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#include "fetch-object.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
42static void git_hash_sha1_init(git_hash_ctx *ctx)
43{
44 git_SHA1_Init(&ctx->sha1);
45}
46
47static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
48{
49 git_SHA1_Update(&ctx->sha1, data, len);
50}
51
52static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
53{
54 git_SHA1_Final(hash, &ctx->sha1);
55}
56
57static void git_hash_unknown_init(git_hash_ctx *ctx)
58{
59 die("trying to init unknown hash");
60}
61
62static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
63{
64 die("trying to update unknown hash");
65}
66
67static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
68{
69 die("trying to finalize unknown hash");
70}
71
72const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
73 {
74 NULL,
75 0x00000000,
76 0,
77 0,
78 git_hash_unknown_init,
79 git_hash_unknown_update,
80 git_hash_unknown_final,
81 NULL,
82 NULL,
83 },
84 {
85 "sha-1",
86 /* "sha1", big-endian */
87 0x73686131,
88 GIT_SHA1_RAWSZ,
89 GIT_SHA1_HEXSZ,
90 git_hash_sha1_init,
91 git_hash_sha1_update,
92 git_hash_sha1_final,
93 &empty_tree_oid,
94 &empty_blob_oid,
95 },
96};
97
98/*
99 * This is meant to hold a *small* number of objects that you would
100 * want read_sha1_file() to be able to return, but yet you do not want
101 * to write them into the object store (e.g. a browse-only
102 * application).
103 */
104static struct cached_object {
105 unsigned char sha1[20];
106 enum object_type type;
107 void *buf;
108 unsigned long size;
109} *cached_objects;
110static int cached_object_nr, cached_object_alloc;
111
112static struct cached_object empty_tree = {
113 EMPTY_TREE_SHA1_BIN_LITERAL,
114 OBJ_TREE,
115 "",
116 0
117};
118
119static struct cached_object *find_cached_object(const unsigned char *sha1)
120{
121 int i;
122 struct cached_object *co = cached_objects;
123
124 for (i = 0; i < cached_object_nr; i++, co++) {
125 if (!hashcmp(co->sha1, sha1))
126 return co;
127 }
128 if (!hashcmp(sha1, empty_tree.sha1))
129 return &empty_tree;
130 return NULL;
131}
132
133
134static int get_conv_flags(unsigned flags)
135{
136 if (flags & HASH_RENORMALIZE)
137 return CONV_EOL_RENORMALIZE;
138 else if (flags & HASH_WRITE_OBJECT)
139 return global_conv_flags_eol;
140 else
141 return 0;
142}
143
144
145int mkdir_in_gitdir(const char *path)
146{
147 if (mkdir(path, 0777)) {
148 int saved_errno = errno;
149 struct stat st;
150 struct strbuf sb = STRBUF_INIT;
151
152 if (errno != EEXIST)
153 return -1;
154 /*
155 * Are we looking at a path in a symlinked worktree
156 * whose original repository does not yet have it?
157 * e.g. .git/rr-cache pointing at its original
158 * repository in which the user hasn't performed any
159 * conflict resolution yet?
160 */
161 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
162 strbuf_readlink(&sb, path, st.st_size) ||
163 !is_absolute_path(sb.buf) ||
164 mkdir(sb.buf, 0777)) {
165 strbuf_release(&sb);
166 errno = saved_errno;
167 return -1;
168 }
169 strbuf_release(&sb);
170 }
171 return adjust_shared_perm(path);
172}
173
174enum scld_error safe_create_leading_directories(char *path)
175{
176 char *next_component = path + offset_1st_component(path);
177 enum scld_error ret = SCLD_OK;
178
179 while (ret == SCLD_OK && next_component) {
180 struct stat st;
181 char *slash = next_component, slash_character;
182
183 while (*slash && !is_dir_sep(*slash))
184 slash++;
185
186 if (!*slash)
187 break;
188
189 next_component = slash + 1;
190 while (is_dir_sep(*next_component))
191 next_component++;
192 if (!*next_component)
193 break;
194
195 slash_character = *slash;
196 *slash = '\0';
197 if (!stat(path, &st)) {
198 /* path exists */
199 if (!S_ISDIR(st.st_mode)) {
200 errno = ENOTDIR;
201 ret = SCLD_EXISTS;
202 }
203 } else if (mkdir(path, 0777)) {
204 if (errno == EEXIST &&
205 !stat(path, &st) && S_ISDIR(st.st_mode))
206 ; /* somebody created it since we checked */
207 else if (errno == ENOENT)
208 /*
209 * Either mkdir() failed because
210 * somebody just pruned the containing
211 * directory, or stat() failed because
212 * the file that was in our way was
213 * just removed. Either way, inform
214 * the caller that it might be worth
215 * trying again:
216 */
217 ret = SCLD_VANISHED;
218 else
219 ret = SCLD_FAILED;
220 } else if (adjust_shared_perm(path)) {
221 ret = SCLD_PERMS;
222 }
223 *slash = slash_character;
224 }
225 return ret;
226}
227
228enum scld_error safe_create_leading_directories_const(const char *path)
229{
230 int save_errno;
231 /* path points to cache entries, so xstrdup before messing with it */
232 char *buf = xstrdup(path);
233 enum scld_error result = safe_create_leading_directories(buf);
234
235 save_errno = errno;
236 free(buf);
237 errno = save_errno;
238 return result;
239}
240
241int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
242{
243 /*
244 * The number of times we will try to remove empty directories
245 * in the way of path. This is only 1 because if another
246 * process is racily creating directories that conflict with
247 * us, we don't want to fight against them.
248 */
249 int remove_directories_remaining = 1;
250
251 /*
252 * The number of times that we will try to create the
253 * directories containing path. We are willing to attempt this
254 * more than once, because another process could be trying to
255 * clean up empty directories at the same time as we are
256 * trying to create them.
257 */
258 int create_directories_remaining = 3;
259
260 /* A scratch copy of path, filled lazily if we need it: */
261 struct strbuf path_copy = STRBUF_INIT;
262
263 int ret, save_errno;
264
265 /* Sanity check: */
266 assert(*path);
267
268retry_fn:
269 ret = fn(path, cb);
270 save_errno = errno;
271 if (!ret)
272 goto out;
273
274 if (errno == EISDIR && remove_directories_remaining-- > 0) {
275 /*
276 * A directory is in the way. Maybe it is empty; try
277 * to remove it:
278 */
279 if (!path_copy.len)
280 strbuf_addstr(&path_copy, path);
281
282 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
283 goto retry_fn;
284 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
285 /*
286 * Maybe the containing directory didn't exist, or
287 * maybe it was just deleted by a process that is
288 * racing with us to clean up empty directories. Try
289 * to create it:
290 */
291 enum scld_error scld_result;
292
293 if (!path_copy.len)
294 strbuf_addstr(&path_copy, path);
295
296 do {
297 scld_result = safe_create_leading_directories(path_copy.buf);
298 if (scld_result == SCLD_OK)
299 goto retry_fn;
300 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
301 }
302
303out:
304 strbuf_release(&path_copy);
305 errno = save_errno;
306 return ret;
307}
308
309static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
310{
311 int i;
312 for (i = 0; i < 20; i++) {
313 static char hex[] = "0123456789abcdef";
314 unsigned int val = sha1[i];
315 strbuf_addch(buf, hex[val >> 4]);
316 strbuf_addch(buf, hex[val & 0xf]);
317 if (!i)
318 strbuf_addch(buf, '/');
319 }
320}
321
322void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
323{
324 strbuf_addstr(buf, get_object_directory());
325 strbuf_addch(buf, '/');
326 fill_sha1_path(buf, sha1);
327}
328
329struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
330{
331 strbuf_setlen(&alt->scratch, alt->base_len);
332 return &alt->scratch;
333}
334
335static const char *alt_sha1_path(struct alternate_object_database *alt,
336 const unsigned char *sha1)
337{
338 struct strbuf *buf = alt_scratch_buf(alt);
339 fill_sha1_path(buf, sha1);
340 return buf->buf;
341}
342
343struct alternate_object_database *alt_odb_list;
344static struct alternate_object_database **alt_odb_tail;
345
346/*
347 * Return non-zero iff the path is usable as an alternate object database.
