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 "mru.h"
28#include "list.h"
29#include "mergesort.h"
30#include "quote.h"
31#include "packfile.h"
32
33const unsigned char null_sha1[GIT_MAX_RAWSZ];
34const struct object_id null_oid;
35const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
37};
38const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
40};
41
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
60 0
61};
62
63static struct cached_object *find_cached_object(const unsigned char *sha1)
64{
65 int i;
66 struct cached_object *co = cached_objects;
67
68 for (i = 0; i < cached_object_nr; i++, co++) {
69 if (!hashcmp(co->sha1, sha1))
70 return co;
71 }
72 if (!hashcmp(sha1, empty_tree.sha1))
73 return &empty_tree;
74 return NULL;
75}
76
77int mkdir_in_gitdir(const char *path)
78{
79 if (mkdir(path, 0777)) {
80 int saved_errno = errno;
81 struct stat st;
82 struct strbuf sb = STRBUF_INIT;
83
84 if (errno != EEXIST)
85 return -1;
86 /*
87 * Are we looking at a path in a symlinked worktree
88 * whose original repository does not yet have it?
89 * e.g. .git/rr-cache pointing at its original
90 * repository in which the user hasn't performed any
91 * conflict resolution yet?
92 */
93 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
94 strbuf_readlink(&sb, path, st.st_size) ||
95 !is_absolute_path(sb.buf) ||
96 mkdir(sb.buf, 0777)) {
97 strbuf_release(&sb);
98 errno = saved_errno;
99 return -1;
100 }
101 strbuf_release(&sb);
102 }
103 return adjust_shared_perm(path);
104}
105
106enum scld_error safe_create_leading_directories(char *path)
107{
108 char *next_component = path + offset_1st_component(path);
109 enum scld_error ret = SCLD_OK;
110
111 while (ret == SCLD_OK && next_component) {
112 struct stat st;
113 char *slash = next_component, slash_character;
114
115 while (*slash && !is_dir_sep(*slash))
116 slash++;
117
118 if (!*slash)
119 break;
120
121 next_component = slash + 1;
122 while (is_dir_sep(*next_component))
123 next_component++;
124 if (!*next_component)
125 break;
126
127 slash_character = *slash;
128 *slash = '\0';
129 if (!stat(path, &st)) {
130 /* path exists */
131 if (!S_ISDIR(st.st_mode)) {
132 errno = ENOTDIR;
133 ret = SCLD_EXISTS;
134 }
135 } else if (mkdir(path, 0777)) {
136 if (errno == EEXIST &&
137 !stat(path, &st) && S_ISDIR(st.st_mode))
138 ; /* somebody created it since we checked */
139 else if (errno == ENOENT)
140 /*
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
147 * trying again:
148 */
149 ret = SCLD_VANISHED;
150 else
151 ret = SCLD_FAILED;
152 } else if (adjust_shared_perm(path)) {
153 ret = SCLD_PERMS;
154 }
155 *slash = slash_character;
156 }
157 return ret;
158}
159
160enum scld_error safe_create_leading_directories_const(const char *path)
161{
162 int save_errno;
163 /* path points to cache entries, so xstrdup before messing with it */
164 char *buf = xstrdup(path);
165 enum scld_error result = safe_create_leading_directories(buf);
166
167 save_errno = errno;
168 free(buf);
169 errno = save_errno;
170 return result;
171}
172
173int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
174{
175 /*
176 * The number of times we will try to remove empty directories
177 * in the way of path. This is only 1 because if another
178 * process is racily creating directories that conflict with
179 * us, we don't want to fight against them.
180 */
181 int remove_directories_remaining = 1;
182
183 /*
184 * The number of times that we will try to create the
185 * directories containing path. We are willing to attempt this
186 * more than once, because another process could be trying to
187 * clean up empty directories at the same time as we are
188 * trying to create them.
189 */
190 int create_directories_remaining = 3;
191
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy = STRBUF_INIT;
194
195 int ret, save_errno;
196
197 /* Sanity check: */
198 assert(*path);
199
200retry_fn:
201 ret = fn(path, cb);
202 save_errno = errno;
203 if (!ret)
204 goto out;
205
206 if (errno == EISDIR && remove_directories_remaining-- > 0) {
207 /*
208 * A directory is in the way. Maybe it is empty; try
209 * to remove it:
210 */
211 if (!path_copy.len)
212 strbuf_addstr(&path_copy, path);
213
214 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
215 goto retry_fn;
216 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
217 /*
218 * Maybe the containing directory didn't exist, or
219 * maybe it was just deleted by a process that is
220 * racing with us to clean up empty directories. Try
221 * to create it:
222 */
223 enum scld_error scld_result;
224
225 if (!path_copy.len)
226 strbuf_addstr(&path_copy, path);
227
228 do {
229 scld_result = safe_create_leading_directories(path_copy.buf);
230 if (scld_result == SCLD_OK)
231 goto retry_fn;
232 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
233 }
234
235out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
239}
240
241static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
242{
243 int i;
244 for (i = 0; i < 20; i++) {
245 static char hex[] = "0123456789abcdef";
246 unsigned int val = sha1[i];
247 strbuf_addch(buf, hex[val >> 4]);
248 strbuf_addch(buf, hex[val & 0xf]);
249 if (!i)
250 strbuf_addch(buf, '/');
251 }
252}
253
254const char *sha1_file_name(const unsigned char *sha1)
255{
256 static struct strbuf buf = STRBUF_INIT;
257
258 strbuf_reset(&buf);
259 strbuf_addf(&buf, "%s/", get_object_directory());
260
261 fill_sha1_path(&buf, sha1);
262 return buf.buf;
263}
264
265struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
266{
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
269}
270
271static const char *alt_sha1_path(struct alternate_object_database *alt,
272 const unsigned char *sha1)
273{
274 struct strbuf *buf = alt_scratch_buf(alt);
275 fill_sha1_path(buf, sha1);
276 return buf->buf;
277}
278
279struct alternate_object_database *alt_odb_list;
280static struct alternate_object_database **alt_odb_tail;
281
282/*
283 * Return non-zero iff the path is usable as an alternate object database.
284 */
285static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
286{
287 struct alternate_object_database *alt;
288
289 /* Detect cases where alternate disappeared */
290 if (!is_directory(path->buf)) {
291 error("object directory %s does not exist; "
292 "check .git/objects/info/alternates.",
293 path->buf);
294 return 0;
295 }
296
297 /*
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
300 */
301 for (alt = alt_odb_list; alt; alt = alt->next) {
302 if (!fspathcmp(path->buf, alt->path))
303 return 0;
304 }
305 if (!fspathcmp(path->buf, normalized_objdir))
306 return 0;
307
308 return 1;
309}
310
311/*
312 * Prepare alternate object database registry.
313 *
314 * The variable alt_odb_list points at the list of struct
315 * alternate_object_database. The elements on this list come from
316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
318 * whose contents is similar to that environment variable but can be
319 * LF separated. Its base points at a statically allocated buffer that
320 * contains "/the/directory/corresponding/to/.git/objects/...", while
321 * its name points just after the slash at the end of ".git/objects/"
322 * in the example above, and has enough space to hold 40-byte hex
323 * SHA1, an extra slash for the first level indirection, and the
324 * terminating NUL.
