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(void *ctx)
43{
44 git_SHA1_Init((git_SHA_CTX *)ctx);
45}
46
47static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
48{
49 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
50}
51
52static void git_hash_sha1_final(unsigned char *hash, void *ctx)
53{
54 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
55}
56
57static void git_hash_unknown_init(void *ctx)
58{
59 die("trying to init unknown hash");
60}
61
62static void git_hash_unknown_update(void *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, void *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 0,
79 git_hash_unknown_init,
80 git_hash_unknown_update,
81 git_hash_unknown_final,
82 NULL,
83 NULL,
84 },
85 {
86 "sha-1",
87 /* "sha1", big-endian */
88 0x73686131,
89 sizeof(git_SHA_CTX),
90 GIT_SHA1_RAWSZ,
91 GIT_SHA1_HEXSZ,
92 git_hash_sha1_init,
93 git_hash_sha1_update,
94 git_hash_sha1_final,
95 &empty_tree_oid,
96 &empty_blob_oid,
97 },
98};
99
100/*
101 * This is meant to hold a *small* number of objects that you would
102 * want read_sha1_file() to be able to return, but yet you do not want
103 * to write them into the object store (e.g. a browse-only
104 * application).
105 */
106static struct cached_object {
107 unsigned char sha1[20];
108 enum object_type type;
109 void *buf;
110 unsigned long size;
111} *cached_objects;
112static int cached_object_nr, cached_object_alloc;
113
114static struct cached_object empty_tree = {
115 EMPTY_TREE_SHA1_BIN_LITERAL,
116 OBJ_TREE,
117 "",
118 0
119};
120
121static struct cached_object *find_cached_object(const unsigned char *sha1)
122{
123 int i;
124 struct cached_object *co = cached_objects;
125
126 for (i = 0; i < cached_object_nr; i++, co++) {
127 if (!hashcmp(co->sha1, sha1))
128 return co;
129 }
130 if (!hashcmp(sha1, empty_tree.sha1))
131 return &empty_tree;
132 return NULL;
133}
134
135
136static int get_conv_flags(unsigned flags)
137{
138 if (flags & HASH_RENORMALIZE)
139 return CONV_EOL_RENORMALIZE;
140 else if (flags & HASH_WRITE_OBJECT)
141 return global_conv_flags_eol;
142 else
143 return 0;
144}
145
146
147int mkdir_in_gitdir(const char *path)
148{
149 if (mkdir(path, 0777)) {
150 int saved_errno = errno;
151 struct stat st;
152 struct strbuf sb = STRBUF_INIT;
153
154 if (errno != EEXIST)
155 return -1;
156 /*
157 * Are we looking at a path in a symlinked worktree
158 * whose original repository does not yet have it?
159 * e.g. .git/rr-cache pointing at its original
160 * repository in which the user hasn't performed any
161 * conflict resolution yet?
162 */
163 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
164 strbuf_readlink(&sb, path, st.st_size) ||
165 !is_absolute_path(sb.buf) ||
166 mkdir(sb.buf, 0777)) {
167 strbuf_release(&sb);
168 errno = saved_errno;
169 return -1;
170 }
171 strbuf_release(&sb);
172 }
173 return adjust_shared_perm(path);
174}
175
176enum scld_error safe_create_leading_directories(char *path)
177{
178 char *next_component = path + offset_1st_component(path);
179 enum scld_error ret = SCLD_OK;
180
181 while (ret == SCLD_OK && next_component) {
182 struct stat st;
183 char *slash = next_component, slash_character;
184
185 while (*slash && !is_dir_sep(*slash))
186 slash++;
187
188 if (!*slash)
189 break;
190
191 next_component = slash + 1;
192 while (is_dir_sep(*next_component))
193 next_component++;
194 if (!*next_component)
195 break;
196
197 slash_character = *slash;
198 *slash = '\0';
199 if (!stat(path, &st)) {
200 /* path exists */
201 if (!S_ISDIR(st.st_mode)) {
202 errno = ENOTDIR;
203 ret = SCLD_EXISTS;
204 }
205 } else if (mkdir(path, 0777)) {
206 if (errno == EEXIST &&
207 !stat(path, &st) && S_ISDIR(st.st_mode))
208 ; /* somebody created it since we checked */
209 else if (errno == ENOENT)
210 /*
211 * Either mkdir() failed because
212 * somebody just pruned the containing
213 * directory, or stat() failed because
214 * the file that was in our way was
215 * just removed. Either way, inform
216 * the caller that it might be worth
217 * trying again:
218 */
219 ret = SCLD_VANISHED;
220 else
221 ret = SCLD_FAILED;
222 } else if (adjust_shared_perm(path)) {
223 ret = SCLD_PERMS;
224 }
225 *slash = slash_character;
226 }
227 return ret;
228}
229
230enum scld_error safe_create_leading_directories_const(const char *path)
231{
232 int save_errno;
233 /* path points to cache entries, so xstrdup before messing with it */
234 char *buf = xstrdup(path);
235 enum scld_error result = safe_create_leading_directories(buf);
236
237 save_errno = errno;
238 free(buf);
239 errno = save_errno;
240 return result;
241}
242
243int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
244{
245 /*
246 * The number of times we will try to remove empty directories
247 * in the way of path. This is only 1 because if another
248 * process is racily creating directories that conflict with
249 * us, we don't want to fight against them.
250 */
251 int remove_directories_remaining = 1;
252
253 /*
254 * The number of times that we will try to create the
255 * directories containing path. We are willing to attempt this
256 * more than once, because another process could be trying to
257 * clean up empty directories at the same time as we are
258 * trying to create them.