348 */
349static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
350{
351 struct alternate_object_database *alt;
352
353 /* Detect cases where alternate disappeared */
354 if (!is_directory(path->buf)) {
355 error("object directory %s does not exist; "
356 "check .git/objects/info/alternates.",
357 path->buf);
358 return 0;
359 }
360
361 /*
362 * Prevent the common mistake of listing the same
363 * thing twice, or object directory itself.
364 */
365 for (alt = alt_odb_list; alt; alt = alt->next) {
366 if (!fspathcmp(path->buf, alt->path))
367 return 0;
368 }
369 if (!fspathcmp(path->buf, normalized_objdir))
370 return 0;
371
372 return 1;
373}
374
375/*
376 * Prepare alternate object database registry.
377 *
378 * The variable alt_odb_list points at the list of struct
379 * alternate_object_database. The elements on this list come from
380 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
381 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
382 * whose contents is similar to that environment variable but can be
383 * LF separated. Its base points at a statically allocated buffer that
384 * contains "/the/directory/corresponding/to/.git/objects/...", while
385 * its name points just after the slash at the end of ".git/objects/"
386 * in the example above, and has enough space to hold 40-byte hex
387 * SHA1, an extra slash for the first level indirection, and the
388 * terminating NUL.
389 */
390static void read_info_alternates(const char * relative_base, int depth);
391static int link_alt_odb_entry(const char *entry, const char *relative_base,
392 int depth, const char *normalized_objdir)
393{
394 struct alternate_object_database *ent;
395 struct strbuf pathbuf = STRBUF_INIT;
396
397 if (!is_absolute_path(entry) && relative_base) {
398 strbuf_realpath(&pathbuf, relative_base, 1);
399 strbuf_addch(&pathbuf, '/');
400 }
401 strbuf_addstr(&pathbuf, entry);
402
403 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
404 error("unable to normalize alternate object path: %s",
405 pathbuf.buf);
406 strbuf_release(&pathbuf);
407 return -1;
408 }
409
410 /*
411 * The trailing slash after the directory name is given by
412 * this function at the end. Remove duplicates.
413 */
414 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
415 strbuf_setlen(&pathbuf, pathbuf.len - 1);
416
417 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
418 strbuf_release(&pathbuf);
419 return -1;
420 }
421
422 ent = alloc_alt_odb(pathbuf.buf);
423
424 /* add the alternate entry */
425 *alt_odb_tail = ent;
426 alt_odb_tail = &(ent->next);
427 ent->next = NULL;
428
429 /* recursively add alternates */
430 read_info_alternates(pathbuf.buf, depth + 1);
431
432 strbuf_release(&pathbuf);
433 return 0;
434}
435
436static const char *parse_alt_odb_entry(const char *string,
437 int sep,
438 struct strbuf *out)
439{
440 const char *end;
441
442 strbuf_reset(out);
443
444 if (*string == '#') {
445 /* comment; consume up to next separator */
446 end = strchrnul(string, sep);
447 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
448 /*
449 * quoted path; unquote_c_style has copied the
450 * data for us and set "end". Broken quoting (e.g.,
451 * an entry that doesn't end with a quote) falls
452 * back to the unquoted case below.
453 */
454 } else {
455 /* normal, unquoted path */
456 end = strchrnul(string, sep);
457 strbuf_add(out, string, end - string);
458 }
459
460 if (*end)
461 end++;
462 return end;
463}
464
465static void link_alt_odb_entries(const char *alt, int sep,
466 const char *relative_base, int depth)
467{
468 struct strbuf objdirbuf = STRBUF_INIT;
469 struct strbuf entry = STRBUF_INIT;
470
471 if (!alt || !*alt)
472 return;
473
474 if (depth > 5) {
475 error("%s: ignoring alternate object stores, nesting too deep.",
476 relative_base);
477 return;
478 }
479
480 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
481 if (strbuf_normalize_path(&objdirbuf) < 0)
482 die("unable to normalize object directory: %s",
483 objdirbuf.buf);
484
485 while (*alt) {
486 alt = parse_alt_odb_entry(alt, sep, &entry);
487 if (!entry.len)
488 continue;
489 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
490 }
491 strbuf_release(&entry);
492 strbuf_release(&objdirbuf);
493}
494
495static void read_info_alternates(const char * relative_base, int depth)
496{
497 char *path;
498 struct strbuf buf = STRBUF_INIT;
499
500 path = xstrfmt("%s/info/alternates", relative_base);
501 if (strbuf_read_file(&buf, path, 1024) < 0) {
502 warn_on_fopen_errors(path);
503 free(path);
504 return;
505 }
506
507 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
508 strbuf_release(&buf);
509 free(path);
510}
511
512struct alternate_object_database *alloc_alt_odb(const char *dir)
513{
514 struct alternate_object_database *ent;
515
516 FLEX_ALLOC_STR(ent, path, dir);
517 strbuf_init(&ent->scratch, 0);
518 strbuf_addf(&ent->scratch, "%s/", dir);
519 ent->base_len = ent->scratch.len;
520
521 return ent;
522}
523
524void add_to_alternates_file(const char *reference)
525{
526 struct lock_file lock = LOCK_INIT;
527 char *alts = git_pathdup("objects/info/alternates");
528 FILE *in, *out;
529 int found = 0;
530
531 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
532 out = fdopen_lock_file(&lock, "w");
533 if (!out)
534 die_errno("unable to fdopen alternates lockfile");
535
536 in = fopen(alts, "r");
537 if (in) {
538 struct strbuf line = STRBUF_INIT;
539
540 while (strbuf_getline(&line, in) != EOF) {
541 if (!strcmp(reference, line.buf)) {
542 found = 1;
543 break;
544 }
545 fprintf_or_die(out, "%s\n", line.buf);
546 }
547
548 strbuf_release(&line);
549 fclose(in);
550 }
551 else if (errno != ENOENT)
552 die_errno("unable to read alternates file");
553
554 if (found) {
555 rollback_lock_file(&lock);
556 } else {
557 fprintf_or_die(out, "%s\n", reference);
558 if (commit_lock_file(&lock))
559 die_errno("unable to move new alternates file into place");
560 if (alt_odb_tail)
561 link_alt_odb_entries(reference, '\n', NULL, 0);
562 }
563 free(alts);
564}
565
566void add_to_alternates_memory(const char *reference)
567{
568 /*
569 * Make sure alternates are initialized, or else our entry may be
570 * overwritten when they are.
571 */
572 prepare_alt_odb();
573
574 link_alt_odb_entries(reference, '\n', NULL, 0);
575}
576
577/*
578 * Compute the exact path an alternate is at and returns it. In case of
579 * error NULL is returned and the human readable error is added to `err`
580 * `path` may be relative and should point to $GITDIR.
581 * `err` must not be null.