325 */
326static void read_info_alternates(const char * relative_base, int depth);
327static int link_alt_odb_entry(const char *entry, const char *relative_base,
328 int depth, const char *normalized_objdir)
329{
330 struct alternate_object_database *ent;
331 struct strbuf pathbuf = STRBUF_INIT;
332
333 if (!is_absolute_path(entry) && relative_base) {
334 strbuf_realpath(&pathbuf, relative_base, 1);
335 strbuf_addch(&pathbuf, '/');
336 }
337 strbuf_addstr(&pathbuf, entry);
338
339 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
340 error("unable to normalize alternate object path: %s",
341 pathbuf.buf);
342 strbuf_release(&pathbuf);
343 return -1;
344 }
345
346 /*
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
349 */
350 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
351 strbuf_setlen(&pathbuf, pathbuf.len - 1);
352
353 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
354 strbuf_release(&pathbuf);
355 return -1;
356 }
357
358 ent = alloc_alt_odb(pathbuf.buf);
359
360 /* add the alternate entry */
361 *alt_odb_tail = ent;
362 alt_odb_tail = &(ent->next);
363 ent->next = NULL;
364
365 /* recursively add alternates */
366 read_info_alternates(pathbuf.buf, depth + 1);
367
368 strbuf_release(&pathbuf);
369 return 0;
370}
371
372static const char *parse_alt_odb_entry(const char *string,
373 int sep,
374 struct strbuf *out)
375{
376 const char *end;
377
378 strbuf_reset(out);
379
380 if (*string == '#') {
381 /* comment; consume up to next separator */
382 end = strchrnul(string, sep);
383 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
384 /*
385 * quoted path; unquote_c_style has copied the
386 * data for us and set "end". Broken quoting (e.g.,
387 * an entry that doesn't end with a quote) falls
388 * back to the unquoted case below.
389 */
390 } else {
391 /* normal, unquoted path */
392 end = strchrnul(string, sep);
393 strbuf_add(out, string, end - string);
394 }
395
396 if (*end)
397 end++;
398 return end;
399}
400
401static void link_alt_odb_entries(const char *alt, int len, int sep,
402 const char *relative_base, int depth)
403{
404 struct strbuf objdirbuf = STRBUF_INIT;
405 struct strbuf entry = STRBUF_INIT;
406
407 if (depth > 5) {
408 error("%s: ignoring alternate object stores, nesting too deep.",
409 relative_base);
410 return;
411 }
412
413 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf) < 0)
415 die("unable to normalize object directory: %s",
416 objdirbuf.buf);
417
418 while (*alt) {
419 alt = parse_alt_odb_entry(alt, sep, &entry);
420 if (!entry.len)
421 continue;
422 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
423 }
424 strbuf_release(&entry);
425 strbuf_release(&objdirbuf);
426}
427
428static void read_info_alternates(const char * relative_base, int depth)
429{
430 char *map;
431 size_t mapsz;
432 struct stat st;
433 char *path;
434 int fd;
435
436 path = xstrfmt("%s/info/alternates", relative_base);
437 fd = git_open(path);
438 free(path);
439 if (fd < 0)
440 return;
441 if (fstat(fd, &st) || (st.st_size == 0)) {
442 close(fd);
443 return;
444 }
445 mapsz = xsize_t(st.st_size);
446 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
447 close(fd);
448
449 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
450
451 munmap(map, mapsz);
452}
453
454struct alternate_object_database *alloc_alt_odb(const char *dir)
455{
456 struct alternate_object_database *ent;
457
458 FLEX_ALLOC_STR(ent, path, dir);
459 strbuf_init(&ent->scratch, 0);
460 strbuf_addf(&ent->scratch, "%s/", dir);
461 ent->base_len = ent->scratch.len;
462
463 return ent;
464}
465
466void add_to_alternates_file(const char *reference)
467{
468 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
469 char *alts = git_pathdup("objects/info/alternates");
470 FILE *in, *out;
471
472 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
473 out = fdopen_lock_file(lock, "w");
474 if (!out)
475 die_errno("unable to fdopen alternates lockfile");
476
477 in = fopen(alts, "r");
478 if (in) {
479 struct strbuf line = STRBUF_INIT;
480 int found = 0;
481
482 while (strbuf_getline(&line, in) != EOF) {
483 if (!strcmp(reference, line.buf)) {
484 found = 1;
485 break;
486 }
487 fprintf_or_die(out, "%s\n", line.buf);
488 }
489
490 strbuf_release(&line);
491 fclose(in);
492
493 if (found) {
494 rollback_lock_file(lock);
495 lock = NULL;
496 }
497 }
498 else if (errno != ENOENT)
499 die_errno("unable to read alternates file");
500
501 if (lock) {
502 fprintf_or_die(out, "%s\n", reference);
503 if (commit_lock_file(lock))
504 die_errno("unable to move new alternates file into place");
505 if (alt_odb_tail)
506 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
507 }
508 free(alts);
509}
510
511void add_to_alternates_memory(const char *reference)
512{
513 /*
514 * Make sure alternates are initialized, or else our entry may be
515 * overwritten when they are.
516 */
517 prepare_alt_odb();
518
519 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
520}
521
522/*
523 * Compute the exact path an alternate is at and returns it. In case of
524 * error NULL is returned and the human readable error is added to `err`
525 * `path` may be relative and should point to $GITDIR.
526 * `err` must not be null.
527 */
528char *compute_alternate_path(const char *path, struct strbuf *err)
529{
530 char *ref_git = NULL;
531 const char *repo, *ref_git_s;
532 int seen_error = 0;
533
534 ref_git_s = real_path_if_valid(path);
535 if (!ref_git_s) {
536 seen_error = 1;
537 strbuf_addf(err, _("path '%s' does not exist"), path);
538 goto out;
539 } else
540 /*
541 * Beware: read_gitfile(), real_path() and mkpath()
542 * return static buffer
543 */
544 ref_git = xstrdup(ref_git_s);
545
546 repo = read_gitfile(ref_git);
547 if (!repo)
548 repo = read_gitfile(mkpath("%s/.git", ref_git));
549 if (repo) {
550 free(ref_git);
551 ref_git = xstrdup(repo);
552 }
553
554 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
555 char *ref_git_git = mkpathdup("%s/.git", ref_git);
556 free(ref_git);
557 ref_git = ref_git_git;
558 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
559 struct strbuf sb = STRBUF_INIT;
560 seen_error = 1;
561 if (get_common_dir(&sb, ref_git)) {
562 strbuf_addf(err,
563 _("reference repository '%s' as a linked "
564 "checkout is not supported yet."),
565 path);
566 goto out;
567 }
568
569 strbuf_addf(err, _("reference repository '%s' is not a "
570 "local repository."), path);
571 goto out;
572 }
573
574 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
575 strbuf_addf(err, _("reference repository '%s' is shallow"),
576 path);
577 seen_error = 1;
578 goto out;
579 }
580
581 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
582 strbuf_addf(err,
583 _("reference repository '%s' is grafted"),
584 path);
585 seen_error = 1;
586 goto out;
587 }
588
589out:
590 if (seen_error) {
591 FREE_AND_NULL(ref_git);
592 }
593
594 return ref_git;
595}
596
597int foreach_alt_odb(alt_odb_fn fn, void *cb)
598{
599 struct alternate_object_database *ent;
600 int r = 0;
601
602 prepare_alt_odb();
603 for (ent = alt_odb_list; ent; ent = ent->next) {
604 r = fn(ent, cb);
605 if (r)
606 break;
607 }
608 return r;
609}
610
611void prepare_alt_odb(void)
612{
613 const char *alt;
614
615 if (alt_odb_tail)
616 return;
617
618 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
619 if (!alt) alt = "";
620
621 alt_odb_tail = &alt_odb_list;
622 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
623
624 read_info_alternates(get_object_directory(), 0);
625}
626
627/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
628static int freshen_file(const char *fn)
629{
630 struct utimbuf t;
631 t.actime = t.modtime = time(NULL);
632 return !utime(fn, &t);
633}
634
635/*
636 * All of the check_and_freshen functions return 1 if the file exists and was
637 * freshened (if freshening was requested), 0 otherwise. If they return
638 * 0, you should not assume that it is safe to skip a write of the object (it
639 * either does not exist on disk, or has a stale mtime and may be subject to
640 * pruning).