259 */
260 int create_directories_remaining = 3;
261
262 /* A scratch copy of path, filled lazily if we need it: */
263 struct strbuf path_copy = STRBUF_INIT;
264
265 int ret, save_errno;
266
267 /* Sanity check: */
268 assert(*path);
269
270retry_fn:
271 ret = fn(path, cb);
272 save_errno = errno;
273 if (!ret)
274 goto out;
275
276 if (errno == EISDIR && remove_directories_remaining-- > 0) {
277 /*
278 * A directory is in the way. Maybe it is empty; try
279 * to remove it:
280 */
281 if (!path_copy.len)
282 strbuf_addstr(&path_copy, path);
283
284 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
285 goto retry_fn;
286 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
287 /*
288 * Maybe the containing directory didn't exist, or
289 * maybe it was just deleted by a process that is
290 * racing with us to clean up empty directories. Try
291 * to create it:
292 */
293 enum scld_error scld_result;
294
295 if (!path_copy.len)
296 strbuf_addstr(&path_copy, path);
297
298 do {
299 scld_result = safe_create_leading_directories(path_copy.buf);
300 if (scld_result == SCLD_OK)
301 goto retry_fn;
302 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
303 }
304
305out:
306 strbuf_release(&path_copy);
307 errno = save_errno;
308 return ret;
309}
310
311static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
312{
313 int i;
314 for (i = 0; i < 20; i++) {
315 static char hex[] = "0123456789abcdef";
316 unsigned int val = sha1[i];
317 strbuf_addch(buf, hex[val >> 4]);
318 strbuf_addch(buf, hex[val & 0xf]);
319 if (!i)
320 strbuf_addch(buf, '/');
321 }
322}
323
324void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
325{
326 strbuf_addstr(buf, get_object_directory());
327 strbuf_addch(buf, '/');
328 fill_sha1_path(buf, sha1);
329}
330
331struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
332{
333 strbuf_setlen(&alt->scratch, alt->base_len);
334 return &alt->scratch;
335}
336
337static const char *alt_sha1_path(struct alternate_object_database *alt,
338 const unsigned char *sha1)
339{
340 struct strbuf *buf = alt_scratch_buf(alt);
341 fill_sha1_path(buf, sha1);
342 return buf->buf;
343}
344
345struct alternate_object_database *alt_odb_list;
346static struct alternate_object_database **alt_odb_tail;
347
348/*
349 * Return non-zero iff the path is usable as an alternate object database.
350 */
351static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
352{
353 struct alternate_object_database *alt;
354
355 /* Detect cases where alternate disappeared */
356 if (!is_directory(path->buf)) {
357 error("object directory %s does not exist; "
358 "check .git/objects/info/alternates.",
359 path->buf);
360 return 0;
361 }
362
363 /*
364 * Prevent the common mistake of listing the same
365 * thing twice, or object directory itself.
366 */
367 for (alt = alt_odb_list; alt; alt = alt->next) {
368 if (!fspathcmp(path->buf, alt->path))
369 return 0;
370 }
371 if (!fspathcmp(path->buf, normalized_objdir))
372 return 0;
373
374 return 1;
375}
376
377/*
378 * Prepare alternate object database registry.
379 *
380 * The variable alt_odb_list points at the list of struct
381 * alternate_object_database. The elements on this list come from
382 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
383 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
384 * whose contents is similar to that environment variable but can be
385 * LF separated. Its base points at a statically allocated buffer that
386 * contains "/the/directory/corresponding/to/.git/objects/...", while
387 * its name points just after the slash at the end of ".git/objects/"
388 * in the example above, and has enough space to hold 40-byte hex
389 * SHA1, an extra slash for the first level indirection, and the
390 * terminating NUL.
391 */
392static void read_info_alternates(const char * relative_base, int depth);
393static int link_alt_odb_entry(const char *entry, const char *relative_base,
394 int depth, const char *normalized_objdir)
395{
396 struct alternate_object_database *ent;
397 struct strbuf pathbuf = STRBUF_INIT;
398
399 if (!is_absolute_path(entry) && relative_base) {
400 strbuf_realpath(&pathbuf, relative_base, 1);
401 strbuf_addch(&pathbuf, '/');
402 }
403 strbuf_addstr(&pathbuf, entry);
404
405 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
406 error("unable to normalize alternate object path: %s",
407 pathbuf.buf);
408 strbuf_release(&pathbuf);
409 return -1;
410 }
411
412 /*
413 * The trailing slash after the directory name is given by
414 * this function at the end. Remove duplicates.
415 */
416 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
417 strbuf_setlen(&pathbuf, pathbuf.len - 1);
418
419 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
420 strbuf_release(&pathbuf);
421 return -1;
422 }
423
424 ent = alloc_alt_odb(pathbuf.buf);
425
426 /* add the alternate entry */
427 *alt_odb_tail = ent;
428 alt_odb_tail = &(ent->next);
429 ent->next = NULL;
430
431 /* recursively add alternates */
432 read_info_alternates(pathbuf.buf, depth + 1);
433
434 strbuf_release(&pathbuf);
435 return 0;
436}
437
438static const char *parse_alt_odb_entry(const char *string,
439 int sep,
440 struct strbuf *out)
441{
442 const char *end;
443
444 strbuf_reset(out);
445
446 if (*string == '#') {
447 /* comment; consume up to next separator */
448 end = strchrnul(string, sep);
449 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
450 /*
451 * quoted path; unquote_c_style has copied the
452 * data for us and set "end". Broken quoting (e.g.,
453 * an entry that doesn't end with a quote) falls
454 * back to the unquoted case below.
