abd3b59b578fc2103c71b45bc74be492917bf3a0
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9#include "cache.h"
10#include "config.h"
11#include "string-list.h"
12#include "lockfile.h"
13#include "delta.h"
14#include "pack.h"
15#include "blob.h"
16#include "commit.h"
17#include "run-command.h"
18#include "tag.h"
19#include "tree.h"
20#include "tree-walk.h"
21#include "refs.h"
22#include "pack-revindex.h"
23#include "sha1-lookup.h"
24#include "bulk-checkin.h"
25#include "streaming.h"
26#include "dir.h"
27#include "mru.h"
28#include "list.h"
29#include "mergesort.h"
30#include "quote.h"
31#include "packfile.h"
32
33const unsigned char null_sha1[20];
34const struct object_id null_oid;
35const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
37};
38const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
40};
41
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
60 0
61};
62
63static struct cached_object *find_cached_object(const unsigned char *sha1)
64{
65 int i;
66 struct cached_object *co = cached_objects;
67
68 for (i = 0; i < cached_object_nr; i++, co++) {
69 if (!hashcmp(co->sha1, sha1))
70 return co;
71 }
72 if (!hashcmp(sha1, empty_tree.sha1))
73 return &empty_tree;
74 return NULL;
75}
76
77int mkdir_in_gitdir(const char *path)
78{
79 if (mkdir(path, 0777)) {
80 int saved_errno = errno;
81 struct stat st;
82 struct strbuf sb = STRBUF_INIT;
83
84 if (errno != EEXIST)
85 return -1;
86 /*
87 * Are we looking at a path in a symlinked worktree
88 * whose original repository does not yet have it?
89 * e.g. .git/rr-cache pointing at its original
90 * repository in which the user hasn't performed any
91 * conflict resolution yet?
92 */
93 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
94 strbuf_readlink(&sb, path, st.st_size) ||
95 !is_absolute_path(sb.buf) ||
96 mkdir(sb.buf, 0777)) {
97 strbuf_release(&sb);
98 errno = saved_errno;
99 return -1;
100 }
101 strbuf_release(&sb);
102 }
103 return adjust_shared_perm(path);
104}
105
106enum scld_error safe_create_leading_directories(char *path)
107{
108 char *next_component = path + offset_1st_component(path);
109 enum scld_error ret = SCLD_OK;
110
111 while (ret == SCLD_OK && next_component) {
112 struct stat st;
113 char *slash = next_component, slash_character;
114
115 while (*slash && !is_dir_sep(*slash))
116 slash++;
117
118 if (!*slash)
119 break;
120
121 next_component = slash + 1;
122 while (is_dir_sep(*next_component))
123 next_component++;
124 if (!*next_component)
125 break;
126
127 slash_character = *slash;
128 *slash = '\0';
129 if (!stat(path, &st)) {
130 /* path exists */
131 if (!S_ISDIR(st.st_mode)) {
132 errno = ENOTDIR;
133 ret = SCLD_EXISTS;
134 }
135 } else if (mkdir(path, 0777)) {
136 if (errno == EEXIST &&
137 !stat(path, &st) && S_ISDIR(st.st_mode))
138 ; /* somebody created it since we checked */
139 else if (errno == ENOENT)
140 /*
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
147 * trying again:
148 */
149 ret = SCLD_VANISHED;
150 else
151 ret = SCLD_FAILED;
152 } else if (adjust_shared_perm(path)) {
153 ret = SCLD_PERMS;
154 }
155 *slash = slash_character;
156 }
157 return ret;
158}
159
160enum scld_error safe_create_leading_directories_const(const char *path)
161{
162 int save_errno;
163 /* path points to cache entries, so xstrdup before messing with it */
164 char *buf = xstrdup(path);
165 enum scld_error result = safe_create_leading_directories(buf);
166
167 save_errno = errno;
168 free(buf);
169 errno = save_errno;
170 return result;
171}
172
173int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
174{
175 /*
176 * The number of times we will try to remove empty directories
177 * in the way of path. This is only 1 because if another
178 * process is racily creating directories that conflict with
179 * us, we don't want to fight against them.
180 */
181 int remove_directories_remaining = 1;
182
183 /*
184 * The number of times that we will try to create the
185 * directories containing path. We are willing to attempt this
186 * more than once, because another process could be trying to
187 * clean up empty directories at the same time as we are
188 * trying to create them.
189 */
190 int create_directories_remaining = 3;
191
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy = STRBUF_INIT;
194
195 int ret, save_errno;
196
197 /* Sanity check: */
198 assert(*path);
199
200retry_fn:
201 ret = fn(path, cb);
202 save_errno = errno;
203 if (!ret)
204 goto out;
205
206 if (errno == EISDIR && remove_directories_remaining-- > 0) {
207 /*
208 * A directory is in the way. Maybe it is empty; try
209 * to remove it:
210 */
211 if (!path_copy.len)
212 strbuf_addstr(&path_copy, path);
213
214 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
215 goto retry_fn;
216 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
217 /*
218 * Maybe the containing directory didn't exist, or
219 * maybe it was just deleted by a process that is
220 * racing with us to clean up empty directories. Try
221 * to create it:
222 */
223 enum scld_error scld_result;
224
225 if (!path_copy.len)
226 strbuf_addstr(&path_copy, path);
227
228 do {
229 scld_result = safe_create_leading_directories(path_copy.buf);
230 if (scld_result == SCLD_OK)
231 goto retry_fn;
232 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
233 }
234
235out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
239}
240
241static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
242{
243 int i;
244 for (i = 0; i < 20; i++) {
245 static char hex[] = "0123456789abcdef";
246 unsigned int val = sha1[i];
247 strbuf_addch(buf, hex[val >> 4]);
248 strbuf_addch(buf, hex[val & 0xf]);
249 if (!i)
250 strbuf_addch(buf, '/');
251 }
252}
253
254const char *sha1_file_name(const unsigned char *sha1)
255{
256 static struct strbuf buf = STRBUF_INIT;
257
258 strbuf_reset(&buf);
259 strbuf_addf(&buf, "%s/", get_object_directory());
260
261 fill_sha1_path(&buf, sha1);
262 return buf.buf;
263}
264
265struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
266{
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
269}
270
271static const char *alt_sha1_path(struct alternate_object_database *alt,
272 const unsigned char *sha1)
273{
274 struct strbuf *buf = alt_scratch_buf(alt);
275 fill_sha1_path(buf, sha1);
276 return buf->buf;
277}
278
279struct alternate_object_database *alt_odb_list;
280static struct alternate_object_database **alt_odb_tail;
281
282/*
283 * Return non-zero iff the path is usable as an alternate object database.
284 */
285static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
286{
287 struct alternate_object_database *alt;
288
289 /* Detect cases where alternate disappeared */
290 if (!is_directory(path->buf)) {
291 error("object directory %s does not exist; "
292 "check .git/objects/info/alternates.",
293 path->buf);
294 return 0;
295 }
296
297 /*
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
300 */
301 for (alt = alt_odb_list; alt; alt = alt->next) {
302 if (!fspathcmp(path->buf, alt->path))
303 return 0;
304 }
305 if (!fspathcmp(path->buf, normalized_objdir))
306 return 0;
307
308 return 1;
309}
310
311/*
312 * Prepare alternate object database registry.
313 *
314 * The variable alt_odb_list points at the list of struct
315 * alternate_object_database. The elements on this list come from
316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
318 * whose contents is similar to that environment variable but can be
319 * LF separated. Its base points at a statically allocated buffer that
320 * contains "/the/directory/corresponding/to/.git/objects/...", while
321 * its name points just after the slash at the end of ".git/objects/"
322 * in the example above, and has enough space to hold 40-byte hex
323 * SHA1, an extra slash for the first level indirection, and the
324 * terminating NUL.
325 */
326static void read_info_alternates(const char * relative_base, int depth);
327static int link_alt_odb_entry(const char *entry, const char *relative_base,
328 int depth, const char *normalized_objdir)
329{
330 struct alternate_object_database *ent;
331 struct strbuf pathbuf = STRBUF_INIT;
332
333 if (!is_absolute_path(entry) && relative_base) {
334 strbuf_realpath(&pathbuf, relative_base, 1);
335 strbuf_addch(&pathbuf, '/');
336 }
337 strbuf_addstr(&pathbuf, entry);
338
339 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
340 error("unable to normalize alternate object path: %s",
341 pathbuf.buf);
342 strbuf_release(&pathbuf);
343 return -1;
344 }
345
346 /*
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
349 */
350 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
351 strbuf_setlen(&pathbuf, pathbuf.len - 1);
352
353 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
354 strbuf_release(&pathbuf);
355 return -1;
356 }
357
358 ent = alloc_alt_odb(pathbuf.buf);
359
360 /* add the alternate entry */
361 *alt_odb_tail = ent;
362 alt_odb_tail = &(ent->next);
363 ent->next = NULL;
364
365 /* recursively add alternates */
366 read_info_alternates(pathbuf.buf, depth + 1);
367
368 strbuf_release(&pathbuf);
369 return 0;
370}
371
372static const char *parse_alt_odb_entry(const char *string,
373 int sep,
374 struct strbuf *out)
375{
376 const char *end;
377
378 strbuf_reset(out);
379
380 if (*string == '#') {
381 /* comment; consume up to next separator */
382 end = strchrnul(string, sep);
383 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
384 /*
385 * quoted path; unquote_c_style has copied the
386 * data for us and set "end". Broken quoting (e.g.,
387 * an entry that doesn't end with a quote) falls
388 * back to the unquoted case below.
389 */
390 } else {
391 /* normal, unquoted path */
392 end = strchrnul(string, sep);
393 strbuf_add(out, string, end - string);
394 }
395
396 if (*end)
397 end++;
398 return end;
399}
400
401static void link_alt_odb_entries(const char *alt, int len, int sep,
402 const char *relative_base, int depth)
403{
404 struct strbuf objdirbuf = STRBUF_INIT;
405 struct strbuf entry = STRBUF_INIT;
406
407 if (depth > 5) {
408 error("%s: ignoring alternate object stores, nesting too deep.",
409 relative_base);
410 return;
411 }
412
413 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf) < 0)
415 die("unable to normalize object directory: %s",
416 objdirbuf.buf);
417
418 while (*alt) {
419 alt = parse_alt_odb_entry(alt, sep, &entry);
420 if (!entry.len)
421 continue;
422 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
423 }
424 strbuf_release(&entry);
425 strbuf_release(&objdirbuf);
426}
427
428static void read_info_alternates(const char * relative_base, int depth)
429{
430 char *map;
431 size_t mapsz;
432 struct stat st;
433 char *path;
434 int fd;
435
436 path = xstrfmt("%s/info/alternates", relative_base);
437 fd = git_open(path);
438 free(path);
439 if (fd < 0)
440 return;
441 if (fstat(fd, &st) || (st.st_size == 0)) {
442 close(fd);
443 return;
444 }
445 mapsz = xsize_t(st.st_size);
446 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
447 close(fd);
448
449 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
450
451 munmap(map, mapsz);
452}
453
454struct alternate_object_database *alloc_alt_odb(const char *dir)
455{
456 struct alternate_object_database *ent;
457
458 FLEX_ALLOC_STR(ent, path, dir);
459 strbuf_init(&ent->scratch, 0);
460 strbuf_addf(&ent->scratch, "%s/", dir);
461 ent->base_len = ent->scratch.len;
462
463 return ent;
464}
465
466void add_to_alternates_file(const char *reference)
467{
468 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
469 char *alts = git_pathdup("objects/info/alternates");
470 FILE *in, *out;
471
472 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
473 out = fdopen_lock_file(lock, "w");
474 if (!out)
475 die_errno("unable to fdopen alternates lockfile");
476
477 in = fopen(alts, "r");
478 if (in) {
479 struct strbuf line = STRBUF_INIT;
480 int found = 0;
481
482 while (strbuf_getline(&line, in) != EOF) {
483 if (!strcmp(reference, line.buf)) {
484 found = 1;
485 break;
486 }
487 fprintf_or_die(out, "%s\n", line.buf);
488 }
489
490 strbuf_release(&line);
491 fclose(in);
492
493 if (found) {
494 rollback_lock_file(lock);
495 lock = NULL;
496 }
497 }
498 else if (errno != ENOENT)
499 die_errno("unable to read alternates file");
500
501 if (lock) {
502 fprintf_or_die(out, "%s\n", reference);
503 if (commit_lock_file(lock))
504 die_errno("unable to move new alternates file into place");
505 if (alt_odb_tail)
506 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
507 }
508 free(alts);
509}
510
511void add_to_alternates_memory(const char *reference)
512{
513 /*
514 * Make sure alternates are initialized, or else our entry may be
515 * overwritten when they are.