641 */
642int check_and_freshen_file(const char *fn, int freshen)
643{
644 if (access(fn, F_OK))
645 return 0;
646 if (freshen && !freshen_file(fn))
647 return 0;
648 return 1;
649}
650
651static int check_and_freshen_local(const unsigned char *sha1, int freshen)
652{
653 return check_and_freshen_file(sha1_file_name(sha1), freshen);
654}
655
656static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
657{
658 struct alternate_object_database *alt;
659 prepare_alt_odb();
660 for (alt = alt_odb_list; alt; alt = alt->next) {
661 const char *path = alt_sha1_path(alt, sha1);
662 if (check_and_freshen_file(path, freshen))
663 return 1;
664 }
665 return 0;
666}
667
668static int check_and_freshen(const unsigned char *sha1, int freshen)
669{
670 return check_and_freshen_local(sha1, freshen) ||
671 check_and_freshen_nonlocal(sha1, freshen);
672}
673
674int has_loose_object_nonlocal(const unsigned char *sha1)
675{
676 return check_and_freshen_nonlocal(sha1, 0);
677}
678
679static int has_loose_object(const unsigned char *sha1)
680{
681 return check_and_freshen(sha1, 0);
682}
683
684static void mmap_limit_check(size_t length)
685{
686 static size_t limit = 0;
687 if (!limit) {
688 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
689 if (!limit)
690 limit = SIZE_MAX;
691 }
692 if (length > limit)
693 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
694 (uintmax_t)length, (uintmax_t)limit);
695}
696
697void *xmmap_gently(void *start, size_t length,
698 int prot, int flags, int fd, off_t offset)
699{
700 void *ret;
701
702 mmap_limit_check(length);
703 ret = mmap(start, length, prot, flags, fd, offset);
704 if (ret == MAP_FAILED) {
705 if (!length)
706 return NULL;
707 release_pack_memory(length);
708 ret = mmap(start, length, prot, flags, fd, offset);
709 }
710 return ret;
711}
712
713void *xmmap(void *start, size_t length,
714 int prot, int flags, int fd, off_t offset)
715{
716 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
717 if (ret == MAP_FAILED)
718 die_errno("mmap failed");
719 return ret;
720}
721
722/*
723 * With an in-core object data in "map", rehash it to make sure the
724 * object name actually matches "sha1" to detect object corruption.
725 * With "map" == NULL, try reading the object named with "sha1" using
726 * the streaming interface and rehash it to do the same.
727 */
728int check_sha1_signature(const unsigned char *sha1, void *map,
729 unsigned long size, const char *type)
730{
731 unsigned char real_sha1[20];
732 enum object_type obj_type;
733 struct git_istream *st;
734 git_SHA_CTX c;
735 char hdr[32];
736 int hdrlen;
737
738 if (map) {
739 hash_sha1_file(map, size, type, real_sha1);
740 return hashcmp(sha1, real_sha1) ? -1 : 0;
741 }
742
743 st = open_istream(sha1, &obj_type, &size, NULL);
744 if (!st)
745 return -1;
746
747 /* Generate the header */
748 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
749
750 /* Sha1.. */
751 git_SHA1_Init(&c);
752 git_SHA1_Update(&c, hdr, hdrlen);
753 for (;;) {
754 char buf[1024 * 16];
755 ssize_t readlen = read_istream(st, buf, sizeof(buf));
756
757 if (readlen < 0) {
758 close_istream(st);
759 return -1;
760 }
761 if (!readlen)
762 break;
763 git_SHA1_Update(&c, buf, readlen);
764 }
765 git_SHA1_Final(real_sha1, &c);
766 close_istream(st);
767 return hashcmp(sha1, real_sha1) ? -1 : 0;
768}
769
770int git_open_cloexec(const char *name, int flags)
771{
772 int fd;
773 static int o_cloexec = O_CLOEXEC;
774
775 fd = open(name, flags | o_cloexec);
776 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
777 /* Try again w/o O_CLOEXEC: the kernel might not support it */
778 o_cloexec &= ~O_CLOEXEC;
779 fd = open(name, flags | o_cloexec);
780 }
781
782#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
783 {
784 static int fd_cloexec = FD_CLOEXEC;
785
786 if (!o_cloexec && 0 <= fd && fd_cloexec) {
787 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
788 int flags = fcntl(fd, F_GETFD);
789 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
790 fd_cloexec = 0;
791 }
792 }
793#endif
794 return fd;
795}
796
797/*
798 * Find "sha1" as a loose object in the local repository or in an alternate.
799 * Returns 0 on success, negative on failure.
800 *
801 * The "path" out-parameter will give the path of the object we found (if any).
802 * Note that it may point to static storage and is only valid until another
803 * call to sha1_file_name(), etc.
804 */
805static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
806 const char **path)
807{
808 struct alternate_object_database *alt;
809
810 *path = sha1_file_name(sha1);
811 if (!lstat(*path, st))
812 return 0;
813
814 prepare_alt_odb();
815 errno = ENOENT;
816 for (alt = alt_odb_list; alt; alt = alt->next) {
817 *path = alt_sha1_path(alt, sha1);
818 if (!lstat(*path, st))
819 return 0;
820 }
821
822 return -1;
823}
824
825/*
826 * Like stat_sha1_file(), but actually open the object and return the
827 * descriptor. See the caveats on the "path" parameter above.
828 */
829static int open_sha1_file(const unsigned char *sha1, const char **path)
830{
831 int fd;
832 struct alternate_object_database *alt;
833 int most_interesting_errno;
834
835 *path = sha1_file_name(sha1);
836 fd = git_open(*path);
837 if (fd >= 0)
838 return fd;
839 most_interesting_errno = errno;
840
841 prepare_alt_odb();
842 for (alt = alt_odb_list; alt; alt = alt->next) {
843 *path = alt_sha1_path(alt, sha1);
844 fd = git_open(*path);
845 if (fd >= 0)
846 return fd;
847 if (most_interesting_errno == ENOENT)
848 most_interesting_errno = errno;
849 }
850 errno = most_interesting_errno;
851 return -1;
852}
853
854/*
855 * Map the loose object at "path" if it is not NULL, or the path found by
856 * searching for a loose object named "sha1".