455 */
456 } else {
457 /* normal, unquoted path */
458 end = strchrnul(string, sep);
459 strbuf_add(out, string, end - string);
460 }
461
462 if (*end)
463 end++;
464 return end;
465}
466
467static void link_alt_odb_entries(const char *alt, int sep,
468 const char *relative_base, int depth)
469{
470 struct strbuf objdirbuf = STRBUF_INIT;
471 struct strbuf entry = STRBUF_INIT;
472
473 if (!alt || !*alt)
474 return;
475
476 if (depth > 5) {
477 error("%s: ignoring alternate object stores, nesting too deep.",
478 relative_base);
479 return;
480 }
481
482 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
483 if (strbuf_normalize_path(&objdirbuf) < 0)
484 die("unable to normalize object directory: %s",
485 objdirbuf.buf);
486
487 while (*alt) {
488 alt = parse_alt_odb_entry(alt, sep, &entry);
489 if (!entry.len)
490 continue;
491 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
492 }
493 strbuf_release(&entry);
494 strbuf_release(&objdirbuf);
495}
496
497static void read_info_alternates(const char * relative_base, int depth)
498{
499 char *path;
500 struct strbuf buf = STRBUF_INIT;
501
502 path = xstrfmt("%s/info/alternates", relative_base);
503 if (strbuf_read_file(&buf, path, 1024) < 0) {
504 warn_on_fopen_errors(path);
505 free(path);
506 return;
507 }
508
509 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
510 strbuf_release(&buf);
511 free(path);
512}
513
514struct alternate_object_database *alloc_alt_odb(const char *dir)
515{
516 struct alternate_object_database *ent;
517
518 FLEX_ALLOC_STR(ent, path, dir);
519 strbuf_init(&ent->scratch, 0);
520 strbuf_addf(&ent->scratch, "%s/", dir);
521 ent->base_len = ent->scratch.len;
522
523 return ent;
524}
525
526void add_to_alternates_file(const char *reference)
527{
528 struct lock_file lock = LOCK_INIT;
529 char *alts = git_pathdup("objects/info/alternates");
530 FILE *in, *out;
531 int found = 0;
532
533 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
534 out = fdopen_lock_file(&lock, "w");
535 if (!out)
536 die_errno("unable to fdopen alternates lockfile");
537
538 in = fopen(alts, "r");
539 if (in) {
540 struct strbuf line = STRBUF_INIT;
541
542 while (strbuf_getline(&line, in) != EOF) {
543 if (!strcmp(reference, line.buf)) {
544 found = 1;
545 break;
546 }
547 fprintf_or_die(out, "%s\n", line.buf);
548 }
549
550 strbuf_release(&line);
551 fclose(in);
552 }
553 else if (errno != ENOENT)
554 die_errno("unable to read alternates file");
555
556 if (found) {
557 rollback_lock_file(&lock);
558 } else {
559 fprintf_or_die(out, "%s\n", reference);
560 if (commit_lock_file(&lock))
561 die_errno("unable to move new alternates file into place");
562 if (alt_odb_tail)
563 link_alt_odb_entries(reference, '\n', NULL, 0);
564 }
565 free(alts);
566}
567
568void add_to_alternates_memory(const char *reference)
569{
570 /*
571 * Make sure alternates are initialized, or else our entry may be
572 * overwritten when they are.
573 */
574 prepare_alt_odb();
575
576 link_alt_odb_entries(reference, '\n', NULL, 0);
577}
578
579/*
580 * Compute the exact path an alternate is at and returns it. In case of
581 * error NULL is returned and the human readable error is added to `err`
582 * `path` may be relative and should point to $GITDIR.
583 * `err` must not be null.
584 */
585char *compute_alternate_path(const char *path, struct strbuf *err)
586{
587 char *ref_git = NULL;
588 const char *repo, *ref_git_s;
589 int seen_error = 0;
590
591 ref_git_s = real_path_if_valid(path);
592 if (!ref_git_s) {
593 seen_error = 1;
594 strbuf_addf(err, _("path '%s' does not exist"), path);
595 goto out;
596 } else
597 /*
598 * Beware: read_gitfile(), real_path() and mkpath()
599 * return static buffer
600 */
601 ref_git = xstrdup(ref_git_s);
602
603 repo = read_gitfile(ref_git);
604 if (!repo)
605 repo = read_gitfile(mkpath("%s/.git", ref_git));
606 if (repo) {
607 free(ref_git);
608 ref_git = xstrdup(repo);
609 }
610
611 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
612 char *ref_git_git = mkpathdup("%s/.git", ref_git);
613 free(ref_git);
614 ref_git = ref_git_git;
615 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
616 struct strbuf sb = STRBUF_INIT;
617 seen_error = 1;
618 if (get_common_dir(&sb, ref_git)) {
619 strbuf_addf(err,
620 _("reference repository '%s' as a linked "
621 "checkout is not supported yet."),
622 path);
623 goto out;
624 }
625
626 strbuf_addf(err, _("reference repository '%s' is not a "
627 "local repository."), path);
628 goto out;
629 }
630
631 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
632 strbuf_addf(err, _("reference repository '%s' is shallow"),
633 path);
634 seen_error = 1;
635 goto out;
636 }
637
638 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
639 strbuf_addf(err,
640 _("reference repository '%s' is grafted"),
641 path);
642 seen_error = 1;
643 goto out;
644 }
645
646out:
647 if (seen_error) {
648 FREE_AND_NULL(ref_git);
649 }
650
651 return ref_git;
652}
653
654int foreach_alt_odb(alt_odb_fn fn, void *cb)
655{
656 struct alternate_object_database *ent;
657 int r = 0;
658
659 prepare_alt_odb();
660 for (ent = alt_odb_list; ent; ent = ent->next) {
661 r = fn(ent, cb);
662 if (r)
663 break;
664 }
665 return r;
666}
667
668void prepare_alt_odb(void)
669{
670 const char *alt;
671
672 if (alt_odb_tail)
673 return;
674
675 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
676
677 alt_odb_tail = &alt_odb_list;
678 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
679
680 read_info_alternates(get_object_directory(), 0);
681}
682
683/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
684static int freshen_file(const char *fn)
685{
686 struct utimbuf t;
687 t.actime = t.modtime = time(NULL);
688 return !utime(fn, &t);
689}
690
691/*
692 * All of the check_and_freshen functions return 1 if the file exists and was
693 * freshened (if freshening was requested), 0 otherwise. If they return
694 * 0, you should not assume that it is safe to skip a write of the object (it
695 * either does not exist on disk, or has a stale mtime and may be subject to
696 * pruning).