516 */
517 prepare_alt_odb();
518
519 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
520}
521
522/*
523 * Compute the exact path an alternate is at and returns it. In case of
524 * error NULL is returned and the human readable error is added to `err`
525 * `path` may be relative and should point to $GITDIR.
526 * `err` must not be null.
527 */
528char *compute_alternate_path(const char *path, struct strbuf *err)
529{
530 char *ref_git = NULL;
531 const char *repo, *ref_git_s;
532 int seen_error = 0;
533
534 ref_git_s = real_path_if_valid(path);
535 if (!ref_git_s) {
536 seen_error = 1;
537 strbuf_addf(err, _("path '%s' does not exist"), path);
538 goto out;
539 } else
540 /*
541 * Beware: read_gitfile(), real_path() and mkpath()
542 * return static buffer
543 */
544 ref_git = xstrdup(ref_git_s);
545
546 repo = read_gitfile(ref_git);
547 if (!repo)
548 repo = read_gitfile(mkpath("%s/.git", ref_git));
549 if (repo) {
550 free(ref_git);
551 ref_git = xstrdup(repo);
552 }
553
554 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
555 char *ref_git_git = mkpathdup("%s/.git", ref_git);
556 free(ref_git);
557 ref_git = ref_git_git;
558 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
559 struct strbuf sb = STRBUF_INIT;
560 seen_error = 1;
561 if (get_common_dir(&sb, ref_git)) {
562 strbuf_addf(err,
563 _("reference repository '%s' as a linked "
564 "checkout is not supported yet."),
565 path);
566 goto out;
567 }
568
569 strbuf_addf(err, _("reference repository '%s' is not a "
570 "local repository."), path);
571 goto out;
572 }
573
574 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
575 strbuf_addf(err, _("reference repository '%s' is shallow"),
576 path);
577 seen_error = 1;
578 goto out;
579 }
580
581 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
582 strbuf_addf(err,
583 _("reference repository '%s' is grafted"),
584 path);
585 seen_error = 1;
586 goto out;
587 }
588
589out:
590 if (seen_error) {
591 FREE_AND_NULL(ref_git);
592 }
593
594 return ref_git;
595}
596
597int foreach_alt_odb(alt_odb_fn fn, void *cb)
598{
599 struct alternate_object_database *ent;
600 int r = 0;
601
602 prepare_alt_odb();
603 for (ent = alt_odb_list; ent; ent = ent->next) {
604 r = fn(ent, cb);
605 if (r)
606 break;
607 }
608 return r;
609}
610
611void prepare_alt_odb(void)
612{
613 const char *alt;
614
615 if (alt_odb_tail)
616 return;
617
618 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
619 if (!alt) alt = "";
620
621 alt_odb_tail = &alt_odb_list;
622 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
623
624 read_info_alternates(get_object_directory(), 0);
625}
626
627/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
628static int freshen_file(const char *fn)
629{
630 struct utimbuf t;
631 t.actime = t.modtime = time(NULL);
632 return !utime(fn, &t);
633}
634
635/*
636 * All of the check_and_freshen functions return 1 if the file exists and was
637 * freshened (if freshening was requested), 0 otherwise. If they return
638 * 0, you should not assume that it is safe to skip a write of the object (it
639 * either does not exist on disk, or has a stale mtime and may be subject to
640 * pruning).
641 */
642int check_and_freshen_file(const char *fn, int freshen)
643{
644 if (access(fn, F_OK))
645 return 0;
646 if (freshen && !freshen_file(fn))
647 return 0;
648 return 1;
649}
650
651static int check_and_freshen_local(const unsigned char *sha1, int freshen)
652{
653 return check_and_freshen_file(sha1_file_name(sha1), freshen);
654}
655
656static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
657{
658 struct alternate_object_database *alt;
659 prepare_alt_odb();
660 for (alt = alt_odb_list; alt; alt = alt->next) {
661 const char *path = alt_sha1_path(alt, sha1);
662 if (check_and_freshen_file(path, freshen))
663 return 1;
664 }
665 return 0;
666}
667
668static int check_and_freshen(const unsigned char *sha1, int freshen)
669{
670 return check_and_freshen_local(sha1, freshen) ||
671 check_and_freshen_nonlocal(sha1, freshen);
672}
673
674int has_loose_object_nonlocal(const unsigned char *sha1)
675{
676 return check_and_freshen_nonlocal(sha1, 0);
677}
678
679static int has_loose_object(const unsigned char *sha1)
680{
681 return check_and_freshen(sha1, 0);
682}
683
684static void mmap_limit_check(size_t length)
685{
686 static size_t limit = 0;
687 if (!limit) {
688 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
689 if (!limit)
690 limit = SIZE_MAX;
691 }
692 if (length > limit)
693 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
694 (uintmax_t)length, (uintmax_t)limit);
695}
696
697void *xmmap_gently(void *start, size_t length,
698 int prot, int flags, int fd, off_t offset)
699{
700 void *ret;
701
702 mmap_limit_check(length);
703 ret = mmap(start, length, prot, flags, fd, offset);
704 if (ret == MAP_FAILED) {
705 if (!length)
706 return NULL;
707 release_pack_memory(length);
708 ret = mmap(start, length, prot, flags, fd, offset);
709 }
710 return ret;
711}
712
713void *xmmap(void *start, size_t length,
714 int prot, int flags, int fd, off_t offset)
715{
716 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
717 if (ret == MAP_FAILED)
718 die_errno("mmap failed");
719 return ret;
720}
721
722void unuse_pack(struct pack_window **w_cursor)
723{
724 struct pack_window *w = *w_cursor;
725 if (w) {
726 w->inuse_cnt--;
727 *w_cursor = NULL;
728 }
729}
730
731static struct packed_git *alloc_packed_git(int extra)
732{
733 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
734 memset(p, 0, sizeof(*p));
735 p->pack_fd = -1;
736 return p;
737}
738
739static void try_to_free_pack_memory(size_t size)
740{
741 release_pack_memory(size);
742}
743
744struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
745{
746 static int have_set_try_to_free_routine;
747 struct stat st;
748 size_t alloc;
749 struct packed_git *p;
750
751 if (!have_set_try_to_free_routine) {
752 have_set_try_to_free_routine = 1;
753 set_try_to_free_routine(try_to_free_pack_memory);
754 }
755
756 /*
757 * Make sure a corresponding .pack file exists and that
758 * the index looks sane.
759 */
760 if (!strip_suffix_mem(path, &path_len, ".idx"))
761 return NULL;
762
763 /*
764 * ".pack" is long enough to hold any suffix we're adding (and
765 * the use xsnprintf double-checks that)
766 */
767 alloc = st_add3(path_len, strlen(".pack"), 1);
768 p = alloc_packed_git(alloc);
769 memcpy(p->pack_name, path, path_len);
770
771 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
772 if (!access(p->pack_name, F_OK))
773 p->pack_keep = 1;
774
775 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
776 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
777 free(p);
778 return NULL;
779 }
780
781 /* ok, it looks sane as far as we can check without
782 * actually mapping the pack file.
783 */
784 p->pack_size = st.st_size;
785 p->pack_local = local;
786 p->mtime = st.st_mtime;
787 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
788 hashclr(p->sha1);
789 return p;
790}
791
792void install_packed_git(struct packed_git *pack)
793{
794 if (pack->pack_fd != -1)
795 pack_open_fds++;
796
797 pack->next = packed_git;
798 packed_git = pack;
799}
800
801void (*report_garbage)(unsigned seen_bits, const char *path);
802
803static void report_helper(const struct string_list *list,
804 int seen_bits, int first, int last)
805{
806 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
807 return;
808
809 for (; first < last; first++)
810 report_garbage(seen_bits, list->items[first].string);
811}
812
813static void report_pack_garbage(struct string_list *list)
814{
815 int i, baselen = -1, first = 0, seen_bits = 0;
816
817 if (!report_garbage)
818 return;
819
820 string_list_sort(list);
821
822 for (i = 0; i < list->nr; i++) {
823 const char *path = list->items[i].string;
824 if (baselen != -1 &&
825 strncmp(path, list->items[first].string, baselen)) {
826 report_helper(list, seen_bits, first, i);
827 baselen = -1;
828 seen_bits = 0;
829 }
830 if (baselen == -1) {
831 const char *dot = strrchr(path, '.');
832 if (!dot) {
833 report_garbage(PACKDIR_FILE_GARBAGE, path);
834 continue;
835 }
836 baselen = dot - path + 1;
837 first = i;
838 }
839 if (!strcmp(path + baselen, "pack"))
840 seen_bits |= 1;
841 else if (!strcmp(path + baselen, "idx"))
842 seen_bits |= 2;
843 }
844 report_helper(list, seen_bits, first, list->nr);
845}
846
847static void prepare_packed_git_one(char *objdir, int local)
848{
849 struct strbuf path = STRBUF_INIT;
850 size_t dirnamelen;
851 DIR *dir;
852 struct dirent *de;
853 struct string_list garbage = STRING_LIST_INIT_DUP;
854
855 strbuf_addstr(&path, objdir);
856 strbuf_addstr(&path, "/pack");
857 dir = opendir(path.buf);
858 if (!dir) {
859 if (errno != ENOENT)
860 error_errno("unable to open object pack directory: %s",
861 path.buf);
862 strbuf_release(&path);
863 return;
864 }
865 strbuf_addch(&path, '/');
866 dirnamelen = path.len;
867 while ((de = readdir(dir)) != NULL) {
868 struct packed_git *p;
869 size_t base_len;
870
871 if (is_dot_or_dotdot(de->d_name))
872 continue;
873
874 strbuf_setlen(&path, dirnamelen);
875 strbuf_addstr(&path, de->d_name);
876
877 base_len = path.len;
878 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
879 /* Don't reopen a pack we already have. */
880 for (p = packed_git; p; p = p->next) {
881 size_t len;
882 if (strip_suffix(p->pack_name, ".pack", &len) &&
883 len == base_len &&
884 !memcmp(p->pack_name, path.buf, len))
885 break;
886 }
887 if (p == NULL &&
888 /*
889 * See if it really is a valid .idx file with
890 * corresponding .pack file that we can map.
891 */
892 (p = add_packed_git(path.buf, path.len, local)) != NULL)
893 install_packed_git(p);
894 }
895
896 if (!report_garbage)
897 continue;
898
899 if (ends_with(de->d_name, ".idx") ||
900 ends_with(de->d_name, ".pack") ||
901 ends_with(de->d_name, ".bitmap") ||
902 ends_with(de->d_name, ".keep"))
903 string_list_append(&garbage, path.buf);
904 else
905 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
906 }
907 closedir(dir);
908 report_pack_garbage(&garbage);
909 string_list_clear(&garbage, 0);
910 strbuf_release(&path);
911}
912
913static int approximate_object_count_valid;
914
915/*
916 * Give a fast, rough count of the number of objects in the repository. This
917 * ignores loose objects completely. If you have a lot of them, then either
918 * you should repack because your performance will be awful, or they are
919 * all unreachable objects about to be pruned, in which case they're not really
920 * interesting as a measure of repo size in the first place.