857 */
858static void *map_sha1_file_1(const char *path,
859 const unsigned char *sha1,
860 unsigned long *size)
861{
862 void *map;
863 int fd;
864
865 if (path)
866 fd = git_open(path);
867 else
868 fd = open_sha1_file(sha1, &path);
869 map = NULL;
870 if (fd >= 0) {
871 struct stat st;
872
873 if (!fstat(fd, &st)) {
874 *size = xsize_t(st.st_size);
875 if (!*size) {
876 /* mmap() is forbidden on empty files */
877 error("object file %s is empty", path);
878 return NULL;
879 }
880 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
881 }
882 close(fd);
883 }
884 return map;
885}
886
887void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
888{
889 return map_sha1_file_1(NULL, sha1, size);
890}
891
892static int unpack_sha1_short_header(git_zstream *stream,
893 unsigned char *map, unsigned long mapsize,
894 void *buffer, unsigned long bufsiz)
895{
896 /* Get the data stream */
897 memset(stream, 0, sizeof(*stream));
898 stream->next_in = map;
899 stream->avail_in = mapsize;
900 stream->next_out = buffer;
901 stream->avail_out = bufsiz;
902
903 git_inflate_init(stream);
904 return git_inflate(stream, 0);
905}
906
907int unpack_sha1_header(git_zstream *stream,
908 unsigned char *map, unsigned long mapsize,
909 void *buffer, unsigned long bufsiz)
910{
911 int status = unpack_sha1_short_header(stream, map, mapsize,
912 buffer, bufsiz);
913
914 if (status < Z_OK)
915 return status;
916
917 /* Make sure we have the terminating NUL */
918 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
919 return -1;
920 return 0;
921}
922
923static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
924 unsigned long mapsize, void *buffer,
925 unsigned long bufsiz, struct strbuf *header)
926{
927 int status;
928
929 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
930 if (status < Z_OK)
931 return -1;
932
933 /*
934 * Check if entire header is unpacked in the first iteration.
935 */
936 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
937 return 0;
938
939 /*
940 * buffer[0..bufsiz] was not large enough. Copy the partial
941 * result out to header, and then append the result of further
942 * reading the stream.
943 */
944 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
945 stream->next_out = buffer;
946 stream->avail_out = bufsiz;
947
948 do {
949 status = git_inflate(stream, 0);
950 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
951 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
952 return 0;
953 stream->next_out = buffer;
954 stream->avail_out = bufsiz;
955 } while (status != Z_STREAM_END);
956 return -1;
957}
958
959static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
960{
961 int bytes = strlen(buffer) + 1;
962 unsigned char *buf = xmallocz(size);
963 unsigned long n;
964 int status = Z_OK;
965
966 n = stream->total_out - bytes;
967 if (n > size)
968 n = size;
969 memcpy(buf, (char *) buffer + bytes, n);
970 bytes = n;
971 if (bytes <= size) {
972 /*
973 * The above condition must be (bytes <= size), not
974 * (bytes < size). In other words, even though we
975 * expect no more output and set avail_out to zero,
976 * the input zlib stream may have bytes that express
977 * "this concludes the stream", and we *do* want to
978 * eat that input.
979 *
980 * Otherwise we would not be able to test that we
981 * consumed all the input to reach the expected size;
982 * we also want to check that zlib tells us that all
983 * went well with status == Z_STREAM_END at the end.
984 */
985 stream->next_out = buf + bytes;
986 stream->avail_out = size - bytes;
987 while (status == Z_OK)
988 status = git_inflate(stream, Z_FINISH);
989 }
990 if (status == Z_STREAM_END && !stream->avail_in) {
991 git_inflate_end(stream);
992 return buf;
993 }
994
995 if (status < 0)
996 error("corrupt loose object '%s'", sha1_to_hex(sha1));
997 else if (stream->avail_in)
998 error("garbage at end of loose object '%s'",
999 sha1_to_hex(sha1));
1000 free(buf);
1001 return NULL;
1002}
1003
1004/*
1005 * We used to just use "sscanf()", but that's actually way
1006 * too permissive for what we want to check. So do an anal
1007 * object header parse by hand.
1008 */
1009static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1010 unsigned int flags)
1011{
1012 const char *type_buf = hdr;
1013 unsigned long size;
1014 int type, type_len = 0;
1015
1016 /*
1017 * The type can be of any size but is followed by
1018 * a space.
1019 */
1020 for (;;) {
1021 char c = *hdr++;
1022 if (!c)
1023 return -1;
1024 if (c == ' ')
1025 break;
1026 type_len++;
1027 }
1028
1029 type = type_from_string_gently(type_buf, type_len, 1);
1030 if (oi->typename)
1031 strbuf_add(oi->typename, type_buf, type_len);
1032 /*
1033 * Set type to 0 if its an unknown object and
1034 * we're obtaining the type using '--allow-unknown-type'
1035 * option.
1036 */
1037 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1038 type = 0;
1039 else if (type < 0)
1040 die("invalid object type");
1041 if (oi->typep)
1042 *oi->typep = type;
1043
1044 /*
1045 * The length must follow immediately, and be in canonical
1046 * decimal format (ie "010" is not valid).
1047 */
1048 size = *hdr++ - '0';
1049 if (size > 9)
1050 return -1;
1051 if (size) {
1052 for (;;) {
1053 unsigned long c = *hdr - '0';
1054 if (c > 9)
1055 break;
1056 hdr++;
1057 size = size * 10 + c;
1058 }
1059 }
1060
1061 if (oi->sizep)
1062 *oi->sizep = size;
1063
1064 /*
1065 * The length must be followed by a zero byte
1066 */
1067 return *hdr ? -1 : type;
1068}
1069
1070int parse_sha1_header(const char *hdr, unsigned long *sizep)
1071{
1072 struct object_info oi = OBJECT_INFO_INIT;
1073
1074 oi.sizep = sizep;
1075 return parse_sha1_header_extended(hdr, &oi, 0);
1076}
1077
1078static int sha1_loose_object_info(const unsigned char *sha1,
1079 struct object_info *oi,
1080 int flags)
1081{
1082 int status = 0;
1083 unsigned long mapsize;
1084 void *map;
1085 git_zstream stream;
1086 char hdr[32];
1087 struct strbuf hdrbuf = STRBUF_INIT;
1088 unsigned long size_scratch;
1089
1090 if (oi->delta_base_sha1)
1091 hashclr(oi->delta_base_sha1);
1092
1093 /*
1094 * If we don't care about type or size, then we don't
1095 * need to look inside the object at all. Note that we
1096 * do not optimize out the stat call, even if the
1097 * caller doesn't care about the disk-size, since our
1098 * return value implicitly indicates whether the
1099 * object even exists.