697 */
698int check_and_freshen_file(const char *fn, int freshen)
699{
700 if (access(fn, F_OK))
701 return 0;
702 if (freshen && !freshen_file(fn))
703 return 0;
704 return 1;
705}
706
707static int check_and_freshen_local(const unsigned char *sha1, int freshen)
708{
709 static struct strbuf buf = STRBUF_INIT;
710
711 strbuf_reset(&buf);
712 sha1_file_name(&buf, sha1);
713
714 return check_and_freshen_file(buf.buf, freshen);
715}
716
717static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
718{
719 struct alternate_object_database *alt;
720 prepare_alt_odb();
721 for (alt = alt_odb_list; alt; alt = alt->next) {
722 const char *path = alt_sha1_path(alt, sha1);
723 if (check_and_freshen_file(path, freshen))
724 return 1;
725 }
726 return 0;
727}
728
729static int check_and_freshen(const unsigned char *sha1, int freshen)
730{
731 return check_and_freshen_local(sha1, freshen) ||
732 check_and_freshen_nonlocal(sha1, freshen);
733}
734
735int has_loose_object_nonlocal(const unsigned char *sha1)
736{
737 return check_and_freshen_nonlocal(sha1, 0);
738}
739
740static int has_loose_object(const unsigned char *sha1)
741{
742 return check_and_freshen(sha1, 0);
743}
744
745static void mmap_limit_check(size_t length)
746{
747 static size_t limit = 0;
748 if (!limit) {
749 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
750 if (!limit)
751 limit = SIZE_MAX;
752 }
753 if (length > limit)
754 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
755 (uintmax_t)length, (uintmax_t)limit);
756}
757
758void *xmmap_gently(void *start, size_t length,
759 int prot, int flags, int fd, off_t offset)
760{
761 void *ret;
762
763 mmap_limit_check(length);
764 ret = mmap(start, length, prot, flags, fd, offset);
765 if (ret == MAP_FAILED) {
766 if (!length)
767 return NULL;
768 release_pack_memory(length);
769 ret = mmap(start, length, prot, flags, fd, offset);
770 }
771 return ret;
772}
773
774void *xmmap(void *start, size_t length,
775 int prot, int flags, int fd, off_t offset)
776{
777 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
778 if (ret == MAP_FAILED)
779 die_errno("mmap failed");
780 return ret;
781}
782
783/*
784 * With an in-core object data in "map", rehash it to make sure the
785 * object name actually matches "sha1" to detect object corruption.
786 * With "map" == NULL, try reading the object named with "sha1" using
787 * the streaming interface and rehash it to do the same.
788 */
789int check_sha1_signature(const unsigned char *sha1, void *map,
790 unsigned long size, const char *type)
791{
792 unsigned char real_sha1[20];
793 enum object_type obj_type;
794 struct git_istream *st;
795 git_SHA_CTX c;
796 char hdr[32];
797 int hdrlen;
798
799 if (map) {
800 hash_sha1_file(map, size, type, real_sha1);
801 return hashcmp(sha1, real_sha1) ? -1 : 0;
802 }
803
804 st = open_istream(sha1, &obj_type, &size, NULL);
805 if (!st)
806 return -1;
807
808 /* Generate the header */
809 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
810
811 /* Sha1.. */
812 git_SHA1_Init(&c);
813 git_SHA1_Update(&c, hdr, hdrlen);
814 for (;;) {
815 char buf[1024 * 16];
816 ssize_t readlen = read_istream(st, buf, sizeof(buf));
817
818 if (readlen < 0) {
819 close_istream(st);
820 return -1;
821 }
822 if (!readlen)
823 break;
824 git_SHA1_Update(&c, buf, readlen);
825 }
826 git_SHA1_Final(real_sha1, &c);
827 close_istream(st);
828 return hashcmp(sha1, real_sha1) ? -1 : 0;
829}
830
831int git_open_cloexec(const char *name, int flags)
832{
833 int fd;
834 static int o_cloexec = O_CLOEXEC;
835
836 fd = open(name, flags | o_cloexec);
837 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
838 /* Try again w/o O_CLOEXEC: the kernel might not support it */
839 o_cloexec &= ~O_CLOEXEC;
840 fd = open(name, flags | o_cloexec);
841 }
842
843#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
844 {
845 static int fd_cloexec = FD_CLOEXEC;
846
847 if (!o_cloexec && 0 <= fd && fd_cloexec) {
848 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
849 int flags = fcntl(fd, F_GETFD);
850 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
851 fd_cloexec = 0;
852 }
853 }
854#endif
855 return fd;
856}
857
858/*
859 * Find "sha1" as a loose object in the local repository or in an alternate.
860 * Returns 0 on success, negative on failure.
861 *
862 * The "path" out-parameter will give the path of the object we found (if any).
863 * Note that it may point to static storage and is only valid until another
864 * call to sha1_file_name(), etc.
865 */
866static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
867 const char **path)
868{
869 struct alternate_object_database *alt;
870 static struct strbuf buf = STRBUF_INIT;
871
872 strbuf_reset(&buf);
873 sha1_file_name(&buf, sha1);
874 *path = buf.buf;
875
876 if (!lstat(*path, st))
877 return 0;
878
879 prepare_alt_odb();
880 errno = ENOENT;
881 for (alt = alt_odb_list; alt; alt = alt->next) {
882 *path = alt_sha1_path(alt, sha1);
883 if (!lstat(*path, st))
884 return 0;
885 }
886
887 return -1;
888}
889
890/*
891 * Like stat_sha1_file(), but actually open the object and return the
892 * descriptor. See the caveats on the "path" parameter above.
893 */
894static int open_sha1_file(const unsigned char *sha1, const char **path)
895{
896 int fd;
897 struct alternate_object_database *alt;
898 int most_interesting_errno;
899 static struct strbuf buf = STRBUF_INIT;
900
901 strbuf_reset(&buf);
902 sha1_file_name(&buf, sha1);
903 *path = buf.buf;
904
905 fd = git_open(*path);
906 if (fd >= 0)
907 return fd;
908 most_interesting_errno = errno;
909
910 prepare_alt_odb();
911 for (alt = alt_odb_list; alt; alt = alt->next) {
912 *path = alt_sha1_path(alt, sha1);
913 fd = git_open(*path);
914 if (fd >= 0)
915 return fd;
916 if (most_interesting_errno == ENOENT)
917 most_interesting_errno = errno;
918 }
919 errno = most_interesting_errno;
920 return -1;
921}
922
923/*
924 * Map the loose object at "path" if it is not NULL, or the path found by
925 * searching for a loose object named "sha1".