921 */
922unsigned long approximate_object_count(void)
923{
924 static unsigned long count;
925 if (!approximate_object_count_valid) {
926 struct packed_git *p;
927
928 prepare_packed_git();
929 count = 0;
930 for (p = packed_git; p; p = p->next) {
931 if (open_pack_index(p))
932 continue;
933 count += p->num_objects;
934 }
935 }
936 return count;
937}
938
939static void *get_next_packed_git(const void *p)
940{
941 return ((const struct packed_git *)p)->next;
942}
943
944static void set_next_packed_git(void *p, void *next)
945{
946 ((struct packed_git *)p)->next = next;
947}
948
949static int sort_pack(const void *a_, const void *b_)
950{
951 const struct packed_git *a = a_;
952 const struct packed_git *b = b_;
953 int st;
954
955 /*
956 * Local packs tend to contain objects specific to our
957 * variant of the project than remote ones. In addition,
958 * remote ones could be on a network mounted filesystem.
959 * Favor local ones for these reasons.
960 */
961 st = a->pack_local - b->pack_local;
962 if (st)
963 return -st;
964
965 /*
966 * Younger packs tend to contain more recent objects,
967 * and more recent objects tend to get accessed more
968 * often.
969 */
970 if (a->mtime < b->mtime)
971 return 1;
972 else if (a->mtime == b->mtime)
973 return 0;
974 return -1;
975}
976
977static void rearrange_packed_git(void)
978{
979 packed_git = llist_mergesort(packed_git, get_next_packed_git,
980 set_next_packed_git, sort_pack);
981}
982
983static void prepare_packed_git_mru(void)
984{
985 struct packed_git *p;
986
987 mru_clear(packed_git_mru);
988 for (p = packed_git; p; p = p->next)
989 mru_append(packed_git_mru, p);
990}
991
992static int prepare_packed_git_run_once = 0;
993void prepare_packed_git(void)
994{
995 struct alternate_object_database *alt;
996
997 if (prepare_packed_git_run_once)
998 return;
999 prepare_packed_git_one(get_object_directory(), 1);
1000 prepare_alt_odb();
1001 for (alt = alt_odb_list; alt; alt = alt->next)
1002 prepare_packed_git_one(alt->path, 0);
1003 rearrange_packed_git();
1004 prepare_packed_git_mru();
1005 prepare_packed_git_run_once = 1;
1006}
1007
1008void reprepare_packed_git(void)
1009{
1010 approximate_object_count_valid = 0;
1011 prepare_packed_git_run_once = 0;
1012 prepare_packed_git();
1013}
1014
1015static void mark_bad_packed_object(struct packed_git *p,
1016 const unsigned char *sha1)
1017{
1018 unsigned i;
1019 for (i = 0; i < p->num_bad_objects; i++)
1020 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1021 return;
1022 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1023 st_mult(GIT_MAX_RAWSZ,
1024 st_add(p->num_bad_objects, 1)));
1025 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1026 p->num_bad_objects++;
1027}
1028
1029static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1030{
1031 struct packed_git *p;
1032 unsigned i;
1033
1034 for (p = packed_git; p; p = p->next)
1035 for (i = 0; i < p->num_bad_objects; i++)
1036 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1037 return p;
1038 return NULL;
1039}
1040
1041/*
1042 * With an in-core object data in "map", rehash it to make sure the
1043 * object name actually matches "sha1" to detect object corruption.
1044 * With "map" == NULL, try reading the object named with "sha1" using
1045 * the streaming interface and rehash it to do the same.
1046 */
1047int check_sha1_signature(const unsigned char *sha1, void *map,
1048 unsigned long size, const char *type)
1049{
1050 unsigned char real_sha1[20];
1051 enum object_type obj_type;
1052 struct git_istream *st;
1053 git_SHA_CTX c;
1054 char hdr[32];
1055 int hdrlen;
1056
1057 if (map) {
1058 hash_sha1_file(map, size, type, real_sha1);
1059 return hashcmp(sha1, real_sha1) ? -1 : 0;
1060 }
1061
1062 st = open_istream(sha1, &obj_type, &size, NULL);
1063 if (!st)
1064 return -1;
1065
1066 /* Generate the header */
1067 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1068
1069 /* Sha1.. */
1070 git_SHA1_Init(&c);
1071 git_SHA1_Update(&c, hdr, hdrlen);
1072 for (;;) {
1073 char buf[1024 * 16];
1074 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1075
1076 if (readlen < 0) {
1077 close_istream(st);
1078 return -1;
1079 }
1080 if (!readlen)
1081 break;
1082 git_SHA1_Update(&c, buf, readlen);
1083 }
1084 git_SHA1_Final(real_sha1, &c);
1085 close_istream(st);
1086 return hashcmp(sha1, real_sha1) ? -1 : 0;
1087}
1088
1089int git_open_cloexec(const char *name, int flags)
1090{
1091 int fd;
1092 static int o_cloexec = O_CLOEXEC;
1093
1094 fd = open(name, flags | o_cloexec);
1095 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1096 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1097 o_cloexec &= ~O_CLOEXEC;
1098 fd = open(name, flags | o_cloexec);
1099 }
1100
1101#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1102 {
1103 static int fd_cloexec = FD_CLOEXEC;
1104
1105 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1106 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1107 int flags = fcntl(fd, F_GETFD);
1108 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1109 fd_cloexec = 0;
1110 }
1111 }
1112#endif
1113 return fd;
1114}
1115
1116/*
1117 * Find "sha1" as a loose object in the local repository or in an alternate.
1118 * Returns 0 on success, negative on failure.
1119 *
1120 * The "path" out-parameter will give the path of the object we found (if any).
1121 * Note that it may point to static storage and is only valid until another
1122 * call to sha1_file_name(), etc.
1123 */
1124static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1125 const char **path)
1126{
1127 struct alternate_object_database *alt;
1128
1129 *path = sha1_file_name(sha1);
1130 if (!lstat(*path, st))
1131 return 0;
1132
1133 prepare_alt_odb();
1134 errno = ENOENT;
1135 for (alt = alt_odb_list; alt; alt = alt->next) {
1136 *path = alt_sha1_path(alt, sha1);
1137 if (!lstat(*path, st))
1138 return 0;
1139 }
1140
1141 return -1;
1142}
1143
1144/*
1145 * Like stat_sha1_file(), but actually open the object and return the
1146 * descriptor. See the caveats on the "path" parameter above.
1147 */
1148static int open_sha1_file(const unsigned char *sha1, const char **path)
1149{
1150 int fd;
1151 struct alternate_object_database *alt;
1152 int most_interesting_errno;
1153
1154 *path = sha1_file_name(sha1);
1155 fd = git_open(*path);
1156 if (fd >= 0)
1157 return fd;
1158 most_interesting_errno = errno;
1159
1160 prepare_alt_odb();
1161 for (alt = alt_odb_list; alt; alt = alt->next) {
1162 *path = alt_sha1_path(alt, sha1);
1163 fd = git_open(*path);
1164 if (fd >= 0)
1165 return fd;
1166 if (most_interesting_errno == ENOENT)
1167 most_interesting_errno = errno;
1168 }
1169 errno = most_interesting_errno;
1170 return -1;
1171}
1172
1173/*
1174 * Map the loose object at "path" if it is not NULL, or the path found by
1175 * searching for a loose object named "sha1".
1176 */
1177static void *map_sha1_file_1(const char *path,
1178 const unsigned char *sha1,
1179 unsigned long *size)
1180{
1181 void *map;
1182 int fd;
1183
1184 if (path)
1185 fd = git_open(path);
1186 else
1187 fd = open_sha1_file(sha1, &path);
1188 map = NULL;
1189 if (fd >= 0) {
1190 struct stat st;
1191
1192 if (!fstat(fd, &st)) {
1193 *size = xsize_t(st.st_size);
1194 if (!*size) {
1195 /* mmap() is forbidden on empty files */
1196 error("object file %s is empty", path);
1197 return NULL;
1198 }
1199 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1200 }
1201 close(fd);
1202 }
1203 return map;
1204}
1205
1206void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1207{
1208 return map_sha1_file_1(NULL, sha1, size);
1209}
1210
1211unsigned long unpack_object_header_buffer(const unsigned char *buf,
1212 unsigned long len, enum object_type *type, unsigned long *sizep)
1213{
1214 unsigned shift;
1215 unsigned long size, c;
1216 unsigned long used = 0;
1217
1218 c = buf[used++];
1219 *type = (c >> 4) & 7;
1220 size = c & 15;
1221 shift = 4;
1222 while (c & 0x80) {
1223 if (len <= used || bitsizeof(long) <= shift) {
1224 error("bad object header");
1225 size = used = 0;
1226 break;
1227 }
1228 c = buf[used++];
1229 size += (c & 0x7f) << shift;
1230 shift += 7;
1231 }
1232 *sizep = size;
1233 return used;
1234}
1235
1236static int unpack_sha1_short_header(git_zstream *stream,
1237 unsigned char *map, unsigned long mapsize,
1238 void *buffer, unsigned long bufsiz)
1239{
1240 /* Get the data stream */
1241 memset(stream, 0, sizeof(*stream));
1242 stream->next_in = map;
1243 stream->avail_in = mapsize;
1244 stream->next_out = buffer;
1245 stream->avail_out = bufsiz;
1246
1247 git_inflate_init(stream);
1248 return git_inflate(stream, 0);
1249}
1250
1251int unpack_sha1_header(git_zstream *stream,
1252 unsigned char *map, unsigned long mapsize,
1253 void *buffer, unsigned long bufsiz)
1254{
1255 int status = unpack_sha1_short_header(stream, map, mapsize,
1256 buffer, bufsiz);
1257
1258 if (status < Z_OK)
1259 return status;
1260
1261 /* Make sure we have the terminating NUL */
1262 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1263 return -1;
1264 return 0;
1265}
1266
1267static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1268 unsigned long mapsize, void *buffer,
1269 unsigned long bufsiz, struct strbuf *header)
1270{
1271 int status;
1272
1273 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1274 if (status < Z_OK)
1275 return -1;
1276
1277 /*
1278 * Check if entire header is unpacked in the first iteration.
1279 */
1280 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1281 return 0;
1282
1283 /*
1284 * buffer[0..bufsiz] was not large enough. Copy the partial
1285 * result out to header, and then append the result of further
1286 * reading the stream.
1287 */
1288 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1289 stream->next_out = buffer;
1290 stream->avail_out = bufsiz;
1291
1292 do {
1293 status = git_inflate(stream, 0);
1294 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1295 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1296 return 0;
1297 stream->next_out = buffer;
1298 stream->avail_out = bufsiz;
1299 } while (status != Z_STREAM_END);
1300 return -1;
1301}
1302
1303static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1304{
1305 int bytes = strlen(buffer) + 1;
1306 unsigned char *buf = xmallocz(size);
1307 unsigned long n;
1308 int status = Z_OK;
1309
1310 n = stream->total_out - bytes;
1311 if (n > size)
1312 n = size;
1313 memcpy(buf, (char *) buffer + bytes, n);
1314 bytes = n;
1315 if (bytes <= size) {
1316 /*
1317 * The above condition must be (bytes <= size), not
1318 * (bytes < size). In other words, even though we
1319 * expect no more output and set avail_out to zero,
1320 * the input zlib stream may have bytes that express
1321 * "this concludes the stream", and we *do* want to
1322 * eat that input.
1323 *
1324 * Otherwise we would not be able to test that we
1325 * consumed all the input to reach the expected size;
1326 * we also want to check that zlib tells us that all
1327 * went well with status == Z_STREAM_END at the end.
1328 */
1329 stream->next_out = buf + bytes;
1330 stream->avail_out = size - bytes;
1331 while (status == Z_OK)
1332 status = git_inflate(stream, Z_FINISH);
1333 }
1334 if (status == Z_STREAM_END && !stream->avail_in) {
1335 git_inflate_end(stream);
1336 return buf;
1337 }
1338
1339 if (status < 0)
1340 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1341 else if (stream->avail_in)
1342 error("garbage at end of loose object '%s'",
1343 sha1_to_hex(sha1));
1344 free(buf);
1345 return NULL;
1346}
1347
1348/*
1349 * We used to just use "sscanf()", but that's actually way
1350 * too permissive for what we want to check. So do an anal
1351 * object header parse by hand.
1352 */
1353static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1354 unsigned int flags)
1355{
1356 const char *type_buf = hdr;
1357 unsigned long size;
1358 int type, type_len = 0;
1359
1360 /*
1361 * The type can be of any size but is followed by
1362 * a space.