1100 */
1101 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1102 const char *path;
1103 struct stat st;
1104 if (stat_sha1_file(sha1, &st, &path) < 0)
1105 return -1;
1106 if (oi->disk_sizep)
1107 *oi->disk_sizep = st.st_size;
1108 return 0;
1109 }
1110
1111 map = map_sha1_file(sha1, &mapsize);
1112 if (!map)
1113 return -1;
1114
1115 if (!oi->sizep)
1116 oi->sizep = &size_scratch;
1117
1118 if (oi->disk_sizep)
1119 *oi->disk_sizep = mapsize;
1120 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1121 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1122 status = error("unable to unpack %s header with --allow-unknown-type",
1123 sha1_to_hex(sha1));
1124 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1125 status = error("unable to unpack %s header",
1126 sha1_to_hex(sha1));
1127 if (status < 0)
1128 ; /* Do nothing */
1129 else if (hdrbuf.len) {
1130 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1131 status = error("unable to parse %s header with --allow-unknown-type",
1132 sha1_to_hex(sha1));
1133 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1134 status = error("unable to parse %s header", sha1_to_hex(sha1));
1135
1136 if (status >= 0 && oi->contentp)
1137 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1138 *oi->sizep, sha1);
1139 else
1140 git_inflate_end(&stream);
1141
1142 munmap(map, mapsize);
1143 if (status && oi->typep)
1144 *oi->typep = status;
1145 if (oi->sizep == &size_scratch)
1146 oi->sizep = NULL;
1147 strbuf_release(&hdrbuf);
1148 oi->whence = OI_LOOSE;
1149 return (status < 0) ? status : 0;
1150}
1151
1152int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1153{
1154 static struct object_info blank_oi = OBJECT_INFO_INIT;
1155 struct pack_entry e;
1156 int rtype;
1157 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1158 lookup_replace_object(sha1) :
1159 sha1;
1160
1161 if (!oi)
1162 oi = &blank_oi;
1163
1164 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1165 struct cached_object *co = find_cached_object(real);
1166 if (co) {
1167 if (oi->typep)
1168 *(oi->typep) = co->type;
1169 if (oi->sizep)
1170 *(oi->sizep) = co->size;
1171 if (oi->disk_sizep)
1172 *(oi->disk_sizep) = 0;
1173 if (oi->delta_base_sha1)
1174 hashclr(oi->delta_base_sha1);
1175 if (oi->typename)
1176 strbuf_addstr(oi->typename, typename(co->type));
1177 if (oi->contentp)
1178 *oi->contentp = xmemdupz(co->buf, co->size);
1179 oi->whence = OI_CACHED;
1180 return 0;
1181 }
1182 }
1183
1184 if (!find_pack_entry(real, &e)) {
1185 /* Most likely it's a loose object. */
1186 if (!sha1_loose_object_info(real, oi, flags))
1187 return 0;
1188
1189 /* Not a loose object; someone else may have just packed it. */
1190 if (flags & OBJECT_INFO_QUICK) {
1191 return -1;
1192 } else {
1193 reprepare_packed_git();
1194 if (!find_pack_entry(real, &e))
1195 return -1;
1196 }
1197 }
1198
1199 if (oi == &blank_oi)
1200 /*
1201 * We know that the caller doesn't actually need the
1202 * information below, so return early.
1203 */
1204 return 0;
1205
1206 rtype = packed_object_info(e.p, e.offset, oi);
1207 if (rtype < 0) {
1208 mark_bad_packed_object(e.p, real);
1209 return sha1_object_info_extended(real, oi, 0);
1210 } else if (oi->whence == OI_PACKED) {
1211 oi->u.packed.offset = e.offset;
1212 oi->u.packed.pack = e.p;
1213 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1214 rtype == OBJ_OFS_DELTA);
1215 }
1216
1217 return 0;
1218}
1219
1220/* returns enum object_type or negative */
1221int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1222{
1223 enum object_type type;
1224 struct object_info oi = OBJECT_INFO_INIT;
1225
1226 oi.typep = &type;
1227 oi.sizep = sizep;
1228 if (sha1_object_info_extended(sha1, &oi,
1229 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1230 return -1;
1231 return type;
1232}
1233
1234static void *read_object(const unsigned char *sha1, enum object_type *type,
1235 unsigned long *size)
1236{
1237 struct object_info oi = OBJECT_INFO_INIT;
1238 void *content;
1239 oi.typep = type;
1240 oi.sizep = size;
1241 oi.contentp = &content;
1242
1243 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1244 return NULL;
1245 return content;
1246}
1247
1248int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1249 unsigned char *sha1)
1250{
1251 struct cached_object *co;
1252
1253 hash_sha1_file(buf, len, typename(type), sha1);
1254 if (has_sha1_file(sha1) || find_cached_object(sha1))
1255 return 0;
1256 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1257 co = &cached_objects[cached_object_nr++];
1258 co->size = len;
1259 co->type = type;
1260 co->buf = xmalloc(len);
1261 memcpy(co->buf, buf, len);
1262 hashcpy(co->sha1, sha1);
1263 return 0;
1264}
1265
1266/*
1267 * This function dies on corrupt objects; the callers who want to
1268 * deal with them should arrange to call read_object() and give error
1269 * messages themselves.
1270 */
1271void *read_sha1_file_extended(const unsigned char *sha1,
1272 enum object_type *type,
1273 unsigned long *size,
1274 int lookup_replace)
1275{
1276 void *data;
1277 const struct packed_git *p;
1278 const char *path;
1279 struct stat st;
1280 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1281 : sha1;
1282
1283 errno = 0;
1284 data = read_object(repl, type, size);
1285 if (data)
1286 return data;
1287
1288 if (errno && errno != ENOENT)
1289 die_errno("failed to read object %s", sha1_to_hex(sha1));
1290
1291 /* die if we replaced an object with one that does not exist */
1292 if (repl != sha1)
1293 die("replacement %s not found for %s",
1294 sha1_to_hex(repl), sha1_to_hex(sha1));
1295
1296 if (!stat_sha1_file(repl, &st, &path))
1297 die("loose object %s (stored in %s) is corrupt",
1298 sha1_to_hex(repl), path);
1299
1300 if ((p = has_packed_and_bad(repl)) != NULL)
1301 die("packed object %s (stored in %s) is corrupt",
1302 sha1_to_hex(repl), p->pack_name);
1303
1304 return NULL;
1305}
1306
1307void *read_object_with_reference(const unsigned char *sha1,
1308 const char *required_type_name,
1309 unsigned long *size,
1310 unsigned char *actual_sha1_return)
1311{
1312 enum object_type type, required_type;
1313 void *buffer;
1314 unsigned long isize;
1315 unsigned char actual_sha1[20];
1316
1317 required_type = type_from_string(required_type_name);
1318 hashcpy(actual_sha1, sha1);
1319 while (1) {
1320 int ref_length = -1;
1321 const char *ref_type = NULL;
1322
1323 buffer = read_sha1_file(actual_sha1, &type, &isize);
1324 if (!buffer)
1325 return NULL;
1326 if (type == required_type) {
1327 *size = isize;
1328 if (actual_sha1_return)
1329 hashcpy(actual_sha1_return, actual_sha1);
1330 return buffer;
1331 }
1332 /* Handle references */
1333 else if (type == OBJ_COMMIT)
1334 ref_type = "tree ";
1335 else if (type == OBJ_TAG)
1336 ref_type = "object ";
1337 else {
1338 free(buffer);
1339 return NULL;
1340 }
1341 ref_length = strlen(ref_type);
1342
1343 if (ref_length + 40 > isize ||
1344 memcmp(buffer, ref_type, ref_length) ||
1345 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1346 free(buffer);
1347 return NULL;
1348 }
1349 free(buffer);
1350 /* Now we have the ID of the referred-to object in
1351 * actual_sha1. Check again. */
1352 }
1353}
1354
1355static void write_sha1_file_prepare(const void *buf, unsigned long len,
1356 const char *type, unsigned char *sha1,
1357 char *hdr, int *hdrlen)
1358{
1359 git_SHA_CTX c;
1360
1361 /* Generate the header */
1362 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1363
1364 /* Sha1.. */
1365 git_SHA1_Init(&c);
1366 git_SHA1_Update(&c, hdr, *hdrlen);
1367 git_SHA1_Update(&c, buf, len);
1368 git_SHA1_Final(sha1, &c);
1369}
1370
1371/*
1372 * Move the just written object into its final resting place.
1373 */
1374int finalize_object_file(const char *tmpfile, const char *filename)
1375{
1376 int ret = 0;
1377
1378 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1379 goto try_rename;
1380 else if (link(tmpfile, filename))
1381 ret = errno;
1382
1383 /*
1384 * Coda hack - coda doesn't like cross-directory links,
1385 * so we fall back to a rename, which will mean that it
1386 * won't be able to check collisions, but that's not a
1387 * big deal.