926 */
927static void *map_sha1_file_1(const char *path,
928 const unsigned char *sha1,
929 unsigned long *size)
930{
931 void *map;
932 int fd;
933
934 if (path)
935 fd = git_open(path);
936 else
937 fd = open_sha1_file(sha1, &path);
938 map = NULL;
939 if (fd >= 0) {
940 struct stat st;
941
942 if (!fstat(fd, &st)) {
943 *size = xsize_t(st.st_size);
944 if (!*size) {
945 /* mmap() is forbidden on empty files */
946 error("object file %s is empty", path);
947 return NULL;
948 }
949 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
950 }
951 close(fd);
952 }
953 return map;
954}
955
956void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
957{
958 return map_sha1_file_1(NULL, sha1, size);
959}
960
961static int unpack_sha1_short_header(git_zstream *stream,
962 unsigned char *map, unsigned long mapsize,
963 void *buffer, unsigned long bufsiz)
964{
965 /* Get the data stream */
966 memset(stream, 0, sizeof(*stream));
967 stream->next_in = map;
968 stream->avail_in = mapsize;
969 stream->next_out = buffer;
970 stream->avail_out = bufsiz;
971
972 git_inflate_init(stream);
973 return git_inflate(stream, 0);
974}
975
976int unpack_sha1_header(git_zstream *stream,
977 unsigned char *map, unsigned long mapsize,
978 void *buffer, unsigned long bufsiz)
979{
980 int status = unpack_sha1_short_header(stream, map, mapsize,
981 buffer, bufsiz);
982
983 if (status < Z_OK)
984 return status;
985
986 /* Make sure we have the terminating NUL */
987 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
988 return -1;
989 return 0;
990}
991
992static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
993 unsigned long mapsize, void *buffer,
994 unsigned long bufsiz, struct strbuf *header)
995{
996 int status;
997
998 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
999 if (status < Z_OK)
1000 return -1;
1001
1002 /*
1003 * Check if entire header is unpacked in the first iteration.
1004 */
1005 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1006 return 0;
1007
1008 /*
1009 * buffer[0..bufsiz] was not large enough. Copy the partial
1010 * result out to header, and then append the result of further
1011 * reading the stream.
1012 */
1013 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1014 stream->next_out = buffer;
1015 stream->avail_out = bufsiz;
1016
1017 do {
1018 status = git_inflate(stream, 0);
1019 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1020 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1021 return 0;
1022 stream->next_out = buffer;
1023 stream->avail_out = bufsiz;
1024 } while (status != Z_STREAM_END);
1025 return -1;
1026}
1027
1028static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1029{
1030 int bytes = strlen(buffer) + 1;
1031 unsigned char *buf = xmallocz(size);
1032 unsigned long n;
1033 int status = Z_OK;
1034
1035 n = stream->total_out - bytes;
1036 if (n > size)
1037 n = size;
1038 memcpy(buf, (char *) buffer + bytes, n);
1039 bytes = n;
1040 if (bytes <= size) {
1041 /*
1042 * The above condition must be (bytes <= size), not
1043 * (bytes < size). In other words, even though we
1044 * expect no more output and set avail_out to zero,
1045 * the input zlib stream may have bytes that express
1046 * "this concludes the stream", and we *do* want to
1047 * eat that input.
1048 *
1049 * Otherwise we would not be able to test that we
1050 * consumed all the input to reach the expected size;
1051 * we also want to check that zlib tells us that all
1052 * went well with status == Z_STREAM_END at the end.
1053 */
1054 stream->next_out = buf + bytes;
1055 stream->avail_out = size - bytes;
1056 while (status == Z_OK)
1057 status = git_inflate(stream, Z_FINISH);
1058 }
1059 if (status == Z_STREAM_END && !stream->avail_in) {
1060 git_inflate_end(stream);
1061 return buf;
1062 }
1063
1064 if (status < 0)
1065 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1066 else if (stream->avail_in)
1067 error("garbage at end of loose object '%s'",
1068 sha1_to_hex(sha1));
1069 free(buf);
1070 return NULL;
1071}
1072
1073/*
1074 * We used to just use "sscanf()", but that's actually way
1075 * too permissive for what we want to check. So do an anal
1076 * object header parse by hand.
1077 */
1078static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1079 unsigned int flags)
1080{
1081 const char *type_buf = hdr;
1082 unsigned long size;
1083 int type, type_len = 0;
1084
1085 /*
1086 * The type can be of any size but is followed by
1087 * a space.
1088 */
1089 for (;;) {
1090 char c = *hdr++;
1091 if (!c)
1092 return -1;
1093 if (c == ' ')
1094 break;
1095 type_len++;
1096 }
1097
1098 type = type_from_string_gently(type_buf, type_len, 1);
1099 if (oi->typename)
1100 strbuf_add(oi->typename, type_buf, type_len);
1101 /*
1102 * Set type to 0 if its an unknown object and
1103 * we're obtaining the type using '--allow-unknown-type'
1104 * option.
1105 */
1106 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1107 type = 0;
1108 else if (type < 0)
1109 die("invalid object type");
1110 if (oi->typep)
1111 *oi->typep = type;
1112
1113 /*
1114 * The length must follow immediately, and be in canonical
1115 * decimal format (ie "010" is not valid).