582 */
583char *compute_alternate_path(const char *path, struct strbuf *err)
584{
585 char *ref_git = NULL;
586 const char *repo, *ref_git_s;
587 int seen_error = 0;
588
589 ref_git_s = real_path_if_valid(path);
590 if (!ref_git_s) {
591 seen_error = 1;
592 strbuf_addf(err, _("path '%s' does not exist"), path);
593 goto out;
594 } else
595 /*
596 * Beware: read_gitfile(), real_path() and mkpath()
597 * return static buffer
598 */
599 ref_git = xstrdup(ref_git_s);
600
601 repo = read_gitfile(ref_git);
602 if (!repo)
603 repo = read_gitfile(mkpath("%s/.git", ref_git));
604 if (repo) {
605 free(ref_git);
606 ref_git = xstrdup(repo);
607 }
608
609 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
610 char *ref_git_git = mkpathdup("%s/.git", ref_git);
611 free(ref_git);
612 ref_git = ref_git_git;
613 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
614 struct strbuf sb = STRBUF_INIT;
615 seen_error = 1;
616 if (get_common_dir(&sb, ref_git)) {
617 strbuf_addf(err,
618 _("reference repository '%s' as a linked "
619 "checkout is not supported yet."),
620 path);
621 goto out;
622 }
623
624 strbuf_addf(err, _("reference repository '%s' is not a "
625 "local repository."), path);
626 goto out;
627 }
628
629 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
630 strbuf_addf(err, _("reference repository '%s' is shallow"),
631 path);
632 seen_error = 1;
633 goto out;
634 }
635
636 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
637 strbuf_addf(err,
638 _("reference repository '%s' is grafted"),
639 path);
640 seen_error = 1;
641 goto out;
642 }
643
644out:
645 if (seen_error) {
646 FREE_AND_NULL(ref_git);
647 }
648
649 return ref_git;
650}
651
652int foreach_alt_odb(alt_odb_fn fn, void *cb)
653{
654 struct alternate_object_database *ent;
655 int r = 0;
656
657 prepare_alt_odb();
658 for (ent = alt_odb_list; ent; ent = ent->next) {
659 r = fn(ent, cb);
660 if (r)
661 break;
662 }
663 return r;
664}
665
666void prepare_alt_odb(void)
667{
668 if (alt_odb_tail)
669 return;
670
671 alt_odb_tail = &alt_odb_list;
672 link_alt_odb_entries(the_repository->alternate_db, PATH_SEP, NULL, 0);
673
674 read_info_alternates(get_object_directory(), 0);
675}
676
677/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
678static int freshen_file(const char *fn)
679{
680 struct utimbuf t;
681 t.actime = t.modtime = time(NULL);
682 return !utime(fn, &t);
683}
684
685/*
686 * All of the check_and_freshen functions return 1 if the file exists and was
687 * freshened (if freshening was requested), 0 otherwise. If they return
688 * 0, you should not assume that it is safe to skip a write of the object (it
689 * either does not exist on disk, or has a stale mtime and may be subject to
690 * pruning).
691 */
692int check_and_freshen_file(const char *fn, int freshen)
693{
694 if (access(fn, F_OK))
695 return 0;
696 if (freshen && !freshen_file(fn))
697 return 0;
698 return 1;
699}
700
701static int check_and_freshen_local(const unsigned char *sha1, int freshen)
702{
703 static struct strbuf buf = STRBUF_INIT;
704
705 strbuf_reset(&buf);
706 sha1_file_name(&buf, sha1);
707
708 return check_and_freshen_file(buf.buf, freshen);
709}
710
711static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
712{
713 struct alternate_object_database *alt;
714 prepare_alt_odb();
715 for (alt = alt_odb_list; alt; alt = alt->next) {
716 const char *path = alt_sha1_path(alt, sha1);
717 if (check_and_freshen_file(path, freshen))
718 return 1;
719 }
720 return 0;
721}
722
723static int check_and_freshen(const unsigned char *sha1, int freshen)
724{
725 return check_and_freshen_local(sha1, freshen) ||
726 check_and_freshen_nonlocal(sha1, freshen);
727}
728
729int has_loose_object_nonlocal(const unsigned char *sha1)
730{
731 return check_and_freshen_nonlocal(sha1, 0);
732}
733
734static int has_loose_object(const unsigned char *sha1)
735{
736 return check_and_freshen(sha1, 0);
737}
738
739static void mmap_limit_check(size_t length)
740{
741 static size_t limit = 0;
742 if (!limit) {
743 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
744 if (!limit)
745 limit = SIZE_MAX;
746 }
747 if (length > limit)
748 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
749 (uintmax_t)length, (uintmax_t)limit);
750}
751
752void *xmmap_gently(void *start, size_t length,
753 int prot, int flags, int fd, off_t offset)
754{
755 void *ret;
756
757 mmap_limit_check(length);
758 ret = mmap(start, length, prot, flags, fd, offset);
759 if (ret == MAP_FAILED) {
760 if (!length)
761 return NULL;
762 release_pack_memory(length);
763 ret = mmap(start, length, prot, flags, fd, offset);
764 }
765 return ret;
766}
767
768void *xmmap(void *start, size_t length,
769 int prot, int flags, int fd, off_t offset)
770{
771 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
772 if (ret == MAP_FAILED)
773 die_errno("mmap failed");
774 return ret;
775}
776
777/*
778 * With an in-core object data in "map", rehash it to make sure the
779 * object name actually matches "sha1" to detect object corruption.
780 * With "map" == NULL, try reading the object named with "sha1" using
781 * the streaming interface and rehash it to do the same.
782 */
783int check_sha1_signature(const unsigned char *sha1, void *map,
784 unsigned long size, const char *type)
785{
786 struct object_id real_oid;
787 enum object_type obj_type;
788 struct git_istream *st;
789 git_hash_ctx c;
790 char hdr[32];
791 int hdrlen;
792
793 if (map) {
794 hash_object_file(map, size, type, &real_oid);
795 return hashcmp(sha1, real_oid.hash) ? -1 : 0;
796 }
797
798 st = open_istream(sha1, &obj_type, &size, NULL);
799 if (!st)
800 return -1;
801
802 /* Generate the header */
803 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
804
805 /* Sha1.. */
806 the_hash_algo->init_fn(&c);
807 the_hash_algo->update_fn(&c, hdr, hdrlen);
808 for (;;) {
809 char buf[1024 * 16];
810 ssize_t readlen = read_istream(st, buf, sizeof(buf));
811
812 if (readlen < 0) {
813 close_istream(st);
814 return -1;
815 }
816 if (!readlen)
817 break;
818 the_hash_algo->update_fn(&c, buf, readlen);
819 }
820 the_hash_algo->final_fn(real_oid.hash, &c);
821 close_istream(st);
822 return hashcmp(sha1, real_oid.hash) ? -1 : 0;
823}
824
825int git_open_cloexec(const char *name, int flags)
826{
827 int fd;
828 static int o_cloexec = O_CLOEXEC;
829
830 fd = open(name, flags | o_cloexec);
831 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
832 /* Try again w/o O_CLOEXEC: the kernel might not support it */
833 o_cloexec &= ~O_CLOEXEC;
834 fd = open(name, flags | o_cloexec);
835 }
836
837#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
838 {
839 static int fd_cloexec = FD_CLOEXEC;
840
841 if (!o_cloexec && 0 <= fd && fd_cloexec) {
842 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
843 int flags = fcntl(fd, F_GETFD);
844 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
845 fd_cloexec = 0;
846 }
847 }
848#endif
849 return fd;
850}
851
852/*
853 * Find "sha1" as a loose object in the local repository or in an alternate.
854 * Returns 0 on success, negative on failure.
855 *
856 * The "path" out-parameter will give the path of the object we found (if any).
857 * Note that it may point to static storage and is only valid until another
858 * call to sha1_file_name(), etc.
859 */
860static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
861 const char **path)
862{
863 struct alternate_object_database *alt;
864 static struct strbuf buf = STRBUF_INIT;
865
866 strbuf_reset(&buf);
867 sha1_file_name(&buf, sha1);
868 *path = buf.buf;
869
870 if (!lstat(*path, st))
871 return 0;
872
873 prepare_alt_odb();
874 errno = ENOENT;
875 for (alt = alt_odb_list; alt; alt = alt->next) {
876 *path = alt_sha1_path(alt, sha1);
877 if (!lstat(*path, st))
878 return 0;
879 }
880
881 return -1;
882}
883
884/*
885 * Like stat_sha1_file(), but actually open the object and return the
886 * descriptor. See the caveats on the "path" parameter above.