1388 *
1389 * The same holds for FAT formatted media.
1390 *
1391 * When this succeeds, we just return. We have nothing
1392 * left to unlink.
1393 */
1394 if (ret && ret != EEXIST) {
1395 try_rename:
1396 if (!rename(tmpfile, filename))
1397 goto out;
1398 ret = errno;
1399 }
1400 unlink_or_warn(tmpfile);
1401 if (ret) {
1402 if (ret != EEXIST) {
1403 return error_errno("unable to write sha1 filename %s", filename);
1404 }
1405 /* FIXME!!! Collision check here ? */
1406 }
1407
1408out:
1409 if (adjust_shared_perm(filename))
1410 return error("unable to set permission to '%s'", filename);
1411 return 0;
1412}
1413
1414static int write_buffer(int fd, const void *buf, size_t len)
1415{
1416 if (write_in_full(fd, buf, len) < 0)
1417 return error_errno("file write error");
1418 return 0;
1419}
1420
1421int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1422 unsigned char *sha1)
1423{
1424 char hdr[32];
1425 int hdrlen = sizeof(hdr);
1426 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1427 return 0;
1428}
1429
1430/* Finalize a file on disk, and close it. */
1431static void close_sha1_file(int fd)
1432{
1433 if (fsync_object_files)
1434 fsync_or_die(fd, "sha1 file");
1435 if (close(fd) != 0)
1436 die_errno("error when closing sha1 file");
1437}
1438
1439/* Size of directory component, including the ending '/' */
1440static inline int directory_size(const char *filename)
1441{
1442 const char *s = strrchr(filename, '/');
1443 if (!s)
1444 return 0;
1445 return s - filename + 1;
1446}
1447
1448/*
1449 * This creates a temporary file in the same directory as the final
1450 * 'filename'
1451 *
1452 * We want to avoid cross-directory filename renames, because those
1453 * can have problems on various filesystems (FAT, NFS, Coda).
1454 */
1455static int create_tmpfile(struct strbuf *tmp, const char *filename)
1456{
1457 int fd, dirlen = directory_size(filename);
1458
1459 strbuf_reset(tmp);
1460 strbuf_add(tmp, filename, dirlen);
1461 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1462 fd = git_mkstemp_mode(tmp->buf, 0444);
1463 if (fd < 0 && dirlen && errno == ENOENT) {
1464 /*
1465 * Make sure the directory exists; note that the contents
1466 * of the buffer are undefined after mkstemp returns an
1467 * error, so we have to rewrite the whole buffer from
1468 * scratch.
1469 */
1470 strbuf_reset(tmp);
1471 strbuf_add(tmp, filename, dirlen - 1);
1472 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1473 return -1;
1474 if (adjust_shared_perm(tmp->buf))
1475 return -1;
1476
1477 /* Try again */
1478 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1479 fd = git_mkstemp_mode(tmp->buf, 0444);
1480 }
1481 return fd;
1482}
1483
1484static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1485 const void *buf, unsigned long len, time_t mtime)
1486{
1487 int fd, ret;
1488 unsigned char compressed[4096];
1489 git_zstream stream;
1490 git_SHA_CTX c;
1491 unsigned char parano_sha1[20];
1492 static struct strbuf tmp_file = STRBUF_INIT;
1493 const char *filename = sha1_file_name(sha1);
1494
1495 fd = create_tmpfile(&tmp_file, filename);
1496 if (fd < 0) {
1497 if (errno == EACCES)
1498 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1499 else
1500 return error_errno("unable to create temporary file");
1501 }
1502
1503 /* Set it up */
1504 git_deflate_init(&stream, zlib_compression_level);
1505 stream.next_out = compressed;
1506 stream.avail_out = sizeof(compressed);
1507 git_SHA1_Init(&c);
1508
1509 /* First header.. */
1510 stream.next_in = (unsigned char *)hdr;
1511 stream.avail_in = hdrlen;
1512 while (git_deflate(&stream, 0) == Z_OK)
1513 ; /* nothing */
1514 git_SHA1_Update(&c, hdr, hdrlen);
1515
1516 /* Then the data itself.. */
1517 stream.next_in = (void *)buf;
1518 stream.avail_in = len;
1519 do {
1520 unsigned char *in0 = stream.next_in;
1521 ret = git_deflate(&stream, Z_FINISH);
1522 git_SHA1_Update(&c, in0, stream.next_in - in0);
1523 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1524 die("unable to write sha1 file");
1525 stream.next_out = compressed;
1526 stream.avail_out = sizeof(compressed);
1527 } while (ret == Z_OK);
1528
1529 if (ret != Z_STREAM_END)
1530 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1531 ret = git_deflate_end_gently(&stream);
1532 if (ret != Z_OK)
1533 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1534 git_SHA1_Final(parano_sha1, &c);
1535 if (hashcmp(sha1, parano_sha1) != 0)
1536 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1537
1538 close_sha1_file(fd);
1539
1540 if (mtime) {
1541 struct utimbuf utb;
1542 utb.actime = mtime;
1543 utb.modtime = mtime;
1544 if (utime(tmp_file.buf, &utb) < 0)
1545 warning_errno("failed utime() on %s", tmp_file.buf);
1546 }
1547
1548 return finalize_object_file(tmp_file.buf, filename);
1549}
1550
1551static int freshen_loose_object(const unsigned char *sha1)
1552{
1553 return check_and_freshen(sha1, 1);
1554}
1555
1556static int freshen_packed_object(const unsigned char *sha1)
1557{
1558 struct pack_entry e;
1559 if (!find_pack_entry(sha1, &e))
1560 return 0;
1561 if (e.p->freshened)
1562 return 1;
1563 if (!freshen_file(e.p->pack_name))
1564 return 0;
1565 e.p->freshened = 1;
1566 return 1;
1567}
1568
1569int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1570{
1571 char hdr[32];
1572 int hdrlen = sizeof(hdr);
1573
1574 /* Normally if we have it in the pack then we do not bother writing
1575 * it out into .git/objects/??/?{38} file.