1116 */
1117 size = *hdr++ - '0';
1118 if (size > 9)
1119 return -1;
1120 if (size) {
1121 for (;;) {
1122 unsigned long c = *hdr - '0';
1123 if (c > 9)
1124 break;
1125 hdr++;
1126 size = size * 10 + c;
1127 }
1128 }
1129
1130 if (oi->sizep)
1131 *oi->sizep = size;
1132
1133 /*
1134 * The length must be followed by a zero byte
1135 */
1136 return *hdr ? -1 : type;
1137}
1138
1139int parse_sha1_header(const char *hdr, unsigned long *sizep)
1140{
1141 struct object_info oi = OBJECT_INFO_INIT;
1142
1143 oi.sizep = sizep;
1144 return parse_sha1_header_extended(hdr, &oi, 0);
1145}
1146
1147static int sha1_loose_object_info(const unsigned char *sha1,
1148 struct object_info *oi,
1149 int flags)
1150{
1151 int status = 0;
1152 unsigned long mapsize;
1153 void *map;
1154 git_zstream stream;
1155 char hdr[32];
1156 struct strbuf hdrbuf = STRBUF_INIT;
1157 unsigned long size_scratch;
1158
1159 if (oi->delta_base_sha1)
1160 hashclr(oi->delta_base_sha1);
1161
1162 /*
1163 * If we don't care about type or size, then we don't
1164 * need to look inside the object at all. Note that we
1165 * do not optimize out the stat call, even if the
1166 * caller doesn't care about the disk-size, since our
1167 * return value implicitly indicates whether the
1168 * object even exists.
1169 */
1170 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1171 const char *path;
1172 struct stat st;
1173 if (stat_sha1_file(sha1, &st, &path) < 0)
1174 return -1;
1175 if (oi->disk_sizep)
1176 *oi->disk_sizep = st.st_size;
1177 return 0;
1178 }
1179
1180 map = map_sha1_file(sha1, &mapsize);
1181 if (!map)
1182 return -1;
1183
1184 if (!oi->sizep)
1185 oi->sizep = &size_scratch;
1186
1187 if (oi->disk_sizep)
1188 *oi->disk_sizep = mapsize;
1189 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1190 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1191 status = error("unable to unpack %s header with --allow-unknown-type",
1192 sha1_to_hex(sha1));
1193 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1194 status = error("unable to unpack %s header",
1195 sha1_to_hex(sha1));
1196 if (status < 0)
1197 ; /* Do nothing */
1198 else if (hdrbuf.len) {
1199 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1200 status = error("unable to parse %s header with --allow-unknown-type",
1201 sha1_to_hex(sha1));
1202 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1203 status = error("unable to parse %s header", sha1_to_hex(sha1));
1204
1205 if (status >= 0 && oi->contentp) {
1206 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1207 *oi->sizep, sha1);
1208 if (!*oi->contentp) {
1209 git_inflate_end(&stream);
1210 status = -1;
1211 }
1212 } else
1213 git_inflate_end(&stream);
1214
1215 munmap(map, mapsize);
1216 if (status && oi->typep)
1217 *oi->typep = status;
1218 if (oi->sizep == &size_scratch)
1219 oi->sizep = NULL;
1220 strbuf_release(&hdrbuf);
1221 oi->whence = OI_LOOSE;
1222 return (status < 0) ? status : 0;
1223}
1224
1225int fetch_if_missing = 1;
1226
1227int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1228{
1229 static struct object_info blank_oi = OBJECT_INFO_INIT;
1230 struct pack_entry e;
1231 int rtype;
1232 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1233 lookup_replace_object(sha1) :
1234 sha1;
1235 int already_retried = 0;
1236
1237 if (is_null_sha1(real))
1238 return -1;
1239
1240 if (!oi)
1241 oi = &blank_oi;
1242
1243 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1244 struct cached_object *co = find_cached_object(real);
1245 if (co) {
1246 if (oi->typep)
1247 *(oi->typep) = co->type;
1248 if (oi->sizep)
1249 *(oi->sizep) = co->size;
1250 if (oi->disk_sizep)
1251 *(oi->disk_sizep) = 0;
1252 if (oi->delta_base_sha1)
1253 hashclr(oi->delta_base_sha1);
1254 if (oi->typename)
1255 strbuf_addstr(oi->typename, typename(co->type));
1256 if (oi->contentp)
1257 *oi->contentp = xmemdupz(co->buf, co->size);
1258 oi->whence = OI_CACHED;
1259 return 0;
1260 }
1261 }
1262
1263 while (1) {
1264 if (find_pack_entry(real, &e))
1265 break;
1266
1267 /* Most likely it's a loose object. */
1268 if (!sha1_loose_object_info(real, oi, flags))
1269 return 0;
1270
1271 /* Not a loose object; someone else may have just packed it. */
1272 reprepare_packed_git();
1273 if (find_pack_entry(real, &e))
1274 break;
1275
1276 /* Check if it is a missing object */
1277 if (fetch_if_missing && repository_format_partial_clone &&
1278 !already_retried) {
1279 /*
1280 * TODO Investigate haveing fetch_object() return
1281 * TODO error/success and stopping the music here.
1282 */
1283 fetch_object(repository_format_partial_clone, real);
1284 already_retried = 1;
1285 continue;
1286 }
1287
1288 return -1;
1289 }
1290
1291 if (oi == &blank_oi)
1292 /*
1293 * We know that the caller doesn't actually need the
1294 * information below, so return early.