1363 */
1364 for (;;) {
1365 char c = *hdr++;
1366 if (!c)
1367 return -1;
1368 if (c == ' ')
1369 break;
1370 type_len++;
1371 }
1372
1373 type = type_from_string_gently(type_buf, type_len, 1);
1374 if (oi->typename)
1375 strbuf_add(oi->typename, type_buf, type_len);
1376 /*
1377 * Set type to 0 if its an unknown object and
1378 * we're obtaining the type using '--allow-unknown-type'
1379 * option.
1380 */
1381 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1382 type = 0;
1383 else if (type < 0)
1384 die("invalid object type");
1385 if (oi->typep)
1386 *oi->typep = type;
1387
1388 /*
1389 * The length must follow immediately, and be in canonical
1390 * decimal format (ie "010" is not valid).
1391 */
1392 size = *hdr++ - '0';
1393 if (size > 9)
1394 return -1;
1395 if (size) {
1396 for (;;) {
1397 unsigned long c = *hdr - '0';
1398 if (c > 9)
1399 break;
1400 hdr++;
1401 size = size * 10 + c;
1402 }
1403 }
1404
1405 if (oi->sizep)
1406 *oi->sizep = size;
1407
1408 /*
1409 * The length must be followed by a zero byte
1410 */
1411 return *hdr ? -1 : type;
1412}
1413
1414int parse_sha1_header(const char *hdr, unsigned long *sizep)
1415{
1416 struct object_info oi = OBJECT_INFO_INIT;
1417
1418 oi.sizep = sizep;
1419 return parse_sha1_header_extended(hdr, &oi, 0);
1420}
1421
1422unsigned long get_size_from_delta(struct packed_git *p,
1423 struct pack_window **w_curs,
1424 off_t curpos)
1425{
1426 const unsigned char *data;
1427 unsigned char delta_head[20], *in;
1428 git_zstream stream;
1429 int st;
1430
1431 memset(&stream, 0, sizeof(stream));
1432 stream.next_out = delta_head;
1433 stream.avail_out = sizeof(delta_head);
1434
1435 git_inflate_init(&stream);
1436 do {
1437 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1438 stream.next_in = in;
1439 st = git_inflate(&stream, Z_FINISH);
1440 curpos += stream.next_in - in;
1441 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1442 stream.total_out < sizeof(delta_head));
1443 git_inflate_end(&stream);
1444 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1445 error("delta data unpack-initial failed");
1446 return 0;
1447 }
1448
1449 /* Examine the initial part of the delta to figure out
1450 * the result size.
1451 */
1452 data = delta_head;
1453
1454 /* ignore base size */
1455 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1456
1457 /* Read the result size */
1458 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1459}
1460
1461static off_t get_delta_base(struct packed_git *p,
1462 struct pack_window **w_curs,
1463 off_t *curpos,
1464 enum object_type type,
1465 off_t delta_obj_offset)
1466{
1467 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1468 off_t base_offset;
1469
1470 /* use_pack() assured us we have [base_info, base_info + 20)
1471 * as a range that we can look at without walking off the
1472 * end of the mapped window. Its actually the hash size
1473 * that is assured. An OFS_DELTA longer than the hash size
1474 * is stupid, as then a REF_DELTA would be smaller to store.
1475 */
1476 if (type == OBJ_OFS_DELTA) {
1477 unsigned used = 0;
1478 unsigned char c = base_info[used++];
1479 base_offset = c & 127;
1480 while (c & 128) {
1481 base_offset += 1;
1482 if (!base_offset || MSB(base_offset, 7))
1483 return 0; /* overflow */
1484 c = base_info[used++];
1485 base_offset = (base_offset << 7) + (c & 127);
1486 }
1487 base_offset = delta_obj_offset - base_offset;
1488 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1489 return 0; /* out of bound */
1490 *curpos += used;
1491 } else if (type == OBJ_REF_DELTA) {
1492 /* The base entry _must_ be in the same pack */
1493 base_offset = find_pack_entry_one(base_info, p);
1494 *curpos += 20;
1495 } else
1496 die("I am totally screwed");
1497 return base_offset;
1498}
1499
1500/*
1501 * Like get_delta_base above, but we return the sha1 instead of the pack
1502 * offset. This means it is cheaper for REF deltas (we do not have to do
1503 * the final object lookup), but more expensive for OFS deltas (we
1504 * have to load the revidx to convert the offset back into a sha1).
1505 */
1506static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1507 struct pack_window **w_curs,
1508 off_t curpos,
1509 enum object_type type,
1510 off_t delta_obj_offset)
1511{
1512 if (type == OBJ_REF_DELTA) {
1513 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1514 return base;
1515 } else if (type == OBJ_OFS_DELTA) {
1516 struct revindex_entry *revidx;
1517 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1518 type, delta_obj_offset);
1519
1520 if (!base_offset)
1521 return NULL;
1522
1523 revidx = find_pack_revindex(p, base_offset);
1524 if (!revidx)
1525 return NULL;
1526
1527 return nth_packed_object_sha1(p, revidx->nr);
1528 } else
1529 return NULL;
1530}
1531
1532int unpack_object_header(struct packed_git *p,
1533 struct pack_window **w_curs,
1534 off_t *curpos,
1535 unsigned long *sizep)
1536{
1537 unsigned char *base;
1538 unsigned long left;
1539 unsigned long used;
1540 enum object_type type;
1541
1542 /* use_pack() assures us we have [base, base + 20) available
1543 * as a range that we can look at. (Its actually the hash
1544 * size that is assured.) With our object header encoding
1545 * the maximum deflated object size is 2^137, which is just
1546 * insane, so we know won't exceed what we have been given.
1547 */
1548 base = use_pack(p, w_curs, *curpos, &left);
1549 used = unpack_object_header_buffer(base, left, &type, sizep);
1550 if (!used) {
1551 type = OBJ_BAD;
1552 } else
1553 *curpos += used;
1554
1555 return type;
1556}
1557
1558static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1559{
1560 int type;
1561 struct revindex_entry *revidx;
1562 const unsigned char *sha1;
1563 revidx = find_pack_revindex(p, obj_offset);
1564 if (!revidx)
1565 return OBJ_BAD;
1566 sha1 = nth_packed_object_sha1(p, revidx->nr);
1567 mark_bad_packed_object(p, sha1);
1568 type = sha1_object_info(sha1, NULL);
1569 if (type <= OBJ_NONE)
1570 return OBJ_BAD;
1571 return type;
1572}
1573
1574#define POI_STACK_PREALLOC 64
1575
1576static enum object_type packed_to_object_type(struct packed_git *p,
1577 off_t obj_offset,
1578 enum object_type type,
1579 struct pack_window **w_curs,
1580 off_t curpos)
1581{
1582 off_t small_poi_stack[POI_STACK_PREALLOC];
1583 off_t *poi_stack = small_poi_stack;
1584 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1585
1586 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1587 off_t base_offset;
1588 unsigned long size;
1589 /* Push the object we're going to leave behind */
1590 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1591 poi_stack_alloc = alloc_nr(poi_stack_nr);
1592 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1593 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1594 } else {
1595 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1596 }
1597 poi_stack[poi_stack_nr++] = obj_offset;
1598 /* If parsing the base offset fails, just unwind */
1599 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1600 if (!base_offset)
1601 goto unwind;
1602 curpos = obj_offset = base_offset;
1603 type = unpack_object_header(p, w_curs, &curpos, &size);
1604 if (type <= OBJ_NONE) {
1605 /* If getting the base itself fails, we first
1606 * retry the base, otherwise unwind */
1607 type = retry_bad_packed_offset(p, base_offset);
1608 if (type > OBJ_NONE)
1609 goto out;
1610 goto unwind;
1611 }
1612 }
1613
1614 switch (type) {
1615 case OBJ_BAD:
1616 case OBJ_COMMIT:
1617 case OBJ_TREE:
1618 case OBJ_BLOB:
1619 case OBJ_TAG:
1620 break;
1621 default:
1622 error("unknown object type %i at offset %"PRIuMAX" in %s",
1623 type, (uintmax_t)obj_offset, p->pack_name);
1624 type = OBJ_BAD;
1625 }
1626
1627out:
1628 if (poi_stack != small_poi_stack)
1629 free(poi_stack);
1630 return type;
1631
1632unwind:
1633 while (poi_stack_nr) {
1634 obj_offset = poi_stack[--poi_stack_nr];
1635 type = retry_bad_packed_offset(p, obj_offset);
1636 if (type > OBJ_NONE)
1637 goto out;
1638 }
1639 type = OBJ_BAD;
1640 goto out;
1641}
1642
1643static struct hashmap delta_base_cache;
1644static size_t delta_base_cached;
1645
1646static LIST_HEAD(delta_base_cache_lru);
1647
1648struct delta_base_cache_key {
1649 struct packed_git *p;
1650 off_t base_offset;
1651};
1652
1653struct delta_base_cache_entry {
1654 struct hashmap hash;
1655 struct delta_base_cache_key key;
1656 struct list_head lru;
1657 void *data;
1658 unsigned long size;
1659 enum object_type type;
1660};
1661
1662static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1663{
1664 unsigned int hash;
1665
1666 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1667 hash += (hash >> 8) + (hash >> 16);
1668 return hash;
1669}
1670
1671static struct delta_base_cache_entry *
1672get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1673{
1674 struct hashmap_entry entry;
1675 struct delta_base_cache_key key;
1676
1677 if (!delta_base_cache.cmpfn)
1678 return NULL;
1679
1680 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1681 key.p = p;
1682 key.base_offset = base_offset;
1683 return hashmap_get(&delta_base_cache, &entry, &key);
1684}
1685
1686static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1687 const struct delta_base_cache_key *b)
1688{
1689 return a->p == b->p && a->base_offset == b->base_offset;
1690}
1691
1692static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1693 const void *va, const void *vb,
1694 const void *vkey)
1695{
1696 const struct delta_base_cache_entry *a = va, *b = vb;
1697 const struct delta_base_cache_key *key = vkey;
1698 if (key)
1699 return !delta_base_cache_key_eq(&a->key, key);
1700 else
1701 return !delta_base_cache_key_eq(&a->key, &b->key);
1702}
1703
1704static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1705{
1706 return !!get_delta_base_cache_entry(p, base_offset);
1707}
1708
1709/*
1710 * Remove the entry from the cache, but do _not_ free the associated
1711 * entry data. The caller takes ownership of the "data" buffer, and
1712 * should copy out any fields it wants before detaching.
1713 */
1714static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1715{
1716 hashmap_remove(&delta_base_cache, ent, &ent->key);
1717 list_del(&ent->lru);
1718 delta_base_cached -= ent->size;
1719 free(ent);
1720}
1721
1722static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1723 unsigned long *base_size, enum object_type *type)
1724{
1725 struct delta_base_cache_entry *ent;
1726
1727 ent = get_delta_base_cache_entry(p, base_offset);
1728 if (!ent)
1729 return unpack_entry(p, base_offset, type, base_size);
1730
1731 if (type)
1732 *type = ent->type;
1733 if (base_size)
1734 *base_size = ent->size;
1735 return xmemdupz(ent->data, ent->size);
1736}
1737
1738static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1739{
1740 free(ent->data);
1741 detach_delta_base_cache_entry(ent);
1742}
1743
1744void clear_delta_base_cache(void)
1745{
1746 struct list_head *lru, *tmp;
1747 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1748 struct delta_base_cache_entry *entry =
1749 list_entry(lru, struct delta_base_cache_entry, lru);
1750 release_delta_base_cache(entry);
1751 }
1752}
1753
1754static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1755 void *base, unsigned long base_size, enum object_type type)
1756{
1757 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1758 struct list_head *lru, *tmp;
1759
1760 delta_base_cached += base_size;
1761
1762 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1763 struct delta_base_cache_entry *f =
1764 list_entry(lru, struct delta_base_cache_entry, lru);
1765 if (delta_base_cached <= delta_base_cache_limit)
1766 break;
1767 release_delta_base_cache(f);
1768 }
1769
1770 ent->key.p = p;
1771 ent->key.base_offset = base_offset;
1772 ent->type = type;
1773 ent->data = base;
1774 ent->size = base_size;
1775 list_add_tail(&ent->lru, &delta_base_cache_lru);
1776
1777 if (!delta_base_cache.cmpfn)
1778 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1779 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1780 hashmap_add(&delta_base_cache, ent);
1781}
1782
1783int packed_object_info(struct packed_git *p, off_t obj_offset,
1784 struct object_info *oi)
1785{
1786 struct pack_window *w_curs = NULL;
1787 unsigned long size;
1788 off_t curpos = obj_offset;
1789 enum object_type type;
1790
1791 /*
1792 * We always get the representation type, but only convert it to
1793 * a "real" type later if the caller is interested.