1576 */
1577 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1578 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1579 return 0;
1580 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1581}
1582
1583int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1584 struct object_id *oid, unsigned flags)
1585{
1586 char *header;
1587 int hdrlen, status = 0;
1588
1589 /* type string, SP, %lu of the length plus NUL must fit this */
1590 hdrlen = strlen(type) + 32;
1591 header = xmalloc(hdrlen);
1592 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1593
1594 if (!(flags & HASH_WRITE_OBJECT))
1595 goto cleanup;
1596 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1597 goto cleanup;
1598 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1599
1600cleanup:
1601 free(header);
1602 return status;
1603}
1604
1605int force_object_loose(const unsigned char *sha1, time_t mtime)
1606{
1607 void *buf;
1608 unsigned long len;
1609 enum object_type type;
1610 char hdr[32];
1611 int hdrlen;
1612 int ret;
1613
1614 if (has_loose_object(sha1))
1615 return 0;
1616 buf = read_object(sha1, &type, &len);
1617 if (!buf)
1618 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1619 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1620 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1621 free(buf);
1622
1623 return ret;
1624}
1625
1626int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1627{
1628 if (!startup_info->have_repository)
1629 return 0;
1630 return sha1_object_info_extended(sha1, NULL,
1631 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1632}
1633
1634int has_object_file(const struct object_id *oid)
1635{
1636 return has_sha1_file(oid->hash);
1637}
1638
1639int has_object_file_with_flags(const struct object_id *oid, int flags)
1640{
1641 return has_sha1_file_with_flags(oid->hash, flags);
1642}
1643
1644static void check_tree(const void *buf, size_t size)
1645{
1646 struct tree_desc desc;
1647 struct name_entry entry;
1648
1649 init_tree_desc(&desc, buf, size);
1650 while (tree_entry(&desc, &entry))
1651 /* do nothing
1652 * tree_entry() will die() on malformed entries */
1653 ;
1654}
1655
1656static void check_commit(const void *buf, size_t size)
1657{
1658 struct commit c;
1659 memset(&c, 0, sizeof(c));
1660 if (parse_commit_buffer(&c, buf, size))
1661 die("corrupt commit");
1662}
1663
1664static void check_tag(const void *buf, size_t size)
1665{
1666 struct tag t;
1667 memset(&t, 0, sizeof(t));
1668 if (parse_tag_buffer(&t, buf, size))
1669 die("corrupt tag");
1670}
1671
1672static int index_mem(unsigned char *sha1, void *buf, size_t size,
1673 enum object_type type,
1674 const char *path, unsigned flags)
1675{
1676 int ret, re_allocated = 0;
1677 int write_object = flags & HASH_WRITE_OBJECT;
1678
1679 if (!type)
1680 type = OBJ_BLOB;
1681
1682 /*
1683 * Convert blobs to git internal format
1684 */
1685 if ((type == OBJ_BLOB) && path) {
1686 struct strbuf nbuf = STRBUF_INIT;
1687 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1688 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1689 buf = strbuf_detach(&nbuf, &size);
1690 re_allocated = 1;
1691 }
1692 }
1693 if (flags & HASH_FORMAT_CHECK) {
1694 if (type == OBJ_TREE)
1695 check_tree(buf, size);
1696 if (type == OBJ_COMMIT)
1697 check_commit(buf, size);
1698 if (type == OBJ_TAG)
1699 check_tag(buf, size);
1700 }
1701
1702 if (write_object)
1703 ret = write_sha1_file(buf, size, typename(type), sha1);
1704 else
1705 ret = hash_sha1_file(buf, size, typename(type), sha1);
1706 if (re_allocated)
1707 free(buf);
1708 return ret;
1709}
1710
1711static int index_stream_convert_blob(unsigned char *sha1, int fd,
1712 const char *path, unsigned flags)
1713{
1714 int ret;
1715 const int write_object = flags & HASH_WRITE_OBJECT;
1716 struct strbuf sbuf = STRBUF_INIT;
1717
1718 assert(path);
1719 assert(would_convert_to_git_filter_fd(path));
1720
1721 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1722 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1723
1724 if (write_object)
1725 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1726 sha1);
1727 else
1728 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1729 sha1);
1730 strbuf_release(&sbuf);
1731 return ret;
1732}
1733
1734static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1735 const char *path, unsigned flags)
1736{
1737 struct strbuf sbuf = STRBUF_INIT;
1738 int ret;
1739
1740 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1741 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1742 else
1743 ret = -1;
1744 strbuf_release(&sbuf);
1745 return ret;
1746}
1747
1748#define SMALL_FILE_SIZE (32*1024)
1749
1750static int index_core(unsigned char *sha1, int fd, size_t size,
1751 enum object_type type, const char *path,
1752 unsigned flags)
1753{
1754 int ret;
1755
1756 if (!size) {
1757 ret = index_mem(sha1, "", size, type, path, flags);
1758 } else if (size <= SMALL_FILE_SIZE) {
1759 char *buf = xmalloc(size);
1760 if (size == read_in_full(fd, buf, size))
1761 ret = index_mem(sha1, buf, size, type, path, flags);
1762 else
1763 ret = error_errno("short read");
1764 free(buf);
1765 } else {
1766 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1767 ret = index_mem(sha1, buf, size, type, path, flags);
1768 munmap(buf, size);
1769 }
1770 return ret;
1771}
1772
1773/*
1774 * This creates one packfile per large blob unless bulk-checkin
1775 * machinery is "plugged".
1776 *
1777 * This also bypasses the usual "convert-to-git" dance, and that is on
1778 * purpose. We could write a streaming version of the converting
1779 * functions and insert that before feeding the data to fast-import
1780 * (or equivalent in-core API described above). However, that is
1781 * somewhat complicated, as we do not know the size of the filter
1782 * result, which we need to know beforehand when writing a git object.
1783 * Since the primary motivation for trying to stream from the working
1784 * tree file and to avoid mmaping it in core is to deal with large
1785 * binary blobs, they generally do not want to get any conversion, and
1786 * callers should avoid this code path when filters are requested.
1787 */
1788static int index_stream(struct object_id *oid, int fd, size_t size,
1789 enum object_type type, const char *path,
1790 unsigned flags)
1791{
1792 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1793}
1794
1795int index_fd(struct object_id *oid, int fd, struct stat *st,
1796 enum object_type type, const char *path, unsigned flags)
1797{
1798 int ret;
1799
1800 /*
1801 * Call xsize_t() only when needed to avoid potentially unnecessary
1802 * die() for large files.
1803 */
1804 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1805 ret = index_stream_convert_blob(oid->hash, fd, path, flags);
1806 else if (!S_ISREG(st->st_mode))
1807 ret = index_pipe(oid->hash, fd, type, path, flags);
1808 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1809 (path && would_convert_to_git(&the_index, path)))
1810 ret = index_core(oid->hash, fd, xsize_t(st->st_size), type, path,
1811 flags);
1812 else
1813 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1814 flags);
1815 close(fd);
1816 return ret;
1817}
1818
1819int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1820{
1821 int fd;
1822 struct strbuf sb = STRBUF_INIT;
1823
1824 switch (st->st_mode & S_IFMT) {
1825 case S_IFREG:
1826 fd = open(path, O_RDONLY);
1827 if (fd < 0)
1828 return error_errno("open(\"%s\")", path);
1829 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1830 return error("%s: failed to insert into database",
1831 path);
1832 break;
1833 case S_IFLNK:
1834 if (strbuf_readlink(&sb, path, st->st_size))