887 */
888static int open_sha1_file(const unsigned char *sha1, const char **path)
889{
890 int fd;
891 struct alternate_object_database *alt;
892 int most_interesting_errno;
893 static struct strbuf buf = STRBUF_INIT;
894
895 strbuf_reset(&buf);
896 sha1_file_name(&buf, sha1);
897 *path = buf.buf;
898
899 fd = git_open(*path);
900 if (fd >= 0)
901 return fd;
902 most_interesting_errno = errno;
903
904 prepare_alt_odb();
905 for (alt = alt_odb_list; alt; alt = alt->next) {
906 *path = alt_sha1_path(alt, sha1);
907 fd = git_open(*path);
908 if (fd >= 0)
909 return fd;
910 if (most_interesting_errno == ENOENT)
911 most_interesting_errno = errno;
912 }
913 errno = most_interesting_errno;
914 return -1;
915}
916
917/*
918 * Map the loose object at "path" if it is not NULL, or the path found by
919 * searching for a loose object named "sha1".
920 */
921static void *map_sha1_file_1(const char *path,
922 const unsigned char *sha1,
923 unsigned long *size)
924{
925 void *map;
926 int fd;
927
928 if (path)
929 fd = git_open(path);
930 else
931 fd = open_sha1_file(sha1, &path);
932 map = NULL;
933 if (fd >= 0) {
934 struct stat st;
935
936 if (!fstat(fd, &st)) {
937 *size = xsize_t(st.st_size);
938 if (!*size) {
939 /* mmap() is forbidden on empty files */
940 error("object file %s is empty", path);
941 return NULL;
942 }
943 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
944 }
945 close(fd);
946 }
947 return map;
948}
949
950void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
951{
952 return map_sha1_file_1(NULL, sha1, size);
953}
954
955static int unpack_sha1_short_header(git_zstream *stream,
956 unsigned char *map, unsigned long mapsize,
957 void *buffer, unsigned long bufsiz)
958{
959 /* Get the data stream */
960 memset(stream, 0, sizeof(*stream));
961 stream->next_in = map;
962 stream->avail_in = mapsize;
963 stream->next_out = buffer;
964 stream->avail_out = bufsiz;
965
966 git_inflate_init(stream);
967 return git_inflate(stream, 0);
968}
969
970int unpack_sha1_header(git_zstream *stream,
971 unsigned char *map, unsigned long mapsize,
972 void *buffer, unsigned long bufsiz)
973{
974 int status = unpack_sha1_short_header(stream, map, mapsize,
975 buffer, bufsiz);
976
977 if (status < Z_OK)
978 return status;
979
980 /* Make sure we have the terminating NUL */
981 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
982 return -1;
983 return 0;
984}
985
986static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
987 unsigned long mapsize, void *buffer,
988 unsigned long bufsiz, struct strbuf *header)
989{
990 int status;
991
992 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
993 if (status < Z_OK)
994 return -1;
995
996 /*
997 * Check if entire header is unpacked in the first iteration.
998 */
999 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1000 return 0;
1001
1002 /*
1003 * buffer[0..bufsiz] was not large enough. Copy the partial
1004 * result out to header, and then append the result of further
1005 * reading the stream.
1006 */
1007 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1008 stream->next_out = buffer;
1009 stream->avail_out = bufsiz;
1010
1011 do {
1012 status = git_inflate(stream, 0);
1013 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1014 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1015 return 0;
1016 stream->next_out = buffer;
1017 stream->avail_out = bufsiz;
1018 } while (status != Z_STREAM_END);
1019 return -1;
1020}
1021
1022static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1023{
1024 int bytes = strlen(buffer) + 1;
1025 unsigned char *buf = xmallocz(size);
1026 unsigned long n;
1027 int status = Z_OK;
1028
1029 n = stream->total_out - bytes;
1030 if (n > size)
1031 n = size;
1032 memcpy(buf, (char *) buffer + bytes, n);
1033 bytes = n;
1034 if (bytes <= size) {
1035 /*
1036 * The above condition must be (bytes <= size), not
1037 * (bytes < size). In other words, even though we
1038 * expect no more output and set avail_out to zero,
1039 * the input zlib stream may have bytes that express
1040 * "this concludes the stream", and we *do* want to
1041 * eat that input.
1042 *
1043 * Otherwise we would not be able to test that we
1044 * consumed all the input to reach the expected size;
1045 * we also want to check that zlib tells us that all
1046 * went well with status == Z_STREAM_END at the end.
1047 */
1048 stream->next_out = buf + bytes;
1049 stream->avail_out = size - bytes;
1050 while (status == Z_OK)
1051 status = git_inflate(stream, Z_FINISH);
1052 }
1053 if (status == Z_STREAM_END && !stream->avail_in) {
1054 git_inflate_end(stream);
1055 return buf;
1056 }
1057
1058 if (status < 0)
1059 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1060 else if (stream->avail_in)
1061 error("garbage at end of loose object '%s'",
1062 sha1_to_hex(sha1));
1063 free(buf);
1064 return NULL;
1065}
1066
1067/*
1068 * We used to just use "sscanf()", but that's actually way
1069 * too permissive for what we want to check. So do an anal
1070 * object header parse by hand.
1071 */
1072static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1073 unsigned int flags)
1074{
1075 const char *type_buf = hdr;
1076 unsigned long size;
1077 int type, type_len = 0;
1078
1079 /*
1080 * The type can be of any size but is followed by
1081 * a space.
1082 */
1083 for (;;) {
1084 char c = *hdr++;
1085 if (!c)
1086 return -1;
1087 if (c == ' ')
1088 break;
1089 type_len++;
1090 }
1091
1092 type = type_from_string_gently(type_buf, type_len, 1);
1093 if (oi->type_name)
1094 strbuf_add(oi->type_name, type_buf, type_len);
1095 /*
1096 * Set type to 0 if its an unknown object and
1097 * we're obtaining the type using '--allow-unknown-type'
1098 * option.
1099 */
1100 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1101 type = 0;
1102 else if (type < 0)
1103 die("invalid object type");
1104 if (oi->typep)
1105 *oi->typep = type;
1106
1107 /*
1108 * The length must follow immediately, and be in canonical
1109 * decimal format (ie "010" is not valid).
1110 */
1111 size = *hdr++ - '0';
1112 if (size > 9)
1113 return -1;
1114 if (size) {
1115 for (;;) {
1116 unsigned long c = *hdr - '0';
1117 if (c > 9)
1118 break;
1119 hdr++;
1120 size = size * 10 + c;
1121 }
1122 }
1123
1124 if (oi->sizep)
1125 *oi->sizep = size;
1126
1127 /*
1128 * The length must be followed by a zero byte
1129 */
1130 return *hdr ? -1 : type;
1131}
1132
1133int parse_sha1_header(const char *hdr, unsigned long *sizep)
1134{
1135 struct object_info oi = OBJECT_INFO_INIT;
1136
1137 oi.sizep = sizep;
1138 return parse_sha1_header_extended(hdr, &oi, 0);
1139}
1140
1141static int sha1_loose_object_info(const unsigned char *sha1,
1142 struct object_info *oi,
1143 int flags)
1144{
1145 int status = 0;
1146 unsigned long mapsize;
1147 void *map;
1148 git_zstream stream;
1149 char hdr[32];
1150 struct strbuf hdrbuf = STRBUF_INIT;
1151 unsigned long size_scratch;
1152
1153 if (oi->delta_base_sha1)
1154 hashclr(oi->delta_base_sha1);
1155
1156 /*
1157 * If we don't care about type or size, then we don't
1158 * need to look inside the object at all. Note that we
1159 * do not optimize out the stat call, even if the
1160 * caller doesn't care about the disk-size, since our
1161 * return value implicitly indicates whether the
1162 * object even exists.