1794 */
1795 if (oi->contentp) {
1796 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1797 &type);
1798 if (!*oi->contentp)
1799 type = OBJ_BAD;
1800 } else {
1801 type = unpack_object_header(p, &w_curs, &curpos, &size);
1802 }
1803
1804 if (!oi->contentp && oi->sizep) {
1805 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1806 off_t tmp_pos = curpos;
1807 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1808 type, obj_offset);
1809 if (!base_offset) {
1810 type = OBJ_BAD;
1811 goto out;
1812 }
1813 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1814 if (*oi->sizep == 0) {
1815 type = OBJ_BAD;
1816 goto out;
1817 }
1818 } else {
1819 *oi->sizep = size;
1820 }
1821 }
1822
1823 if (oi->disk_sizep) {
1824 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1825 *oi->disk_sizep = revidx[1].offset - obj_offset;
1826 }
1827
1828 if (oi->typep || oi->typename) {
1829 enum object_type ptot;
1830 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1831 curpos);
1832 if (oi->typep)
1833 *oi->typep = ptot;
1834 if (oi->typename) {
1835 const char *tn = typename(ptot);
1836 if (tn)
1837 strbuf_addstr(oi->typename, tn);
1838 }
1839 if (ptot < 0) {
1840 type = OBJ_BAD;
1841 goto out;
1842 }
1843 }
1844
1845 if (oi->delta_base_sha1) {
1846 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1847 const unsigned char *base;
1848
1849 base = get_delta_base_sha1(p, &w_curs, curpos,
1850 type, obj_offset);
1851 if (!base) {
1852 type = OBJ_BAD;
1853 goto out;
1854 }
1855
1856 hashcpy(oi->delta_base_sha1, base);
1857 } else
1858 hashclr(oi->delta_base_sha1);
1859 }
1860
1861 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1862 OI_PACKED;
1863
1864out:
1865 unuse_pack(&w_curs);
1866 return type;
1867}
1868
1869static void *unpack_compressed_entry(struct packed_git *p,
1870 struct pack_window **w_curs,
1871 off_t curpos,
1872 unsigned long size)
1873{
1874 int st;
1875 git_zstream stream;
1876 unsigned char *buffer, *in;
1877
1878 buffer = xmallocz_gently(size);
1879 if (!buffer)
1880 return NULL;
1881 memset(&stream, 0, sizeof(stream));
1882 stream.next_out = buffer;
1883 stream.avail_out = size + 1;
1884
1885 git_inflate_init(&stream);
1886 do {
1887 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1888 stream.next_in = in;
1889 st = git_inflate(&stream, Z_FINISH);
1890 if (!stream.avail_out)
1891 break; /* the payload is larger than it should be */
1892 curpos += stream.next_in - in;
1893 } while (st == Z_OK || st == Z_BUF_ERROR);
1894 git_inflate_end(&stream);
1895 if ((st != Z_STREAM_END) || stream.total_out != size) {
1896 free(buffer);
1897 return NULL;
1898 }
1899
1900 return buffer;
1901}
1902
1903static void *read_object(const unsigned char *sha1, enum object_type *type,
1904 unsigned long *size);
1905
1906static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1907{
1908 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1909 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1910 p->pack_name, (uintmax_t)obj_offset);
1911}
1912
1913int do_check_packed_object_crc;
1914
1915#define UNPACK_ENTRY_STACK_PREALLOC 64
1916struct unpack_entry_stack_ent {
1917 off_t obj_offset;
1918 off_t curpos;
1919 unsigned long size;
1920};
1921
1922void *unpack_entry(struct packed_git *p, off_t obj_offset,
1923 enum object_type *final_type, unsigned long *final_size)
1924{
1925 struct pack_window *w_curs = NULL;
1926 off_t curpos = obj_offset;
1927 void *data = NULL;
1928 unsigned long size;
1929 enum object_type type;
1930 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1931 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1932 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1933 int base_from_cache = 0;
1934
1935 write_pack_access_log(p, obj_offset);
1936
1937 /* PHASE 1: drill down to the innermost base object */
1938 for (;;) {
1939 off_t base_offset;
1940 int i;
1941 struct delta_base_cache_entry *ent;
1942
1943 ent = get_delta_base_cache_entry(p, curpos);
1944 if (ent) {
1945 type = ent->type;
1946 data = ent->data;
1947 size = ent->size;
1948 detach_delta_base_cache_entry(ent);
1949 base_from_cache = 1;
1950 break;
1951 }
1952
1953 if (do_check_packed_object_crc && p->index_version > 1) {
1954 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1955 off_t len = revidx[1].offset - obj_offset;
1956 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1957 const unsigned char *sha1 =
1958 nth_packed_object_sha1(p, revidx->nr);
1959 error("bad packed object CRC for %s",
1960 sha1_to_hex(sha1));
1961 mark_bad_packed_object(p, sha1);
1962 data = NULL;
1963 goto out;
1964 }
1965 }
1966
1967 type = unpack_object_header(p, &w_curs, &curpos, &size);
1968 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1969 break;
1970
1971 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1972 if (!base_offset) {
1973 error("failed to validate delta base reference "
1974 "at offset %"PRIuMAX" from %s",
1975 (uintmax_t)curpos, p->pack_name);
1976 /* bail to phase 2, in hopes of recovery */
1977 data = NULL;
1978 break;
1979 }
1980
1981 /* push object, proceed to base */
1982 if (delta_stack_nr >= delta_stack_alloc
1983 && delta_stack == small_delta_stack) {
1984 delta_stack_alloc = alloc_nr(delta_stack_nr);
1985 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1986 memcpy(delta_stack, small_delta_stack,
1987 sizeof(*delta_stack)*delta_stack_nr);
1988 } else {
1989 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1990 }
1991 i = delta_stack_nr++;
1992 delta_stack[i].obj_offset = obj_offset;
1993 delta_stack[i].curpos = curpos;
1994 delta_stack[i].size = size;
1995
1996 curpos = obj_offset = base_offset;
1997 }
1998
1999 /* PHASE 2: handle the base */
2000 switch (type) {
2001 case OBJ_OFS_DELTA:
2002 case OBJ_REF_DELTA:
2003 if (data)
2004 die("BUG: unpack_entry: left loop at a valid delta");
2005 break;
2006 case OBJ_COMMIT:
2007 case OBJ_TREE:
2008 case OBJ_BLOB:
2009 case OBJ_TAG:
2010 if (!base_from_cache)
2011 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2012 break;
2013 default:
2014 data = NULL;
2015 error("unknown object type %i at offset %"PRIuMAX" in %s",
2016 type, (uintmax_t)obj_offset, p->pack_name);
2017 }
2018
2019 /* PHASE 3: apply deltas in order */
2020
2021 /* invariants:
2022 * 'data' holds the base data, or NULL if there was corruption
2023 */
2024 while (delta_stack_nr) {
2025 void *delta_data;
2026 void *base = data;
2027 void *external_base = NULL;
2028 unsigned long delta_size, base_size = size;
2029 int i;
2030
2031 data = NULL;
2032
2033 if (base)
2034 add_delta_base_cache(p, obj_offset, base, base_size, type);
2035
2036 if (!base) {
2037 /*
2038 * We're probably in deep shit, but let's try to fetch
2039 * the required base anyway from another pack or loose.
2040 * This is costly but should happen only in the presence
2041 * of a corrupted pack, and is better than failing outright.
2042 */
2043 struct revindex_entry *revidx;
2044 const unsigned char *base_sha1;
2045 revidx = find_pack_revindex(p, obj_offset);
2046 if (revidx) {
2047 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2048 error("failed to read delta base object %s"
2049 " at offset %"PRIuMAX" from %s",
2050 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2051 p->pack_name);
2052 mark_bad_packed_object(p, base_sha1);
2053 base = read_object(base_sha1, &type, &base_size);
2054 external_base = base;
2055 }
2056 }
2057
2058 i = --delta_stack_nr;
2059 obj_offset = delta_stack[i].obj_offset;
2060 curpos = delta_stack[i].curpos;
2061 delta_size = delta_stack[i].size;
2062
2063 if (!base)
2064 continue;
2065
2066 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2067
2068 if (!delta_data) {
2069 error("failed to unpack compressed delta "
2070 "at offset %"PRIuMAX" from %s",
2071 (uintmax_t)curpos, p->pack_name);
2072 data = NULL;
2073 free(external_base);
2074 continue;
2075 }
2076
2077 data = patch_delta(base, base_size,
2078 delta_data, delta_size,
2079 &size);
2080
2081 /*
2082 * We could not apply the delta; warn the user, but keep going.
2083 * Our failure will be noticed either in the next iteration of
2084 * the loop, or if this is the final delta, in the caller when
2085 * we return NULL. Those code paths will take care of making
2086 * a more explicit warning and retrying with another copy of
2087 * the object.
2088 */
2089 if (!data)
2090 error("failed to apply delta");
2091
2092 free(delta_data);
2093 free(external_base);
2094 }
2095
2096 if (final_type)
2097 *final_type = type;
2098 if (final_size)
2099 *final_size = size;
2100
2101out:
2102 unuse_pack(&w_curs);
2103
2104 if (delta_stack != small_delta_stack)
2105 free(delta_stack);
2106
2107 return data;
2108}
2109
2110const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2111 uint32_t n)
2112{
2113 const unsigned char *index = p->index_data;
2114 if (!index) {
2115 if (open_pack_index(p))
2116 return NULL;
2117 index = p->index_data;
2118 }
2119 if (n >= p->num_objects)
2120 return NULL;
2121 index += 4 * 256;
2122 if (p->index_version == 1) {
2123 return index + 24 * n + 4;
2124 } else {
2125 index += 8;
2126 return index + 20 * n;
2127 }
2128}
2129
2130const struct object_id *nth_packed_object_oid(struct object_id *oid,
2131 struct packed_git *p,
2132 uint32_t n)
2133{
2134 const unsigned char *hash = nth_packed_object_sha1(p, n);
2135 if (!hash)
2136 return NULL;
2137 hashcpy(oid->hash, hash);
2138 return oid;
2139}
2140
2141void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2142{
2143 const unsigned char *ptr = vptr;
2144 const unsigned char *start = p->index_data;
2145 const unsigned char *end = start + p->index_size;
2146 if (ptr < start)
2147 die(_("offset before start of pack index for %s (corrupt index?)"),
2148 p->pack_name);
2149 /* No need to check for underflow; .idx files must be at least 8 bytes */
2150 if (ptr >= end - 8)
2151 die(_("offset beyond end of pack index for %s (truncated index?)"),
2152 p->pack_name);
2153}
2154
2155off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2156{
2157 const unsigned char *index = p->index_data;
2158 index += 4 * 256;
2159 if (p->index_version == 1) {
2160 return ntohl(*((uint32_t *)(index + 24 * n)));
2161 } else {
2162 uint32_t off;
2163 index += 8 + p->num_objects * (20 + 4);
2164 off = ntohl(*((uint32_t *)(index + 4 * n)));
2165 if (!(off & 0x80000000))
2166 return off;
2167 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2168 check_pack_index_ptr(p, index);
2169 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2170 ntohl(*((uint32_t *)(index + 4)));
2171 }
2172}
2173
2174off_t find_pack_entry_one(const unsigned char *sha1,
2175 struct packed_git *p)
2176{
2177 const uint32_t *level1_ofs = p->index_data;
2178 const unsigned char *index = p->index_data;
2179 unsigned hi, lo, stride;
2180 static int debug_lookup = -1;
2181
2182 if (debug_lookup < 0)
2183 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2184
2185 if (!index) {
2186 if (open_pack_index(p))
2187 return 0;
2188 level1_ofs = p->index_data;
2189 index = p->index_data;
2190 }
2191 if (p->index_version > 1) {
2192 level1_ofs += 2;
2193 index += 8;
2194 }
2195 index += 4 * 256;
2196 hi = ntohl(level1_ofs[*sha1]);
2197 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2198 if (p->index_version > 1) {
2199 stride = 20;
2200 } else {
2201 stride = 24;
2202 index += 4;
2203 }
2204
2205 if (debug_lookup)
2206 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2207 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2208
2209 while (lo < hi) {
2210 unsigned mi = (lo + hi) / 2;
2211 int cmp = hashcmp(index + mi * stride, sha1);
2212
2213 if (debug_lookup)
2214 printf("lo %u hi %u rg %u mi %u\n",
2215 lo, hi, hi - lo, mi);
2216 if (!cmp)
2217 return nth_packed_object_offset(p, mi);
2218 if (cmp > 0)
2219 hi = mi;
2220 else
2221 lo = mi+1;
2222 }
2223 return 0;
2224}
2225
2226int is_pack_valid(struct packed_git *p)
2227{
2228 /* An already open pack is known to be valid. */
2229 if (p->pack_fd != -1)
2230 return 1;
2231
2232 /* If the pack has one window completely covering the
2233 * file size, the pack is known to be valid even if
2234 * the descriptor is not currently open.