1835 return error_errno("readlink(\"%s\")", path);
1836 if (!(flags & HASH_WRITE_OBJECT))
1837 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1838 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1839 return error("%s: failed to insert into database",
1840 path);
1841 strbuf_release(&sb);
1842 break;
1843 case S_IFDIR:
1844 return resolve_gitlink_ref(path, "HEAD", oid->hash);
1845 default:
1846 return error("%s: unsupported file type", path);
1847 }
1848 return 0;
1849}
1850
1851int read_pack_header(int fd, struct pack_header *header)
1852{
1853 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
1854 /* "eof before pack header was fully read" */
1855 return PH_ERROR_EOF;
1856
1857 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1858 /* "protocol error (pack signature mismatch detected)" */
1859 return PH_ERROR_PACK_SIGNATURE;
1860 if (!pack_version_ok(header->hdr_version))
1861 /* "protocol error (pack version unsupported)" */
1862 return PH_ERROR_PROTOCOL;
1863 return 0;
1864}
1865
1866void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1867{
1868 enum object_type type = sha1_object_info(sha1, NULL);
1869 if (type < 0)
1870 die("%s is not a valid object", sha1_to_hex(sha1));
1871 if (type != expect)
1872 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1873 typename(expect));
1874}
1875
1876int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1877 struct strbuf *path,
1878 each_loose_object_fn obj_cb,
1879 each_loose_cruft_fn cruft_cb,
1880 each_loose_subdir_fn subdir_cb,
1881 void *data)
1882{
1883 size_t origlen, baselen;
1884 DIR *dir;
1885 struct dirent *de;
1886 int r = 0;
1887
1888 if (subdir_nr > 0xff)
1889 BUG("invalid loose object subdirectory: %x", subdir_nr);
1890
1891 origlen = path->len;
1892 strbuf_complete(path, '/');
1893 strbuf_addf(path, "%02x", subdir_nr);
1894 baselen = path->len;
1895
1896 dir = opendir(path->buf);
1897 if (!dir) {
1898 if (errno != ENOENT)
1899 r = error_errno("unable to open %s", path->buf);
1900 strbuf_setlen(path, origlen);
1901 return r;
1902 }
1903
1904 while ((de = readdir(dir))) {
1905 if (is_dot_or_dotdot(de->d_name))
1906 continue;
1907
1908 strbuf_setlen(path, baselen);
1909 strbuf_addf(path, "/%s", de->d_name);
1910
1911 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1912 char hex[GIT_MAX_HEXSZ+1];
1913 struct object_id oid;
1914
1915 xsnprintf(hex, sizeof(hex), "%02x%s",
1916 subdir_nr, de->d_name);
1917 if (!get_oid_hex(hex, &oid)) {
1918 if (obj_cb) {
1919 r = obj_cb(&oid, path->buf, data);
1920 if (r)
1921 break;
1922 }
1923 continue;
1924 }
1925 }
1926
1927 if (cruft_cb) {
1928 r = cruft_cb(de->d_name, path->buf, data);
1929 if (r)
1930 break;
1931 }
1932 }
1933 closedir(dir);
1934
1935 strbuf_setlen(path, baselen);
1936 if (!r && subdir_cb)
1937 r = subdir_cb(subdir_nr, path->buf, data);
1938
1939 strbuf_setlen(path, origlen);
1940
1941 return r;
1942}
1943
1944int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1945 each_loose_object_fn obj_cb,
1946 each_loose_cruft_fn cruft_cb,
1947 each_loose_subdir_fn subdir_cb,
1948 void *data)
1949{
1950 int r = 0;
1951 int i;
1952
1953 for (i = 0; i < 256; i++) {
1954 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1955 subdir_cb, data);
1956 if (r)
1957 break;
1958 }
1959
1960 return r;
1961}
1962
1963int for_each_loose_file_in_objdir(const char *path,
1964 each_loose_object_fn obj_cb,
1965 each_loose_cruft_fn cruft_cb,
1966 each_loose_subdir_fn subdir_cb,
1967 void *data)
1968{
1969 struct strbuf buf = STRBUF_INIT;
1970 int r;
1971
1972 strbuf_addstr(&buf, path);
1973 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1974 subdir_cb, data);
1975 strbuf_release(&buf);
1976
1977 return r;
1978}
1979
1980struct loose_alt_odb_data {
1981 each_loose_object_fn *cb;
1982 void *data;
1983};
1984
1985static int loose_from_alt_odb(struct alternate_object_database *alt,
1986 void *vdata)
1987{
1988 struct loose_alt_odb_data *data = vdata;
1989 struct strbuf buf = STRBUF_INIT;
1990 int r;
1991
1992 strbuf_addstr(&buf, alt->path);
1993 r = for_each_loose_file_in_objdir_buf(&buf,
1994 data->cb, NULL, NULL,
1995 data->data);
1996 strbuf_release(&buf);
1997 return r;
1998}
1999
2000int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2001{
2002 struct loose_alt_odb_data alt;
2003 int r;
2004
2005 r = for_each_loose_file_in_objdir(get_object_directory(),
2006 cb, NULL, NULL, data);
2007 if (r)
2008 return r;
2009
2010 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2011 return 0;
2012
2013 alt.cb = cb;
2014 alt.data = data;
2015 return foreach_alt_odb(loose_from_alt_odb, &alt);
2016}
2017
2018static int check_stream_sha1(git_zstream *stream,
2019 const char *hdr,
2020 unsigned long size,
2021 const char *path,
2022 const unsigned char *expected_sha1)
2023{
2024 git_SHA_CTX c;
2025 unsigned char real_sha1[GIT_MAX_RAWSZ];
2026 unsigned char buf[4096];
2027 unsigned long total_read;
2028 int status = Z_OK;
2029
2030 git_SHA1_Init(&c);
2031 git_SHA1_Update(&c, hdr, stream->total_out);
2032
2033 /*
2034 * We already read some bytes into hdr, but the ones up to the NUL
2035 * do not count against the object's content size.
2036 */
2037 total_read = stream->total_out - strlen(hdr) - 1;
2038
2039 /*
2040 * This size comparison must be "<=" to read the final zlib packets;
2041 * see the comment in unpack_sha1_rest for details.
2042 */
2043 while (total_read <= size &&
2044 (status == Z_OK || status == Z_BUF_ERROR)) {
2045 stream->next_out = buf;
2046 stream->avail_out = sizeof(buf);
2047 if (size - total_read < stream->avail_out)
2048 stream->avail_out = size - total_read;
2049 status = git_inflate(stream, Z_FINISH);
2050 git_SHA1_Update(&c, buf, stream->next_out - buf);
2051 total_read += stream->next_out - buf;
2052 }
2053 git_inflate_end(stream);
2054
2055 if (status != Z_STREAM_END) {
2056 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2057 return -1;
2058 }
2059 if (stream->avail_in) {
2060 error("garbage at end of loose object '%s'",
2061 sha1_to_hex(expected_sha1));
2062 return -1;
2063 }
2064
2065 git_SHA1_Final(real_sha1, &c);
2066 if (hashcmp(expected_sha1, real_sha1)) {
2067 error("sha1 mismatch for %s (expected %s)", path,
2068 sha1_to_hex(expected_sha1));
2069 return -1;
2070 }
2071
2072 return 0;
2073}
2074
2075int read_loose_object(const char *path,
2076 const unsigned char *expected_sha1,
2077 enum object_type *type,
2078 unsigned long *size,
2079 void **contents)
2080{
2081 int ret = -1;
2082 void *map = NULL;
2083 unsigned long mapsize;
2084 git_zstream stream;
2085 char hdr[32];
2086
2087 *contents = NULL;
2088
2089 map = map_sha1_file_1(path, NULL, &mapsize);
2090 if (!map) {
2091 error_errno("unable to mmap %s", path);
2092 goto out;
2093 }
2094
2095 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2096 error("unable to unpack header of %s", path);
2097 goto out;
2098 }
2099
2100 *type = parse_sha1_header(hdr, size);
2101 if (*type < 0) {
2102 error("unable to parse header of %s", path);
2103 git_inflate_end(&stream);
2104 goto out;
2105 }
2106
2107 if (*type == OBJ_BLOB) {
2108 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2109 goto out;
2110 } else {
2111 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2112 if (!*contents) {
2113 error("unable to unpack contents of %s", path);
2114 git_inflate_end(&stream);
2115 goto out;
2116 }
2117 if (check_sha1_signature(expected_sha1, *contents,
2118 *size, typename(*type))) {
2119 error("sha1 mismatch for %s (expected %s)", path,
2120 sha1_to_hex(expected_sha1));
2121 free(*contents);
2122 goto out;
2123 }
2124 }
2125
2126 ret = 0; /* everything checks out */
2127
2128out:
2129 if (map)
2130 munmap(map, mapsize);
2131 return ret;
2132}