1163 */
1164 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1165 const char *path;
1166 struct stat st;
1167 if (stat_sha1_file(sha1, &st, &path) < 0)
1168 return -1;
1169 if (oi->disk_sizep)
1170 *oi->disk_sizep = st.st_size;
1171 return 0;
1172 }
1173
1174 map = map_sha1_file(sha1, &mapsize);
1175 if (!map)
1176 return -1;
1177
1178 if (!oi->sizep)
1179 oi->sizep = &size_scratch;
1180
1181 if (oi->disk_sizep)
1182 *oi->disk_sizep = mapsize;
1183 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1184 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1185 status = error("unable to unpack %s header with --allow-unknown-type",
1186 sha1_to_hex(sha1));
1187 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1188 status = error("unable to unpack %s header",
1189 sha1_to_hex(sha1));
1190 if (status < 0)
1191 ; /* Do nothing */
1192 else if (hdrbuf.len) {
1193 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1194 status = error("unable to parse %s header with --allow-unknown-type",
1195 sha1_to_hex(sha1));
1196 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1197 status = error("unable to parse %s header", sha1_to_hex(sha1));
1198
1199 if (status >= 0 && oi->contentp) {
1200 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1201 *oi->sizep, sha1);
1202 if (!*oi->contentp) {
1203 git_inflate_end(&stream);
1204 status = -1;
1205 }
1206 } else
1207 git_inflate_end(&stream);
1208
1209 munmap(map, mapsize);
1210 if (status && oi->typep)
1211 *oi->typep = status;
1212 if (oi->sizep == &size_scratch)
1213 oi->sizep = NULL;
1214 strbuf_release(&hdrbuf);
1215 oi->whence = OI_LOOSE;
1216 return (status < 0) ? status : 0;
1217}
1218
1219int fetch_if_missing = 1;
1220
1221int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1222{
1223 static struct object_info blank_oi = OBJECT_INFO_INIT;
1224 struct pack_entry e;
1225 int rtype;
1226 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1227 lookup_replace_object(sha1) :
1228 sha1;
1229 int already_retried = 0;
1230
1231 if (is_null_sha1(real))
1232 return -1;
1233
1234 if (!oi)
1235 oi = &blank_oi;
1236
1237 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1238 struct cached_object *co = find_cached_object(real);
1239 if (co) {
1240 if (oi->typep)
1241 *(oi->typep) = co->type;
1242 if (oi->sizep)
1243 *(oi->sizep) = co->size;
1244 if (oi->disk_sizep)
1245 *(oi->disk_sizep) = 0;
1246 if (oi->delta_base_sha1)
1247 hashclr(oi->delta_base_sha1);
1248 if (oi->type_name)
1249 strbuf_addstr(oi->type_name, type_name(co->type));
1250 if (oi->contentp)
1251 *oi->contentp = xmemdupz(co->buf, co->size);
1252 oi->whence = OI_CACHED;
1253 return 0;
1254 }
1255 }
1256
1257 while (1) {
1258 if (find_pack_entry(real, &e))
1259 break;
1260
1261 /* Most likely it's a loose object. */
1262 if (!sha1_loose_object_info(real, oi, flags))
1263 return 0;
1264
1265 /* Not a loose object; someone else may have just packed it. */
1266 reprepare_packed_git();
1267 if (find_pack_entry(real, &e))
1268 break;
1269
1270 /* Check if it is a missing object */
1271 if (fetch_if_missing && repository_format_partial_clone &&
1272 !already_retried) {
1273 /*
1274 * TODO Investigate haveing fetch_object() return
1275 * TODO error/success and stopping the music here.
1276 */
1277 fetch_object(repository_format_partial_clone, real);
1278 already_retried = 1;
1279 continue;
1280 }
1281
1282 return -1;
1283 }
1284
1285 if (oi == &blank_oi)
1286 /*
1287 * We know that the caller doesn't actually need the
1288 * information below, so return early.
1289 */
1290 return 0;
1291 rtype = packed_object_info(e.p, e.offset, oi);
1292 if (rtype < 0) {
1293 mark_bad_packed_object(e.p, real);
1294 return sha1_object_info_extended(real, oi, 0);
1295 } else if (oi->whence == OI_PACKED) {
1296 oi->u.packed.offset = e.offset;
1297 oi->u.packed.pack = e.p;
1298 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1299 rtype == OBJ_OFS_DELTA);
1300 }
1301
1302 return 0;
1303}
1304
1305/* returns enum object_type or negative */
1306int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1307{
1308 enum object_type type;
1309 struct object_info oi = OBJECT_INFO_INIT;
1310
1311 oi.typep = &type;
1312 oi.sizep = sizep;
1313 if (sha1_object_info_extended(sha1, &oi,
1314 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1315 return -1;
1316 return type;
1317}
1318
1319static void *read_object(const unsigned char *sha1, enum object_type *type,
1320 unsigned long *size)
1321{
1322 struct object_info oi = OBJECT_INFO_INIT;
1323 void *content;
1324 oi.typep = type;
1325 oi.sizep = size;
1326 oi.contentp = &content;
1327
1328 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1329 return NULL;
1330 return content;
1331}
1332
1333int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1334 struct object_id *oid)
1335{
1336 struct cached_object *co;
1337
1338 hash_object_file(buf, len, type_name(type), oid);
1339 if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1340 return 0;
1341 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1342 co = &cached_objects[cached_object_nr++];
1343 co->size = len;
1344 co->type = type;
1345 co->buf = xmalloc(len);
1346 memcpy(co->buf, buf, len);
1347 hashcpy(co->sha1, oid->hash);
1348 return 0;
1349}
1350
1351/*
1352 * This function dies on corrupt objects; the callers who want to
1353 * deal with them should arrange to call read_object() and give error
1354 * messages themselves.
1355 */
1356void *read_sha1_file_extended(const unsigned char *sha1,
1357 enum object_type *type,
1358 unsigned long *size,
1359 int lookup_replace)
1360{
1361 void *data;
1362 const struct packed_git *p;
1363 const char *path;
1364 struct stat st;
1365 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1366 : sha1;
1367
1368 errno = 0;
1369 data = read_object(repl, type, size);
1370 if (data)
1371 return data;
1372
1373 if (errno && errno != ENOENT)
1374 die_errno("failed to read object %s", sha1_to_hex(sha1));
1375
1376 /* die if we replaced an object with one that does not exist */
1377 if (repl != sha1)
1378 die("replacement %s not found for %s",
1379 sha1_to_hex(repl), sha1_to_hex(sha1));
1380
1381 if (!stat_sha1_file(repl, &st, &path))
1382 die("loose object %s (stored in %s) is corrupt",
1383 sha1_to_hex(repl), path);
1384
1385 if ((p = has_packed_and_bad(repl)) != NULL)
1386 die("packed object %s (stored in %s) is corrupt",
1387 sha1_to_hex(repl), p->pack_name);
1388
1389 return NULL;
1390}
1391
1392void *read_object_with_reference(const unsigned char *sha1,
1393 const char *required_type_name,
1394 unsigned long *size,
1395 unsigned char *actual_sha1_return)
1396{
1397 enum object_type type, required_type;
1398 void *buffer;
1399 unsigned long isize;
1400 unsigned char actual_sha1[20];
1401
1402 required_type = type_from_string(required_type_name);
1403 hashcpy(actual_sha1, sha1);
1404 while (1) {
1405 int ref_length = -1;
1406 const char *ref_type = NULL;
1407
1408 buffer = read_sha1_file(actual_sha1, &type, &isize);
1409 if (!buffer)
1410 return NULL;
1411 if (type == required_type) {
1412 *size = isize;
1413 if (actual_sha1_return)
1414 hashcpy(actual_sha1_return, actual_sha1);
1415 return buffer;
1416 }
1417 /* Handle references */
1418 else if (type == OBJ_COMMIT)
1419 ref_type = "tree ";
1420 else if (type == OBJ_TAG)
1421 ref_type = "object ";
1422 else {
1423 free(buffer);
1424 return NULL;
1425 }
1426 ref_length = strlen(ref_type);
1427
1428 if (ref_length + 40 > isize ||
1429 memcmp(buffer, ref_type, ref_length) ||
1430 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1431 free(buffer);
1432 return NULL;
1433 }
1434 free(buffer);
1435 /* Now we have the ID of the referred-to object in
1436 * actual_sha1. Check again. */
1437 }
1438}
1439
1440static void write_object_file_prepare(const void *buf, unsigned long len,
1441 const char *type, struct object_id *oid,
1442 char *hdr, int *hdrlen)
1443{
1444 git_hash_ctx c;
1445
1446 /* Generate the header */
1447 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1448
1449 /* Sha1.. */
1450 the_hash_algo->init_fn(&c);
1451 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1452 the_hash_algo->update_fn(&c, buf, len);
1453 the_hash_algo->final_fn(oid->hash, &c);
1454}
1455
1456/*
1457 * Move the just written object into its final resting place.