1295 */
1296 return 0;
1297 rtype = packed_object_info(e.p, e.offset, oi);
1298 if (rtype < 0) {
1299 mark_bad_packed_object(e.p, real);
1300 return sha1_object_info_extended(real, oi, 0);
1301 } else if (oi->whence == OI_PACKED) {
1302 oi->u.packed.offset = e.offset;
1303 oi->u.packed.pack = e.p;
1304 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1305 rtype == OBJ_OFS_DELTA);
1306 }
1307
1308 return 0;
1309}
1310
1311/* returns enum object_type or negative */
1312int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1313{
1314 enum object_type type;
1315 struct object_info oi = OBJECT_INFO_INIT;
1316
1317 oi.typep = &type;
1318 oi.sizep = sizep;
1319 if (sha1_object_info_extended(sha1, &oi,
1320 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1321 return -1;
1322 return type;
1323}
1324
1325static void *read_object(const unsigned char *sha1, enum object_type *type,
1326 unsigned long *size)
1327{
1328 struct object_info oi = OBJECT_INFO_INIT;
1329 void *content;
1330 oi.typep = type;
1331 oi.sizep = size;
1332 oi.contentp = &content;
1333
1334 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1335 return NULL;
1336 return content;
1337}
1338
1339int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1340 unsigned char *sha1)
1341{
1342 struct cached_object *co;
1343
1344 hash_sha1_file(buf, len, typename(type), sha1);
1345 if (has_sha1_file(sha1) || find_cached_object(sha1))
1346 return 0;
1347 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1348 co = &cached_objects[cached_object_nr++];
1349 co->size = len;
1350 co->type = type;
1351 co->buf = xmalloc(len);
1352 memcpy(co->buf, buf, len);
1353 hashcpy(co->sha1, sha1);
1354 return 0;
1355}
1356
1357/*
1358 * This function dies on corrupt objects; the callers who want to
1359 * deal with them should arrange to call read_object() and give error
1360 * messages themselves.
1361 */
1362void *read_sha1_file_extended(const unsigned char *sha1,
1363 enum object_type *type,
1364 unsigned long *size,
1365 int lookup_replace)
1366{
1367 void *data;
1368 const struct packed_git *p;
1369 const char *path;
1370 struct stat st;
1371 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1372 : sha1;
1373
1374 errno = 0;
1375 data = read_object(repl, type, size);
1376 if (data)
1377 return data;
1378
1379 if (errno && errno != ENOENT)
1380 die_errno("failed to read object %s", sha1_to_hex(sha1));
1381
1382 /* die if we replaced an object with one that does not exist */
1383 if (repl != sha1)
1384 die("replacement %s not found for %s",
1385 sha1_to_hex(repl), sha1_to_hex(sha1));
1386
1387 if (!stat_sha1_file(repl, &st, &path))
1388 die("loose object %s (stored in %s) is corrupt",
1389 sha1_to_hex(repl), path);
1390
1391 if ((p = has_packed_and_bad(repl)) != NULL)
1392 die("packed object %s (stored in %s) is corrupt",
1393 sha1_to_hex(repl), p->pack_name);
1394
1395 return NULL;
1396}
1397
1398void *read_object_with_reference(const unsigned char *sha1,
1399 const char *required_type_name,
1400 unsigned long *size,
1401 unsigned char *actual_sha1_return)
1402{
1403 enum object_type type, required_type;
1404 void *buffer;
1405 unsigned long isize;
1406 unsigned char actual_sha1[20];
1407
1408 required_type = type_from_string(required_type_name);
1409 hashcpy(actual_sha1, sha1);
1410 while (1) {
1411 int ref_length = -1;
1412 const char *ref_type = NULL;
1413
1414 buffer = read_sha1_file(actual_sha1, &type, &isize);
1415 if (!buffer)
1416 return NULL;
1417 if (type == required_type) {
1418 *size = isize;
1419 if (actual_sha1_return)
1420 hashcpy(actual_sha1_return, actual_sha1);
1421 return buffer;
1422 }
1423 /* Handle references */
1424 else if (type == OBJ_COMMIT)
1425 ref_type = "tree ";
1426 else if (type == OBJ_TAG)
1427 ref_type = "object ";
1428 else {
1429 free(buffer);
1430 return NULL;
1431 }
1432 ref_length = strlen(ref_type);
1433
1434 if (ref_length + 40 > isize ||
1435 memcmp(buffer, ref_type, ref_length) ||
1436 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1437 free(buffer);
1438 return NULL;
1439 }
1440 free(buffer);
1441 /* Now we have the ID of the referred-to object in
1442 * actual_sha1. Check again. */
1443 }
1444}
1445
1446static void write_sha1_file_prepare(const void *buf, unsigned long len,
1447 const char *type, unsigned char *sha1,
1448 char *hdr, int *hdrlen)
1449{
1450 git_SHA_CTX c;
1451
1452 /* Generate the header */
1453 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1454
1455 /* Sha1.. */
1456 git_SHA1_Init(&c);
1457 git_SHA1_Update(&c, hdr, *hdrlen);
1458 git_SHA1_Update(&c, buf, len);
1459 git_SHA1_Final(sha1, &c);
1460}
1461
1462/*
1463 * Move the just written object into its final resting place.
1464 */
1465int finalize_object_file(const char *tmpfile, const char *filename)
1466{
1467 int ret = 0;
1468
1469 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1470 goto try_rename;
1471 else if (link(tmpfile, filename))
1472 ret = errno;
1473
1474 /*
1475 * Coda hack - coda doesn't like cross-directory links,
1476 * so we fall back to a rename, which will mean that it
1477 * won't be able to check collisions, but that's not a
1478 * big deal.
1479 *
1480 * The same holds for FAT formatted media.
1481 *
1482 * When this succeeds, we just return. We have nothing
1483 * left to unlink.
1484 */
1485 if (ret && ret != EEXIST) {
1486 try_rename:
1487 if (!rename(tmpfile, filename))
1488 goto out;
1489 ret = errno;
1490 }
1491 unlink_or_warn(tmpfile);
1492 if (ret) {
1493 if (ret != EEXIST) {
1494 return error_errno("unable to write sha1 filename %s", filename);
1495 }
1496 /* FIXME!!! Collision check here ? */
1497 }
1498
1499out:
1500 if (adjust_shared_perm(filename))
1501 return error("unable to set permission to '%s'", filename);
1502 return 0;
1503}
1504
1505static int write_buffer(int fd, const void *buf, size_t len)
1506{
1507 if (write_in_full(fd, buf, len) < 0)
1508 return error_errno("file write error");
1509 return 0;
1510}
1511
1512int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1513 unsigned char *sha1)
1514{
1515 char hdr[32];
1516 int hdrlen = sizeof(hdr);
1517 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1518 return 0;
1519}
1520
1521/* Finalize a file on disk, and close it. */
1522static void close_sha1_file(int fd)
1523{
1524 if (fsync_object_files)
1525 fsync_or_die(fd, "sha1 file");
1526 if (close(fd) != 0)
1527 die_errno("error when closing sha1 file");
1528}
1529
1530/* Size of directory component, including the ending '/' */
1531static inline int directory_size(const char *filename)
1532{
1533 const char *s = strrchr(filename, '/');
1534 if (!s)
1535 return 0;
1536 return s - filename + 1;
1537}
1538
1539/*
1540 * This creates a temporary file in the same directory as the final
1541 * 'filename'
1542 *
1543 * We want to avoid cross-directory filename renames, because those
1544 * can have problems on various filesystems (FAT, NFS, Coda).