2235 */
2236 if (p->windows) {
2237 struct pack_window *w = p->windows;
2238
2239 if (!w->offset && w->len == p->pack_size)
2240 return 1;
2241 }
2242
2243 /* Force the pack to open to prove its valid. */
2244 return !open_packed_git(p);
2245}
2246
2247static int fill_pack_entry(const unsigned char *sha1,
2248 struct pack_entry *e,
2249 struct packed_git *p)
2250{
2251 off_t offset;
2252
2253 if (p->num_bad_objects) {
2254 unsigned i;
2255 for (i = 0; i < p->num_bad_objects; i++)
2256 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2257 return 0;
2258 }
2259
2260 offset = find_pack_entry_one(sha1, p);
2261 if (!offset)
2262 return 0;
2263
2264 /*
2265 * We are about to tell the caller where they can locate the
2266 * requested object. We better make sure the packfile is
2267 * still here and can be accessed before supplying that
2268 * answer, as it may have been deleted since the index was
2269 * loaded!
2270 */
2271 if (!is_pack_valid(p))
2272 return 0;
2273 e->offset = offset;
2274 e->p = p;
2275 hashcpy(e->sha1, sha1);
2276 return 1;
2277}
2278
2279/*
2280 * Iff a pack file contains the object named by sha1, return true and
2281 * store its location to e.
2282 */
2283static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2284{
2285 struct mru_entry *p;
2286
2287 prepare_packed_git();
2288 if (!packed_git)
2289 return 0;
2290
2291 for (p = packed_git_mru->head; p; p = p->next) {
2292 if (fill_pack_entry(sha1, e, p->item)) {
2293 mru_mark(packed_git_mru, p);
2294 return 1;
2295 }
2296 }
2297 return 0;
2298}
2299
2300struct packed_git *find_sha1_pack(const unsigned char *sha1,
2301 struct packed_git *packs)
2302{
2303 struct packed_git *p;
2304
2305 for (p = packs; p; p = p->next) {
2306 if (find_pack_entry_one(sha1, p))
2307 return p;
2308 }
2309 return NULL;
2310
2311}
2312
2313static int sha1_loose_object_info(const unsigned char *sha1,
2314 struct object_info *oi,
2315 int flags)
2316{
2317 int status = 0;
2318 unsigned long mapsize;
2319 void *map;
2320 git_zstream stream;
2321 char hdr[32];
2322 struct strbuf hdrbuf = STRBUF_INIT;
2323 unsigned long size_scratch;
2324
2325 if (oi->delta_base_sha1)
2326 hashclr(oi->delta_base_sha1);
2327
2328 /*
2329 * If we don't care about type or size, then we don't
2330 * need to look inside the object at all. Note that we
2331 * do not optimize out the stat call, even if the
2332 * caller doesn't care about the disk-size, since our
2333 * return value implicitly indicates whether the
2334 * object even exists.
2335 */
2336 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2337 const char *path;
2338 struct stat st;
2339 if (stat_sha1_file(sha1, &st, &path) < 0)
2340 return -1;
2341 if (oi->disk_sizep)
2342 *oi->disk_sizep = st.st_size;
2343 return 0;
2344 }
2345
2346 map = map_sha1_file(sha1, &mapsize);
2347 if (!map)
2348 return -1;
2349
2350 if (!oi->sizep)
2351 oi->sizep = &size_scratch;
2352
2353 if (oi->disk_sizep)
2354 *oi->disk_sizep = mapsize;
2355 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2356 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2357 status = error("unable to unpack %s header with --allow-unknown-type",
2358 sha1_to_hex(sha1));
2359 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2360 status = error("unable to unpack %s header",
2361 sha1_to_hex(sha1));
2362 if (status < 0)
2363 ; /* Do nothing */
2364 else if (hdrbuf.len) {
2365 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2366 status = error("unable to parse %s header with --allow-unknown-type",
2367 sha1_to_hex(sha1));
2368 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2369 status = error("unable to parse %s header", sha1_to_hex(sha1));
2370
2371 if (status >= 0 && oi->contentp)
2372 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2373 *oi->sizep, sha1);
2374 else
2375 git_inflate_end(&stream);
2376
2377 munmap(map, mapsize);
2378 if (status && oi->typep)
2379 *oi->typep = status;
2380 if (oi->sizep == &size_scratch)
2381 oi->sizep = NULL;
2382 strbuf_release(&hdrbuf);
2383 oi->whence = OI_LOOSE;
2384 return (status < 0) ? status : 0;
2385}
2386
2387int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2388{
2389 static struct object_info blank_oi = OBJECT_INFO_INIT;
2390 struct pack_entry e;
2391 int rtype;
2392 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2393 lookup_replace_object(sha1) :
2394 sha1;
2395
2396 if (!oi)
2397 oi = &blank_oi;
2398
2399 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2400 struct cached_object *co = find_cached_object(real);
2401 if (co) {
2402 if (oi->typep)
2403 *(oi->typep) = co->type;
2404 if (oi->sizep)
2405 *(oi->sizep) = co->size;
2406 if (oi->disk_sizep)
2407 *(oi->disk_sizep) = 0;
2408 if (oi->delta_base_sha1)
2409 hashclr(oi->delta_base_sha1);
2410 if (oi->typename)
2411 strbuf_addstr(oi->typename, typename(co->type));
2412 if (oi->contentp)
2413 *oi->contentp = xmemdupz(co->buf, co->size);
2414 oi->whence = OI_CACHED;
2415 return 0;
2416 }
2417 }
2418
2419 if (!find_pack_entry(real, &e)) {
2420 /* Most likely it's a loose object. */
2421 if (!sha1_loose_object_info(real, oi, flags))
2422 return 0;
2423
2424 /* Not a loose object; someone else may have just packed it. */
2425 if (flags & OBJECT_INFO_QUICK) {
2426 return -1;
2427 } else {
2428 reprepare_packed_git();
2429 if (!find_pack_entry(real, &e))
2430 return -1;
2431 }
2432 }
2433
2434 if (oi == &blank_oi)
2435 /*
2436 * We know that the caller doesn't actually need the
2437 * information below, so return early.
2438 */
2439 return 0;
2440
2441 rtype = packed_object_info(e.p, e.offset, oi);
2442 if (rtype < 0) {
2443 mark_bad_packed_object(e.p, real);
2444 return sha1_object_info_extended(real, oi, 0);
2445 } else if (oi->whence == OI_PACKED) {
2446 oi->u.packed.offset = e.offset;
2447 oi->u.packed.pack = e.p;
2448 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2449 rtype == OBJ_OFS_DELTA);
2450 }
2451
2452 return 0;
2453}
2454
2455/* returns enum object_type or negative */
2456int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2457{
2458 enum object_type type;
2459 struct object_info oi = OBJECT_INFO_INIT;
2460
2461 oi.typep = &type;
2462 oi.sizep = sizep;
2463 if (sha1_object_info_extended(sha1, &oi,
2464 OBJECT_INFO_LOOKUP_REPLACE) < 0)
2465 return -1;
2466 return type;
2467}
2468
2469int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2470 unsigned char *sha1)
2471{
2472 struct cached_object *co;
2473
2474 hash_sha1_file(buf, len, typename(type), sha1);
2475 if (has_sha1_file(sha1) || find_cached_object(sha1))
2476 return 0;
2477 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2478 co = &cached_objects[cached_object_nr++];
2479 co->size = len;
2480 co->type = type;
2481 co->buf = xmalloc(len);
2482 memcpy(co->buf, buf, len);
2483 hashcpy(co->sha1, sha1);
2484 return 0;
2485}
2486
2487static void *read_object(const unsigned char *sha1, enum object_type *type,
2488 unsigned long *size)
2489{
2490 struct object_info oi = OBJECT_INFO_INIT;
2491 void *content;
2492 oi.typep = type;
2493 oi.sizep = size;
2494 oi.contentp = &content;
2495
2496 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
2497 return NULL;
2498 return content;
2499}
2500
2501/*
2502 * This function dies on corrupt objects; the callers who want to
2503 * deal with them should arrange to call read_object() and give error
2504 * messages themselves.
2505 */
2506void *read_sha1_file_extended(const unsigned char *sha1,
2507 enum object_type *type,
2508 unsigned long *size,
2509 int lookup_replace)
2510{
2511 void *data;
2512 const struct packed_git *p;
2513 const char *path;
2514 struct stat st;
2515 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
2516 : sha1;
2517
2518 errno = 0;
2519 data = read_object(repl, type, size);
2520 if (data)
2521 return data;
2522
2523 if (errno && errno != ENOENT)
2524 die_errno("failed to read object %s", sha1_to_hex(sha1));
2525
2526 /* die if we replaced an object with one that does not exist */
2527 if (repl != sha1)
2528 die("replacement %s not found for %s",
2529 sha1_to_hex(repl), sha1_to_hex(sha1));
2530
2531 if (!stat_sha1_file(repl, &st, &path))
2532 die("loose object %s (stored in %s) is corrupt",
2533 sha1_to_hex(repl), path);
2534
2535 if ((p = has_packed_and_bad(repl)) != NULL)
2536 die("packed object %s (stored in %s) is corrupt",
2537 sha1_to_hex(repl), p->pack_name);
2538
2539 return NULL;
2540}
2541
2542void *read_object_with_reference(const unsigned char *sha1,
2543 const char *required_type_name,
2544 unsigned long *size,
2545 unsigned char *actual_sha1_return)
2546{
2547 enum object_type type, required_type;
2548 void *buffer;
2549 unsigned long isize;
2550 unsigned char actual_sha1[20];
2551
2552 required_type = type_from_string(required_type_name);
2553 hashcpy(actual_sha1, sha1);
2554 while (1) {
2555 int ref_length = -1;
2556 const char *ref_type = NULL;
2557
2558 buffer = read_sha1_file(actual_sha1, &type, &isize);
2559 if (!buffer)
2560 return NULL;
2561 if (type == required_type) {
2562 *size = isize;
2563 if (actual_sha1_return)
2564 hashcpy(actual_sha1_return, actual_sha1);
2565 return buffer;
2566 }
2567 /* Handle references */
2568 else if (type == OBJ_COMMIT)
2569 ref_type = "tree ";
2570 else if (type == OBJ_TAG)
2571 ref_type = "object ";
2572 else {
2573 free(buffer);
2574 return NULL;
2575 }
2576 ref_length = strlen(ref_type);
2577
2578 if (ref_length + 40 > isize ||
2579 memcmp(buffer, ref_type, ref_length) ||
2580 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2581 free(buffer);
2582 return NULL;
2583 }
2584 free(buffer);
2585 /* Now we have the ID of the referred-to object in
2586 * actual_sha1. Check again. */
2587 }
2588}
2589
2590static void write_sha1_file_prepare(const void *buf, unsigned long len,
2591 const char *type, unsigned char *sha1,
2592 char *hdr, int *hdrlen)
2593{
2594 git_SHA_CTX c;
2595
2596 /* Generate the header */
2597 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2598
2599 /* Sha1.. */
2600 git_SHA1_Init(&c);
2601 git_SHA1_Update(&c, hdr, *hdrlen);
2602 git_SHA1_Update(&c, buf, len);
2603 git_SHA1_Final(sha1, &c);
2604}
2605
2606/*
2607 * Move the just written object into its final resting place.