1458 */
1459int finalize_object_file(const char *tmpfile, const char *filename)
1460{
1461 int ret = 0;
1462
1463 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1464 goto try_rename;
1465 else if (link(tmpfile, filename))
1466 ret = errno;
1467
1468 /*
1469 * Coda hack - coda doesn't like cross-directory links,
1470 * so we fall back to a rename, which will mean that it
1471 * won't be able to check collisions, but that's not a
1472 * big deal.
1473 *
1474 * The same holds for FAT formatted media.
1475 *
1476 * When this succeeds, we just return. We have nothing
1477 * left to unlink.
1478 */
1479 if (ret && ret != EEXIST) {
1480 try_rename:
1481 if (!rename(tmpfile, filename))
1482 goto out;
1483 ret = errno;
1484 }
1485 unlink_or_warn(tmpfile);
1486 if (ret) {
1487 if (ret != EEXIST) {
1488 return error_errno("unable to write sha1 filename %s", filename);
1489 }
1490 /* FIXME!!! Collision check here ? */
1491 }
1492
1493out:
1494 if (adjust_shared_perm(filename))
1495 return error("unable to set permission to '%s'", filename);
1496 return 0;
1497}
1498
1499static int write_buffer(int fd, const void *buf, size_t len)
1500{
1501 if (write_in_full(fd, buf, len) < 0)
1502 return error_errno("file write error");
1503 return 0;
1504}
1505
1506int hash_object_file(const void *buf, unsigned long len, const char *type,
1507 struct object_id *oid)
1508{
1509 char hdr[32];
1510 int hdrlen = sizeof(hdr);
1511 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1512 return 0;
1513}
1514
1515/* Finalize a file on disk, and close it. */
1516static void close_sha1_file(int fd)
1517{
1518 if (fsync_object_files)
1519 fsync_or_die(fd, "sha1 file");
1520 if (close(fd) != 0)
1521 die_errno("error when closing sha1 file");
1522}
1523
1524/* Size of directory component, including the ending '/' */
1525static inline int directory_size(const char *filename)
1526{
1527 const char *s = strrchr(filename, '/');
1528 if (!s)
1529 return 0;
1530 return s - filename + 1;
1531}
1532
1533/*
1534 * This creates a temporary file in the same directory as the final
1535 * 'filename'
1536 *
1537 * We want to avoid cross-directory filename renames, because those
1538 * can have problems on various filesystems (FAT, NFS, Coda).
1539 */
1540static int create_tmpfile(struct strbuf *tmp, const char *filename)
1541{
1542 int fd, dirlen = directory_size(filename);
1543
1544 strbuf_reset(tmp);
1545 strbuf_add(tmp, filename, dirlen);
1546 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1547 fd = git_mkstemp_mode(tmp->buf, 0444);
1548 if (fd < 0 && dirlen && errno == ENOENT) {
1549 /*
1550 * Make sure the directory exists; note that the contents
1551 * of the buffer are undefined after mkstemp returns an
1552 * error, so we have to rewrite the whole buffer from
1553 * scratch.
1554 */
1555 strbuf_reset(tmp);
1556 strbuf_add(tmp, filename, dirlen - 1);
1557 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1558 return -1;
1559 if (adjust_shared_perm(tmp->buf))
1560 return -1;
1561
1562 /* Try again */
1563 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1564 fd = git_mkstemp_mode(tmp->buf, 0444);
1565 }
1566 return fd;
1567}
1568
1569static int write_loose_object(const struct object_id *oid, char *hdr,
1570 int hdrlen, const void *buf, unsigned long len,
1571 time_t mtime)
1572{
1573 int fd, ret;
1574 unsigned char compressed[4096];
1575 git_zstream stream;
1576 git_hash_ctx c;
1577 struct object_id parano_oid;
1578 static struct strbuf tmp_file = STRBUF_INIT;
1579 static struct strbuf filename = STRBUF_INIT;
1580
1581 strbuf_reset(&filename);
1582 sha1_file_name(&filename, oid->hash);
1583
1584 fd = create_tmpfile(&tmp_file, filename.buf);
1585 if (fd < 0) {
1586 if (errno == EACCES)
1587 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1588 else
1589 return error_errno("unable to create temporary file");
1590 }
1591
1592 /* Set it up */
1593 git_deflate_init(&stream, zlib_compression_level);
1594 stream.next_out = compressed;
1595 stream.avail_out = sizeof(compressed);
1596 the_hash_algo->init_fn(&c);
1597
1598 /* First header.. */
1599 stream.next_in = (unsigned char *)hdr;
1600 stream.avail_in = hdrlen;
1601 while (git_deflate(&stream, 0) == Z_OK)
1602 ; /* nothing */
1603 the_hash_algo->update_fn(&c, hdr, hdrlen);
1604
1605 /* Then the data itself.. */
1606 stream.next_in = (void *)buf;
1607 stream.avail_in = len;
1608 do {
1609 unsigned char *in0 = stream.next_in;
1610 ret = git_deflate(&stream, Z_FINISH);
1611 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1612 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1613 die("unable to write sha1 file");
1614 stream.next_out = compressed;
1615 stream.avail_out = sizeof(compressed);
1616 } while (ret == Z_OK);
1617
1618 if (ret != Z_STREAM_END)
1619 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1620 ret);
1621 ret = git_deflate_end_gently(&stream);
1622 if (ret != Z_OK)
1623 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1624 ret);
1625 the_hash_algo->final_fn(parano_oid.hash, &c);
1626 if (oidcmp(oid, ¶no_oid) != 0)
1627 die("confused by unstable object source data for %s",
1628 oid_to_hex(oid));
1629
1630 close_sha1_file(fd);
1631
1632 if (mtime) {
1633 struct utimbuf utb;
1634 utb.actime = mtime;
1635 utb.modtime = mtime;
1636 if (utime(tmp_file.buf, &utb) < 0)
1637 warning_errno("failed utime() on %s", tmp_file.buf);
1638 }
1639
1640 return finalize_object_file(tmp_file.buf, filename.buf);
1641}
1642
1643static int freshen_loose_object(const unsigned char *sha1)
1644{
1645 return check_and_freshen(sha1, 1);
1646}
1647
1648static int freshen_packed_object(const unsigned char *sha1)
1649{
1650 struct pack_entry e;
1651 if (!find_pack_entry(sha1, &e))
1652 return 0;
1653 if (e.p->freshened)
1654 return 1;
1655 if (!freshen_file(e.p->pack_name))
1656 return 0;
1657 e.p->freshened = 1;
1658 return 1;
1659}
1660
1661int write_object_file(const void *buf, unsigned long len, const char *type,
1662 struct object_id *oid)
1663{
1664 char hdr[32];
1665 int hdrlen = sizeof(hdr);
1666
1667 /* Normally if we have it in the pack then we do not bother writing
1668 * it out into .git/objects/??/?{38} file.