1545 */
1546static int create_tmpfile(struct strbuf *tmp, const char *filename)
1547{
1548 int fd, dirlen = directory_size(filename);
1549
1550 strbuf_reset(tmp);
1551 strbuf_add(tmp, filename, dirlen);
1552 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1553 fd = git_mkstemp_mode(tmp->buf, 0444);
1554 if (fd < 0 && dirlen && errno == ENOENT) {
1555 /*
1556 * Make sure the directory exists; note that the contents
1557 * of the buffer are undefined after mkstemp returns an
1558 * error, so we have to rewrite the whole buffer from
1559 * scratch.
1560 */
1561 strbuf_reset(tmp);
1562 strbuf_add(tmp, filename, dirlen - 1);
1563 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1564 return -1;
1565 if (adjust_shared_perm(tmp->buf))
1566 return -1;
1567
1568 /* Try again */
1569 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1570 fd = git_mkstemp_mode(tmp->buf, 0444);
1571 }
1572 return fd;
1573}
1574
1575static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1576 const void *buf, unsigned long len, time_t mtime)
1577{
1578 int fd, ret;
1579 unsigned char compressed[4096];
1580 git_zstream stream;
1581 git_SHA_CTX c;
1582 unsigned char parano_sha1[20];
1583 static struct strbuf tmp_file = STRBUF_INIT;
1584 static struct strbuf filename = STRBUF_INIT;
1585
1586 strbuf_reset(&filename);
1587 sha1_file_name(&filename, sha1);
1588
1589 fd = create_tmpfile(&tmp_file, filename.buf);
1590 if (fd < 0) {
1591 if (errno == EACCES)
1592 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1593 else
1594 return error_errno("unable to create temporary file");
1595 }
1596
1597 /* Set it up */
1598 git_deflate_init(&stream, zlib_compression_level);
1599 stream.next_out = compressed;
1600 stream.avail_out = sizeof(compressed);
1601 git_SHA1_Init(&c);
1602
1603 /* First header.. */
1604 stream.next_in = (unsigned char *)hdr;
1605 stream.avail_in = hdrlen;
1606 while (git_deflate(&stream, 0) == Z_OK)
1607 ; /* nothing */
1608 git_SHA1_Update(&c, hdr, hdrlen);
1609
1610 /* Then the data itself.. */
1611 stream.next_in = (void *)buf;
1612 stream.avail_in = len;
1613 do {
1614 unsigned char *in0 = stream.next_in;
1615 ret = git_deflate(&stream, Z_FINISH);
1616 git_SHA1_Update(&c, in0, stream.next_in - in0);
1617 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1618 die("unable to write sha1 file");
1619 stream.next_out = compressed;
1620 stream.avail_out = sizeof(compressed);
1621 } while (ret == Z_OK);
1622
1623 if (ret != Z_STREAM_END)
1624 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1625 ret = git_deflate_end_gently(&stream);
1626 if (ret != Z_OK)
1627 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1628 git_SHA1_Final(parano_sha1, &c);
1629 if (hashcmp(sha1, parano_sha1) != 0)
1630 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1631
1632 close_sha1_file(fd);
1633
1634 if (mtime) {
1635 struct utimbuf utb;
1636 utb.actime = mtime;
1637 utb.modtime = mtime;
1638 if (utime(tmp_file.buf, &utb) < 0)
1639 warning_errno("failed utime() on %s", tmp_file.buf);
1640 }
1641
1642 return finalize_object_file(tmp_file.buf, filename.buf);
1643}
1644
1645static int freshen_loose_object(const unsigned char *sha1)
1646{
1647 return check_and_freshen(sha1, 1);
1648}
1649
1650static int freshen_packed_object(const unsigned char *sha1)
1651{
1652 struct pack_entry e;
1653 if (!find_pack_entry(sha1, &e))
1654 return 0;
1655 if (e.p->freshened)
1656 return 1;
1657 if (!freshen_file(e.p->pack_name))
1658 return 0;
1659 e.p->freshened = 1;
1660 return 1;
1661}
1662
1663int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1664{
1665 char hdr[32];
1666 int hdrlen = sizeof(hdr);
1667
1668 /* Normally if we have it in the pack then we do not bother writing
1669 * it out into .git/objects/??/?{38} file.
1670 */
1671 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1672 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1673 return 0;
1674 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1675}
1676
1677int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1678 struct object_id *oid, 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_sha1_file_prepare(buf, len, type, oid->hash, 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->hash, header, hdrlen, buf, len, 0);
1693
1694cleanup:
1695 free(header);
1696 return status;
1697}
1698
1699int force_object_loose(const unsigned char *sha1, 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(sha1))
1709 return 0;
1710 buf = read_object(sha1, &type, &len);
1711 if (!buf)
1712 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1713 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1714 ret = write_loose_object(sha1, 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_sha1_file(buf, size, typename(type), oid->hash);
1798 else
1799 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
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_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1820 oid->hash);
1821 else
1822 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1823 oid->hash);
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_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1938 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
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 typename(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_SHA_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 git_SHA1_Init(&c);
2130 git_SHA1_Update(&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 git_SHA1_Update(&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 git_SHA1_Final(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, typename(*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}