf, data);
1928
1929 strbuf_setlen(path, origlen);
1930
1931 return r;
1932}
1933
1934int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1935 each_loose_object_fn obj_cb,
1936 each_loose_cruft_fn cruft_cb,
1937 each_loose_subdir_fn subdir_cb,
1938 void *data)
1939{
1940 int r = 0;
1941 int i;
1942
1943 for (i = 0; i < 256; i++) {
1944 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1945 subdir_cb, data);
1946 if (r)
1947 break;
1948 }
1949
1950 return r;
1951}
1952
1953int for_each_loose_file_in_objdir(const char *path,
1954 each_loose_object_fn obj_cb,
1955 each_loose_cruft_fn cruft_cb,
1956 each_loose_subdir_fn subdir_cb,
1957 void *data)
1958{
1959 struct strbuf buf = STRBUF_INIT;
1960 int r;
1961
1962 strbuf_addstr(&buf, path);
1963 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1964 subdir_cb, data);
1965 strbuf_release(&buf);
1966
1967 return r;
1968}
1969
1970struct loose_alt_odb_data {
1971 each_loose_object_fn *cb;
1972 void *data;
1973};
1974
1975static int loose_from_alt_odb(struct alternate_object_database *alt,
1976 void *vdata)
1977{
1978 struct loose_alt_odb_data *data = vdata;
1979 struct strbuf buf = STRBUF_INIT;
1980 int r;
1981
1982 strbuf_addstr(&buf, alt->path);
1983 r = for_each_loose_file_in_objdir_buf(&buf,
1984 data->cb, NULL, NULL,
1985 data->data);
1986 strbuf_release(&buf);
1987 return r;
1988}
1989
1990int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
1991{
1992 struct loose_alt_odb_data alt;
1993 int r;
1994
1995 r = for_each_loose_file_in_objdir(get_object_directory(),
1996 cb, NULL, NULL, data);
1997 if (r)
1998 return r;
1999
2000 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2001 return 0;
2002
2003 alt.cb = cb;
2004 alt.data = data;
2005 return foreach_alt_odb(loose_from_alt_odb, &alt);
2006}
2007
2008static int check_stream_sha1(git_zstream *stream,
2009 const char *hdr,
2010 unsigned long size,
2011 const char *path,
2012 const unsigned char *expected_sha1)
2013{
2014 git_SHA_CTX c;
2015 unsigned char real_sha1[GIT_MAX_RAWSZ];
2016 unsigned char buf[4096];
2017 unsigned long total_read;
2018 int status = Z_OK;
2019
2020 git_SHA1_Init(&c);
2021 git_SHA1_Update(&c, hdr, stream->total_out);
2022
2023 /*
2024 * We already read some bytes into hdr, but the ones up to the NUL
2025 * do not count against the object's content size.
2026 */
2027 total_read = stream->total_out - strlen(hdr) - 1;
2028
2029 /*
2030 * This size comparison must be "<=" to read the final zlib packets;
2031 * see the comment in unpack_sha1_rest for details.
2032 */
2033 while (total_read <= size &&
2034 (status == Z_OK || status == Z_BUF_ERROR)) {
2035 stream->next_out = buf;
2036 stream->avail_out = sizeof(buf);
2037 if (size - total_read < stream->avail_out)
2038 stream->avail_out = size - total_read;
2039 status = git_inflate(stream, Z_FINISH);
2040 git_SHA1_Update(&c, buf, stream->next_out - buf);
2041 total_read += stream->next_out - buf;
2042 }
2043 git_inflate_end(stream);
2044
2045 if (status != Z_STREAM_END) {
2046 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2047 return -1;
2048 }
2049 if (stream->avail_in) {
2050 error("garbage at end of loose object '%s'",
2051 sha1_to_hex(expected_sha1));
2052 return -1;
2053 }
2054
2055 git_SHA1_Final(real_sha1, &c);
2056 if (hashcmp(expected_sha1, real_sha1)) {
2057 error("sha1 mismatch for %s (expected %s)", path,
2058 sha1_to_hex(expected_sha1));
2059 return -1;
2060 }
2061
2062 return 0;
2063}
2064
2065int read_loose_object(const char *path,
2066 const unsigned char *expected_sha1,
2067 enum object_type *type,
2068 unsigned long *size,
2069 void **contents)
2070{
2071 int ret = -1;
2072 void *map = NULL;
2073 unsigned long mapsize;
2074 git_zstream stream;
2075 char hdr[32];
2076
2077 *contents = NULL;
2078
2079 map = map_sha1_file_1(path, NULL, &mapsize);
2080 if (!map) {
2081 error_errno("unable to mmap %s", path);
2082 goto out;
2083 }
2084
2085 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2086 error("unable to unpack header of %s", path);
2087 goto out;
2088 }
2089
2090 *type = parse_sha1_header(hdr, size);
2091 if (*type < 0) {
2092 error("unable to parse header of %s", path);
2093 git_inflate_end(&stream);
2094 goto out;
2095 }
2096
2097 if (*type == OBJ_BLOB) {
2098 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2099 goto out;
2100 } else {
2101 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2102 if (!*contents) {
2103 error("unable to unpack contents of %s", path);
2104 git_inflate_end(&stream);
2105 goto out;
2106 }
2107 if (check_sha1_signature(expected_sha1, *contents,
2108 *size, typename(*type))) {
2109 error("sha1 mismatch for %s (expected %s)", path,
2110 sha1_to_hex(expected_sha1));
2111 free(*contents);
2112 goto out;
2113 }
2114 }
2115
2116 ret = 0; /* everything checks out */
2117
2118out:
2119 if (map)
2120 munmap(map, mapsize);
2121 return ret;
2122}