1669 */
1670 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1671 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1672 return 0;
1673 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1674}
1675
1676int hash_object_file_literally(const void *buf, unsigned long len,
1677 const char *type, struct object_id *oid,
1678 unsigned flags)
1679{
1680 char *header;
1681 int hdrlen, status = 0;
1682
1683 /* type string, SP, %lu of the length plus NUL must fit this */
1684 hdrlen = strlen(type) + 32;
1685 header = xmalloc(hdrlen);
1686 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1687
1688 if (!(flags & HASH_WRITE_OBJECT))
1689 goto cleanup;
1690 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1691 goto cleanup;
1692 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1693
1694cleanup:
1695 free(header);
1696 return status;
1697}
1698
1699int force_object_loose(const struct object_id *oid, time_t mtime)
1700{
1701 void *buf;
1702 unsigned long len;
1703 enum object_type type;
1704 char hdr[32];
1705 int hdrlen;
1706 int ret;
1707
1708 if (has_loose_object(oid->hash))
1709 return 0;
1710 buf = read_object(oid->hash, &type, &len);
1711 if (!buf)
1712 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1713 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1714 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1715 free(buf);
1716
1717 return ret;
1718}
1719
1720int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1721{
1722 if (!startup_info->have_repository)
1723 return 0;
1724 return sha1_object_info_extended(sha1, NULL,
1725 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1726}
1727
1728int has_object_file(const struct object_id *oid)
1729{
1730 return has_sha1_file(oid->hash);
1731}
1732
1733int has_object_file_with_flags(const struct object_id *oid, int flags)
1734{
1735 return has_sha1_file_with_flags(oid->hash, flags);
1736}
1737
1738static void check_tree(const void *buf, size_t size)
1739{
1740 struct tree_desc desc;
1741 struct name_entry entry;
1742
1743 init_tree_desc(&desc, buf, size);
1744 while (tree_entry(&desc, &entry))
1745 /* do nothing
1746 * tree_entry() will die() on malformed entries */
1747 ;
1748}
1749
1750static void check_commit(const void *buf, size_t size)
1751{
1752 struct commit c;
1753 memset(&c, 0, sizeof(c));
1754 if (parse_commit_buffer(&c, buf, size))
1755 die("corrupt commit");
1756}
1757
1758static void check_tag(const void *buf, size_t size)
1759{
1760 struct tag t;
1761 memset(&t, 0, sizeof(t));
1762 if (parse_tag_buffer(&t, buf, size))
1763 die("corrupt tag");
1764}
1765
1766static int index_mem(struct object_id *oid, void *buf, size_t size,
1767 enum object_type type,
1768 const char *path, unsigned flags)
1769{
1770 int ret, re_allocated = 0;
1771 int write_object = flags & HASH_WRITE_OBJECT;
1772
1773 if (!type)
1774 type = OBJ_BLOB;
1775
1776 /*
1777 * Convert blobs to git internal format
1778 */
1779 if ((type == OBJ_BLOB) && path) {
1780 struct strbuf nbuf = STRBUF_INIT;
1781 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1782 get_conv_flags(flags))) {
1783 buf = strbuf_detach(&nbuf, &size);
1784 re_allocated = 1;
1785 }
1786 }
1787 if (flags & HASH_FORMAT_CHECK) {
1788 if (type == OBJ_TREE)
1789 check_tree(buf, size);
1790 if (type == OBJ_COMMIT)
1791 check_commit(buf, size);
1792 if (type == OBJ_TAG)
1793 check_tag(buf, size);
1794 }
1795
1796 if (write_object)
1797 ret = write_object_file(buf, size, type_name(type), oid);
1798 else
1799 ret = hash_object_file(buf, size, type_name(type), oid);
1800 if (re_allocated)
1801 free(buf);
1802 return ret;
1803}
1804
1805static int index_stream_convert_blob(struct object_id *oid, int fd,
1806 const char *path, unsigned flags)
1807{
1808 int ret;
1809 const int write_object = flags & HASH_WRITE_OBJECT;
1810 struct strbuf sbuf = STRBUF_INIT;
1811
1812 assert(path);
1813 assert(would_convert_to_git_filter_fd(path));
1814
1815 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1816 get_conv_flags(flags));
1817
1818 if (write_object)
1819 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1820 oid);
1821 else
1822 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1823 oid);
1824 strbuf_release(&sbuf);
1825 return ret;
1826}
1827
1828static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1829 const char *path, unsigned flags)
1830{
1831 struct strbuf sbuf = STRBUF_INIT;
1832 int ret;
1833
1834 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1835 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1836 else
1837 ret = -1;
1838 strbuf_release(&sbuf);
1839 return ret;
1840}
1841
1842#define SMALL_FILE_SIZE (32*1024)
1843
1844static int index_core(struct object_id *oid, int fd, size_t size,
1845 enum object_type type, const char *path,
1846 unsigned flags)
1847{
1848 int ret;
1849
1850 if (!size) {
1851 ret = index_mem(oid, "", size, type, path, flags);
1852 } else if (size <= SMALL_FILE_SIZE) {
1853 char *buf = xmalloc(size);
1854 ssize_t read_result = read_in_full(fd, buf, size);
1855 if (read_result < 0)
1856 ret = error_errno("read error while indexing %s",
1857 path ? path : "<unknown>");
1858 else if (read_result != size)
1859 ret = error("short read while indexing %s",
1860 path ? path : "<unknown>");
1861 else
1862 ret = index_mem(oid, buf, size, type, path, flags);
1863 free(buf);
1864 } else {
1865 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1866 ret = index_mem(oid, buf, size, type, path, flags);
1867 munmap(buf, size);
1868 }
1869 return ret;
1870}
1871
1872/*
1873 * This creates one packfile per large blob unless bulk-checkin
1874 * machinery is "plugged".
1875 *
1876 * This also bypasses the usual "convert-to-git" dance, and that is on
1877 * purpose. We could write a streaming version of the converting
1878 * functions and insert that before feeding the data to fast-import
1879 * (or equivalent in-core API described above). However, that is
1880 * somewhat complicated, as we do not know the size of the filter
1881 * result, which we need to know beforehand when writing a git object.
1882 * Since the primary motivation for trying to stream from the working
1883 * tree file and to avoid mmaping it in core is to deal with large
1884 * binary blobs, they generally do not want to get any conversion, and
1885 * callers should avoid this code path when filters are requested.
1886 */
1887static int index_stream(struct object_id *oid, int fd, size_t size,
1888 enum object_type type, const char *path,
1889 unsigned flags)
1890{
1891 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1892}
1893
1894int index_fd(struct object_id *oid, int fd, struct stat *st,
1895 enum object_type type, const char *path, unsigned flags)
1896{
1897 int ret;
1898
1899 /*
1900 * Call xsize_t() only when needed to avoid potentially unnecessary
1901 * die() for large files.