2608 */
2609int finalize_object_file(const char *tmpfile, const char *filename)
2610{
2611 int ret = 0;
2612
2613 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2614 goto try_rename;
2615 else if (link(tmpfile, filename))
2616 ret = errno;
2617
2618 /*
2619 * Coda hack - coda doesn't like cross-directory links,
2620 * so we fall back to a rename, which will mean that it
2621 * won't be able to check collisions, but that's not a
2622 * big deal.
2623 *
2624 * The same holds for FAT formatted media.
2625 *
2626 * When this succeeds, we just return. We have nothing
2627 * left to unlink.
2628 */
2629 if (ret && ret != EEXIST) {
2630 try_rename:
2631 if (!rename(tmpfile, filename))
2632 goto out;
2633 ret = errno;
2634 }
2635 unlink_or_warn(tmpfile);
2636 if (ret) {
2637 if (ret != EEXIST) {
2638 return error_errno("unable to write sha1 filename %s", filename);
2639 }
2640 /* FIXME!!! Collision check here ? */
2641 }
2642
2643out:
2644 if (adjust_shared_perm(filename))
2645 return error("unable to set permission to '%s'", filename);
2646 return 0;
2647}
2648
2649static int write_buffer(int fd, const void *buf, size_t len)
2650{
2651 if (write_in_full(fd, buf, len) < 0)
2652 return error_errno("file write error");
2653 return 0;
2654}
2655
2656int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2657 unsigned char *sha1)
2658{
2659 char hdr[32];
2660 int hdrlen = sizeof(hdr);
2661 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2662 return 0;
2663}
2664
2665/* Finalize a file on disk, and close it. */
2666static void close_sha1_file(int fd)
2667{
2668 if (fsync_object_files)
2669 fsync_or_die(fd, "sha1 file");
2670 if (close(fd) != 0)
2671 die_errno("error when closing sha1 file");
2672}
2673
2674/* Size of directory component, including the ending '/' */
2675static inline int directory_size(const char *filename)
2676{
2677 const char *s = strrchr(filename, '/');
2678 if (!s)
2679 return 0;
2680 return s - filename + 1;
2681}
2682
2683/*
2684 * This creates a temporary file in the same directory as the final
2685 * 'filename'
2686 *
2687 * We want to avoid cross-directory filename renames, because those
2688 * can have problems on various filesystems (FAT, NFS, Coda).
2689 */
2690static int create_tmpfile(struct strbuf *tmp, const char *filename)
2691{
2692 int fd, dirlen = directory_size(filename);
2693
2694 strbuf_reset(tmp);
2695 strbuf_add(tmp, filename, dirlen);
2696 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2697 fd = git_mkstemp_mode(tmp->buf, 0444);
2698 if (fd < 0 && dirlen && errno == ENOENT) {
2699 /*
2700 * Make sure the directory exists; note that the contents
2701 * of the buffer are undefined after mkstemp returns an
2702 * error, so we have to rewrite the whole buffer from
2703 * scratch.
2704 */
2705 strbuf_reset(tmp);
2706 strbuf_add(tmp, filename, dirlen - 1);
2707 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
2708 return -1;
2709 if (adjust_shared_perm(tmp->buf))
2710 return -1;
2711
2712 /* Try again */
2713 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
2714 fd = git_mkstemp_mode(tmp->buf, 0444);
2715 }
2716 return fd;
2717}
2718
2719static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2720 const void *buf, unsigned long len, time_t mtime)
2721{
2722 int fd, ret;
2723 unsigned char compressed[4096];
2724 git_zstream stream;
2725 git_SHA_CTX c;
2726 unsigned char parano_sha1[20];
2727 static struct strbuf tmp_file = STRBUF_INIT;
2728 const char *filename = sha1_file_name(sha1);
2729
2730 fd = create_tmpfile(&tmp_file, filename);
2731 if (fd < 0) {
2732 if (errno == EACCES)
2733 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2734 else
2735 return error_errno("unable to create temporary file");
2736 }
2737
2738 /* Set it up */
2739 git_deflate_init(&stream, zlib_compression_level);
2740 stream.next_out = compressed;
2741 stream.avail_out = sizeof(compressed);
2742 git_SHA1_Init(&c);
2743
2744 /* First header.. */
2745 stream.next_in = (unsigned char *)hdr;
2746 stream.avail_in = hdrlen;
2747 while (git_deflate(&stream, 0) == Z_OK)
2748 ; /* nothing */
2749 git_SHA1_Update(&c, hdr, hdrlen);
2750
2751 /* Then the data itself.. */
2752 stream.next_in = (void *)buf;
2753 stream.avail_in = len;
2754 do {
2755 unsigned char *in0 = stream.next_in;
2756 ret = git_deflate(&stream, Z_FINISH);
2757 git_SHA1_Update(&c, in0, stream.next_in - in0);
2758 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2759 die("unable to write sha1 file");
2760 stream.next_out = compressed;
2761 stream.avail_out = sizeof(compressed);
2762 } while (ret == Z_OK);
2763
2764 if (ret != Z_STREAM_END)
2765 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2766 ret = git_deflate_end_gently(&stream);
2767 if (ret != Z_OK)
2768 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2769 git_SHA1_Final(parano_sha1, &c);
2770 if (hashcmp(sha1, parano_sha1) != 0)
2771 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2772
2773 close_sha1_file(fd);
2774
2775 if (mtime) {
2776 struct utimbuf utb;
2777 utb.actime = mtime;
2778 utb.modtime = mtime;
2779 if (utime(tmp_file.buf, &utb) < 0)
2780 warning_errno("failed utime() on %s", tmp_file.buf);
2781 }
2782
2783 return finalize_object_file(tmp_file.buf, filename);
2784}
2785
2786static int freshen_loose_object(const unsigned char *sha1)
2787{
2788 return check_and_freshen(sha1, 1);
2789}
2790
2791static int freshen_packed_object(const unsigned char *sha1)
2792{
2793 struct pack_entry e;
2794 if (!find_pack_entry(sha1, &e))
2795 return 0;
2796 if (e.p->freshened)
2797 return 1;
2798 if (!freshen_file(e.p->pack_name))
2799 return 0;
2800 e.p->freshened = 1;
2801 return 1;
2802}
2803
2804int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
2805{
2806 char hdr[32];
2807 int hdrlen = sizeof(hdr);
2808
2809 /* Normally if we have it in the pack then we do not bother writing
2810 * it out into .git/objects/??/?{38} file.
2811 */
2812 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2813 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2814 return 0;
2815 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2816}
2817
2818int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
2819 unsigned char *sha1, unsigned flags)
2820{
2821 char *header;
2822 int hdrlen, status = 0;
2823
2824 /* type string, SP, %lu of the length plus NUL must fit this */
2825 hdrlen = strlen(type) + 32;
2826 header = xmalloc(hdrlen);
2827 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
2828
2829 if (!(flags & HASH_WRITE_OBJECT))
2830 goto cleanup;
2831 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2832 goto cleanup;
2833 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
2834
2835cleanup:
2836 free(header);
2837 return status;
2838}
2839
2840int force_object_loose(const unsigned char *sha1, time_t mtime)
2841{
2842 void *buf;
2843 unsigned long len;
2844 enum object_type type;
2845 char hdr[32];
2846 int hdrlen;
2847 int ret;
2848
2849 if (has_loose_object(sha1))
2850 return 0;
2851 buf = read_object(sha1, &type, &len);
2852 if (!buf)
2853 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2854 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
2855 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2856 free(buf);
2857
2858 return ret;
2859}
2860
2861int has_pack_index(const unsigned char *sha1)
2862{
2863 struct stat st;
2864 if (stat(sha1_pack_index_name(sha1), &st))
2865 return 0;
2866 return 1;
2867}
2868
2869int has_sha1_pack(const unsigned char *sha1)
2870{
2871 struct pack_entry e;
2872 return find_pack_entry(sha1, &e);
2873}
2874
2875int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
2876{
2877 if (!startup_info->have_repository)
2878 return 0;
2879 return sha1_object_info_extended(sha1, NULL,
2880 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
2881}
2882
2883int has_object_file(const struct object_id *oid)
2884{
2885 return has_sha1_file(oid->hash);
2886}
2887
2888int has_object_file_with_flags(const struct object_id *oid, int flags)
2889{
2890 return has_sha1_file_with_flags(oid->hash, flags);
2891}
2892
2893static void check_tree(const void *buf, size_t size)
2894{
2895 struct tree_desc desc;
2896 struct name_entry entry;
2897
2898 init_tree_desc(&desc, buf, size);
2899 while (tree_entry(&desc, &entry))
2900 /* do nothing
2901 * tree_entry() will die() on malformed entries */
2902 ;
2903}
2904
2905static void check_commit(const void *buf, size_t size)
2906{
2907 struct commit c;
2908 memset(&c, 0, sizeof(c));
2909 if (parse_commit_buffer(&c, buf, size))
2910 die("corrupt commit");
2911}
2912
2913static void check_tag(const void *buf, size_t size)
2914{
2915 struct tag t;
2916 memset(&t, 0, sizeof(t));
2917 if (parse_tag_buffer(&t, buf, size))
2918 die("corrupt tag");
2919}
2920
2921static int index_mem(unsigned char *sha1, void *buf, size_t size,
2922 enum object_type type,
2923 const char *path, unsigned flags)
2924{
2925 int ret, re_allocated = 0;
2926 int write_object = flags & HASH_WRITE_OBJECT;
2927
2928 if (!type)
2929 type = OBJ_BLOB;
2930
2931 /*
2932 * Convert blobs to git internal format
2933 */
2934 if ((type == OBJ_BLOB) && path) {
2935 struct strbuf nbuf = STRBUF_INIT;
2936 if (convert_to_git(&the_index, path, buf, size, &nbuf,
2937 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2938 buf = strbuf_detach(&nbuf, &size);
2939 re_allocated = 1;
2940 }
2941 }
2942 if (flags & HASH_FORMAT_CHECK) {
2943 if (type == OBJ_TREE)
2944 check_tree(buf, size);
2945 if (type == OBJ_COMMIT)
2946 check_commit(buf, size);
2947 if (type == OBJ_TAG)
2948 check_tag(buf, size);
2949 }
2950
2951 if (write_object)
2952 ret = write_sha1_file(buf, size, typename(type), sha1);
2953 else
2954 ret = hash_sha1_file(buf, size, typename(type), sha1);
2955 if (re_allocated)
2956 free(buf);
2957 return ret;
2958}
2959
2960static int index_stream_convert_blob(unsigned char *sha1, int fd,
2961 const char *path, unsigned flags)
2962{
2963 int ret;
2964 const int write_object = flags & HASH_WRITE_OBJECT;
2965 struct strbuf sbuf = STRBUF_INIT;
2966
2967 assert(path);
2968 assert(would_convert_to_git_filter_fd(path));
2969
2970 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
2971 write_object ? safe_crlf : SAFE_CRLF_FALSE);
2972
2973 if (write_object)
2974 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2975 sha1);
2976 else
2977 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2978 sha1);
2979 strbuf_release(&sbuf);
2980 return ret;
2981}
2982
2983static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2984 const char *path, unsigned flags)
2985{
2986 struct strbuf sbuf = STRBUF_INIT;
2987 int ret;
2988
2989 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2990 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2991 else
2992 ret = -1;
2993 strbuf_release(&sbuf);
2994 return ret;
2995}
2996
2997#define SMALL_FILE_SIZE (32*1024)
2998
2999static int index_core(unsigned char *sha1, int fd, size_t size,
3000 enum object_type type, const char *path,
3001 unsigned flags)
3002{
3003 int ret;
3004
3005 if (!size) {
3006 ret = index_mem(sha1, "", size, type, path, flags);
3007 } else if (size <= SMALL_FILE_SIZE) {
3008 char *buf = xmalloc(size);
3009 if (size == read_in_full(fd, buf, size))
3010 ret = index_mem(sha1, buf, size, type, path, flags);
3011 else
3012 ret = error_errno("short read");
3013 free(buf);
3014 } else {
3015 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3016 ret = index_mem(sha1, buf, size, type, path, flags);
3017 munmap(buf, size);
3018 }
3019 return ret;
3020}
3021
3022/*
3023 * This creates one packfile per large blob unless bulk-checkin
3024 * machinery is "plugged".