1902 */
1903 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1904 ret = index_stream_convert_blob(oid, fd, path, flags);
1905 else if (!S_ISREG(st->st_mode))
1906 ret = index_pipe(oid, fd, type, path, flags);
1907 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1908 (path && would_convert_to_git(&the_index, path)))
1909 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1910 flags);
1911 else
1912 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1913 flags);
1914 close(fd);
1915 return ret;
1916}
1917
1918int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1919{
1920 int fd;
1921 struct strbuf sb = STRBUF_INIT;
1922 int rc = 0;
1923
1924 switch (st->st_mode & S_IFMT) {
1925 case S_IFREG:
1926 fd = open(path, O_RDONLY);
1927 if (fd < 0)
1928 return error_errno("open(\"%s\")", path);
1929 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1930 return error("%s: failed to insert into database",
1931 path);
1932 break;
1933 case S_IFLNK:
1934 if (strbuf_readlink(&sb, path, st->st_size))
1935 return error_errno("readlink(\"%s\")", path);
1936 if (!(flags & HASH_WRITE_OBJECT))
1937 hash_object_file(sb.buf, sb.len, blob_type, oid);
1938 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1939 rc = error("%s: failed to insert into database", path);
1940 strbuf_release(&sb);
1941 break;
1942 case S_IFDIR:
1943 return resolve_gitlink_ref(path, "HEAD", oid);
1944 default:
1945 return error("%s: unsupported file type", path);
1946 }
1947 return rc;
1948}
1949
1950int read_pack_header(int fd, struct pack_header *header)
1951{
1952 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1953 /* "eof before pack header was fully read" */
1954 return PH_ERROR_EOF;
1955
1956 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1957 /* "protocol error (pack signature mismatch detected)" */
1958 return PH_ERROR_PACK_SIGNATURE;
1959 if (!pack_version_ok(header->hdr_version))
1960 /* "protocol error (pack version unsupported)" */
1961 return PH_ERROR_PROTOCOL;
1962 return 0;
1963}
1964
1965void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1966{
1967 enum object_type type = sha1_object_info(sha1, NULL);
1968 if (type < 0)
1969 die("%s is not a valid object", sha1_to_hex(sha1));
1970 if (type != expect)
1971 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1972 type_name(expect));
1973}
1974
1975int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1976 struct strbuf *path,
1977 each_loose_object_fn obj_cb,
1978 each_loose_cruft_fn cruft_cb,
1979 each_loose_subdir_fn subdir_cb,
1980 void *data)
1981{
1982 size_t origlen, baselen;
1983 DIR *dir;
1984 struct dirent *de;
1985 int r = 0;
1986 struct object_id oid;
1987
1988 if (subdir_nr > 0xff)
1989 BUG("invalid loose object subdirectory: %x", subdir_nr);
1990
1991 origlen = path->len;
1992 strbuf_complete(path, '/');
1993 strbuf_addf(path, "%02x", subdir_nr);
1994
1995 dir = opendir(path->buf);
1996 if (!dir) {
1997 if (errno != ENOENT)
1998 r = error_errno("unable to open %s", path->buf);
1999 strbuf_setlen(path, origlen);
2000 return r;
2001 }
2002
2003 oid.hash[0] = subdir_nr;
2004 strbuf_addch(path, '/');
2005 baselen = path->len;
2006
2007 while ((de = readdir(dir))) {
2008 size_t namelen;
2009 if (is_dot_or_dotdot(de->d_name))
2010 continue;
2011
2012 namelen = strlen(de->d_name);
2013 strbuf_setlen(path, baselen);
2014 strbuf_add(path, de->d_name, namelen);
2015 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2016 !hex_to_bytes(oid.hash + 1, de->d_name,
2017 GIT_SHA1_RAWSZ - 1)) {
2018 if (obj_cb) {
2019 r = obj_cb(&oid, path->buf, data);
2020 if (r)
2021 break;
2022 }
2023 continue;
2024 }
2025
2026 if (cruft_cb) {
2027 r = cruft_cb(de->d_name, path->buf, data);
2028 if (r)
2029 break;
2030 }
2031 }
2032 closedir(dir);
2033
2034 strbuf_setlen(path, baselen - 1);
2035 if (!r && subdir_cb)
2036 r = subdir_cb(subdir_nr, path->buf, data);
2037
2038 strbuf_setlen(path, origlen);
2039
2040 return r;
2041}
2042
2043int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2044 each_loose_object_fn obj_cb,
2045 each_loose_cruft_fn cruft_cb,
2046 each_loose_subdir_fn subdir_cb,
2047 void *data)
2048{
2049 int r = 0;
2050 int i;
2051
2052 for (i = 0; i < 256; i++) {
2053 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2054 subdir_cb, data);
2055 if (r)
2056 break;
2057 }
2058
2059 return r;
2060}
2061
2062int for_each_loose_file_in_objdir(const char *path,
2063 each_loose_object_fn obj_cb,
2064 each_loose_cruft_fn cruft_cb,
2065 each_loose_subdir_fn subdir_cb,
2066 void *data)
2067{
2068 struct strbuf buf = STRBUF_INIT;
2069 int r;
2070
2071 strbuf_addstr(&buf, path);
2072 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2073 subdir_cb, data);
2074 strbuf_release(&buf);
2075
2076 return r;
2077}
2078
2079struct loose_alt_odb_data {
2080 each_loose_object_fn *cb;
2081 void *data;
2082};
2083
2084static int loose_from_alt_odb(struct alternate_object_database *alt,
2085 void *vdata)
2086{
2087 struct loose_alt_odb_data *data = vdata;
2088 struct strbuf buf = STRBUF_INIT;
2089 int r;
2090
2091 strbuf_addstr(&buf, alt->path);
2092 r = for_each_loose_file_in_objdir_buf(&buf,
2093 data->cb, NULL, NULL,
2094 data->data);
2095 strbuf_release(&buf);
2096 return r;
2097}
2098
2099int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2100{
2101 struct loose_alt_odb_data alt;
2102 int r;
2103
2104 r = for_each_loose_file_in_objdir(get_object_directory(),
2105 cb, NULL, NULL, data);
2106 if (r)
2107 return r;
2108
2109 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2110 return 0;
2111
2112 alt.cb = cb;
2113 alt.data = data;
2114 return foreach_alt_odb(loose_from_alt_odb, &alt);
2115}
2116
2117static int check_stream_sha1(git_zstream *stream,
2118 const char *hdr,
2119 unsigned long size,
2120 const char *path,
2121 const unsigned char *expected_sha1)
2122{
2123 git_hash_ctx c;
2124 unsigned char real_sha1[GIT_MAX_RAWSZ];
2125 unsigned char buf[4096];
2126 unsigned long total_read;
2127 int status = Z_OK;
2128
2129 the_hash_algo->init_fn(&c);
2130 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2131
2132 /*
2133 * We already read some bytes into hdr, but the ones up to the NUL
2134 * do not count against the object's content size.
2135 */
2136 total_read = stream->total_out - strlen(hdr) - 1;
2137
2138 /*
2139 * This size comparison must be "<=" to read the final zlib packets;
2140 * see the comment in unpack_sha1_rest for details.
2141 */
2142 while (total_read <= size &&
2143 (status == Z_OK || status == Z_BUF_ERROR)) {
2144 stream->next_out = buf;
2145 stream->avail_out = sizeof(buf);
2146 if (size - total_read < stream->avail_out)
2147 stream->avail_out = size - total_read;
2148 status = git_inflate(stream, Z_FINISH);
2149 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2150 total_read += stream->next_out - buf;
2151 }
2152 git_inflate_end(stream);
2153
2154 if (status != Z_STREAM_END) {
2155 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2156 return -1;
2157 }
2158 if (stream->avail_in) {
2159 error("garbage at end of loose object '%s'",
2160 sha1_to_hex(expected_sha1));
2161 return -1;
2162 }
2163
2164 the_hash_algo->final_fn(real_sha1, &c);
2165 if (hashcmp(expected_sha1, real_sha1)) {
2166 error("sha1 mismatch for %s (expected %s)", path,
2167 sha1_to_hex(expected_sha1));
2168 return -1;
2169 }
2170
2171 return 0;
2172}
2173
2174int read_loose_object(const char *path,
2175 const unsigned char *expected_sha1,
2176 enum object_type *type,
2177 unsigned long *size,
2178 void **contents)
2179{
2180 int ret = -1;
2181 void *map = NULL;
2182 unsigned long mapsize;
2183 git_zstream stream;
2184 char hdr[32];
2185
2186 *contents = NULL;
2187
2188 map = map_sha1_file_1(path, NULL, &mapsize);
2189 if (!map) {
2190 error_errno("unable to mmap %s", path);
2191 goto out;
2192 }
2193
2194 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2195 error("unable to unpack header of %s", path);
2196 goto out;
2197 }
2198
2199 *type = parse_sha1_header(hdr, size);
2200 if (*type < 0) {
2201 error("unable to parse header of %s", path);
2202 git_inflate_end(&stream);
2203 goto out;
2204 }
2205
2206 if (*type == OBJ_BLOB) {
2207 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2208 goto out;
2209 } else {
2210 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2211 if (!*contents) {
2212 error("unable to unpack contents of %s", path);
2213 git_inflate_end(&stream);
2214 goto out;
2215 }
2216 if (check_sha1_signature(expected_sha1, *contents,
2217 *size, type_name(*type))) {
2218 error("sha1 mismatch for %s (expected %s)", path,
2219 sha1_to_hex(expected_sha1));
2220 free(*contents);
2221 goto out;
2222 }
2223 }
2224
2225 ret = 0; /* everything checks out */
2226
2227out:
2228 if (map)
2229 munmap(map, mapsize);
2230 return ret;
2231}