3025 *
3026 * This also bypasses the usual "convert-to-git" dance, and that is on
3027 * purpose. We could write a streaming version of the converting
3028 * functions and insert that before feeding the data to fast-import
3029 * (or equivalent in-core API described above). However, that is
3030 * somewhat complicated, as we do not know the size of the filter
3031 * result, which we need to know beforehand when writing a git object.
3032 * Since the primary motivation for trying to stream from the working
3033 * tree file and to avoid mmaping it in core is to deal with large
3034 * binary blobs, they generally do not want to get any conversion, and
3035 * callers should avoid this code path when filters are requested.
3036 */
3037static int index_stream(unsigned char *sha1, int fd, size_t size,
3038 enum object_type type, const char *path,
3039 unsigned flags)
3040{
3041 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3042}
3043
3044int index_fd(unsigned char *sha1, int fd, struct stat *st,
3045 enum object_type type, const char *path, unsigned flags)
3046{
3047 int ret;
3048
3049 /*
3050 * Call xsize_t() only when needed to avoid potentially unnecessary
3051 * die() for large files.
3052 */
3053 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3054 ret = index_stream_convert_blob(sha1, fd, path, flags);
3055 else if (!S_ISREG(st->st_mode))
3056 ret = index_pipe(sha1, fd, type, path, flags);
3057 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3058 (path && would_convert_to_git(&the_index, path)))
3059 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3060 flags);
3061 else
3062 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3063 flags);
3064 close(fd);
3065 return ret;
3066}
3067
3068int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3069{
3070 int fd;
3071 struct strbuf sb = STRBUF_INIT;
3072
3073 switch (st->st_mode & S_IFMT) {
3074 case S_IFREG:
3075 fd = open(path, O_RDONLY);
3076 if (fd < 0)
3077 return error_errno("open(\"%s\")", path);
3078 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3079 return error("%s: failed to insert into database",
3080 path);
3081 break;
3082 case S_IFLNK:
3083 if (strbuf_readlink(&sb, path, st->st_size))
3084 return error_errno("readlink(\"%s\")", path);
3085 if (!(flags & HASH_WRITE_OBJECT))
3086 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3087 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3088 return error("%s: failed to insert into database",
3089 path);
3090 strbuf_release(&sb);
3091 break;
3092 case S_IFDIR:
3093 return resolve_gitlink_ref(path, "HEAD", sha1);
3094 default:
3095 return error("%s: unsupported file type", path);
3096 }
3097 return 0;
3098}
3099
3100int read_pack_header(int fd, struct pack_header *header)
3101{
3102 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3103 /* "eof before pack header was fully read" */
3104 return PH_ERROR_EOF;
3105
3106 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3107 /* "protocol error (pack signature mismatch detected)" */
3108 return PH_ERROR_PACK_SIGNATURE;
3109 if (!pack_version_ok(header->hdr_version))
3110 /* "protocol error (pack version unsupported)" */
3111 return PH_ERROR_PROTOCOL;
3112 return 0;
3113}
3114
3115void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3116{
3117 enum object_type type = sha1_object_info(sha1, NULL);
3118 if (type < 0)
3119 die("%s is not a valid object", sha1_to_hex(sha1));
3120 if (type != expect)
3121 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3122 typename(expect));
3123}
3124
3125int for_each_file_in_obj_subdir(unsigned int subdir_nr,
3126 struct strbuf *path,
3127 each_loose_object_fn obj_cb,
3128 each_loose_cruft_fn cruft_cb,
3129 each_loose_subdir_fn subdir_cb,
3130 void *data)
3131{
3132 size_t origlen, baselen;
3133 DIR *dir;
3134 struct dirent *de;
3135 int r = 0;
3136
3137 if (subdir_nr > 0xff)
3138 BUG("invalid loose object subdirectory: %x", subdir_nr);
3139
3140 origlen = path->len;
3141 strbuf_complete(path, '/');
3142 strbuf_addf(path, "%02x", subdir_nr);
3143 baselen = path->len;
3144
3145 dir = opendir(path->buf);
3146 if (!dir) {
3147 if (errno != ENOENT)
3148 r = error_errno("unable to open %s", path->buf);
3149 strbuf_setlen(path, origlen);
3150 return r;
3151 }
3152
3153 while ((de = readdir(dir))) {
3154 if (is_dot_or_dotdot(de->d_name))
3155 continue;
3156
3157 strbuf_setlen(path, baselen);
3158 strbuf_addf(path, "/%s", de->d_name);
3159
3160 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3161 char hex[GIT_MAX_HEXSZ+1];
3162 struct object_id oid;
3163
3164 xsnprintf(hex, sizeof(hex), "%02x%s",
3165 subdir_nr, de->d_name);
3166 if (!get_oid_hex(hex, &oid)) {
3167 if (obj_cb) {
3168 r = obj_cb(&oid, path->buf, data);
3169 if (r)
3170 break;
3171 }
3172 continue;
3173 }
3174 }
3175
3176 if (cruft_cb) {
3177 r = cruft_cb(de->d_name, path->buf, data);
3178 if (r)
3179 break;
3180 }
3181 }
3182 closedir(dir);
3183
3184 strbuf_setlen(path, baselen);
3185 if (!r && subdir_cb)
3186 r = subdir_cb(subdir_nr, path->buf, data);
3187
3188 strbuf_setlen(path, origlen);
3189
3190 return r;
3191}
3192
3193int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3194 each_loose_object_fn obj_cb,
3195 each_loose_cruft_fn cruft_cb,
3196 each_loose_subdir_fn subdir_cb,
3197 void *data)
3198{
3199 int r = 0;
3200 int i;
3201
3202 for (i = 0; i < 256; i++) {
3203 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3204 subdir_cb, data);
3205 if (r)
3206 break;
3207 }
3208
3209 return r;
3210}
3211
3212int for_each_loose_file_in_objdir(const char *path,
3213 each_loose_object_fn obj_cb,
3214 each_loose_cruft_fn cruft_cb,
3215 each_loose_subdir_fn subdir_cb,
3216 void *data)
3217{
3218 struct strbuf buf = STRBUF_INIT;
3219 int r;
3220
3221 strbuf_addstr(&buf, path);
3222 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3223 subdir_cb, data);
3224 strbuf_release(&buf);
3225
3226 return r;
3227}
3228
3229struct loose_alt_odb_data {
3230 each_loose_object_fn *cb;
3231 void *data;
3232};
3233
3234static int loose_from_alt_odb(struct alternate_object_database *alt,
3235 void *vdata)
3236{
3237 struct loose_alt_odb_data *data = vdata;
3238 struct strbuf buf = STRBUF_INIT;
3239 int r;
3240
3241 strbuf_addstr(&buf, alt->path);
3242 r = for_each_loose_file_in_objdir_buf(&buf,
3243 data->cb, NULL, NULL,
3244 data->data);
3245 strbuf_release(&buf);
3246 return r;
3247}
3248
3249int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3250{
3251 struct loose_alt_odb_data alt;
3252 int r;
3253
3254 r = for_each_loose_file_in_objdir(get_object_directory(),
3255 cb, NULL, NULL, data);
3256 if (r)
3257 return r;
3258
3259 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3260 return 0;
3261
3262 alt.cb = cb;
3263 alt.data = data;
3264 return foreach_alt_odb(loose_from_alt_odb, &alt);
3265}
3266
3267static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3268{
3269 uint32_t i;
3270 int r = 0;
3271
3272 for (i = 0; i < p->num_objects; i++) {
3273 struct object_id oid;
3274
3275 if (!nth_packed_object_oid(&oid, p, i))
3276 return error("unable to get sha1 of object %u in %s",
3277 i, p->pack_name);
3278
3279 r = cb(&oid, p, i, data);
3280 if (r)
3281 break;
3282 }
3283 return r;
3284}
3285
3286int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3287{
3288 struct packed_git *p;
3289 int r = 0;
3290 int pack_errors = 0;
3291
3292 prepare_packed_git();
3293 for (p = packed_git; p; p = p->next) {
3294 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3295 continue;
3296 if (open_pack_index(p)) {
3297 pack_errors = 1;
3298 continue;
3299 }
3300 r = for_each_object_in_pack(p, cb, data);
3301 if (r)
3302 break;
3303 }
3304 return r ? r : pack_errors;
3305}
3306
3307static int check_stream_sha1(git_zstream *stream,
3308 const char *hdr,
3309 unsigned long size,
3310 const char *path,
3311 const unsigned char *expected_sha1)
3312{
3313 git_SHA_CTX c;
3314 unsigned char real_sha1[GIT_MAX_RAWSZ];
3315 unsigned char buf[4096];
3316 unsigned long total_read;
3317 int status = Z_OK;
3318
3319 git_SHA1_Init(&c);
3320 git_SHA1_Update(&c, hdr, stream->total_out);
3321
3322 /*
3323 * We already read some bytes into hdr, but the ones up to the NUL
3324 * do not count against the object's content size.
3325 */
3326 total_read = stream->total_out - strlen(hdr) - 1;
3327
3328 /*
3329 * This size comparison must be "<=" to read the final zlib packets;
3330 * see the comment in unpack_sha1_rest for details.
3331 */
3332 while (total_read <= size &&
3333 (status == Z_OK || status == Z_BUF_ERROR)) {
3334 stream->next_out = buf;
3335 stream->avail_out = sizeof(buf);
3336 if (size - total_read < stream->avail_out)
3337 stream->avail_out = size - total_read;
3338 status = git_inflate(stream, Z_FINISH);
3339 git_SHA1_Update(&c, buf, stream->next_out - buf);
3340 total_read += stream->next_out - buf;
3341 }
3342 git_inflate_end(stream);
3343
3344 if (status != Z_STREAM_END) {
3345 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3346 return -1;
3347 }
3348 if (stream->avail_in) {
3349 error("garbage at end of loose object '%s'",
3350 sha1_to_hex(expected_sha1));
3351 return -1;
3352 }
3353
3354 git_SHA1_Final(real_sha1, &c);
3355 if (hashcmp(expected_sha1, real_sha1)) {
3356 error("sha1 mismatch for %s (expected %s)", path,
3357 sha1_to_hex(expected_sha1));
3358 return -1;
3359 }
3360
3361 return 0;
3362}
3363
3364int read_loose_object(const char *path,
3365 const unsigned char *expected_sha1,
3366 enum object_type *type,
3367 unsigned long *size,
3368 void **contents)
3369{
3370 int ret = -1;
3371 void *map = NULL;
3372 unsigned long mapsize;
3373 git_zstream stream;
3374 char hdr[32];
3375
3376 *contents = NULL;
3377
3378 map = map_sha1_file_1(path, NULL, &mapsize);
3379 if (!map) {
3380 error_errno("unable to mmap %s", path);
3381 goto out;
3382 }
3383
3384 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3385 error("unable to unpack header of %s", path);
3386 goto out;
3387 }
3388
3389 *type = parse_sha1_header(hdr, size);
3390 if (*type < 0) {
3391 error("unable to parse header of %s", path);
3392 git_inflate_end(&stream);
3393 goto out;
3394 }
3395
3396 if (*type == OBJ_BLOB) {
3397 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3398 goto out;
3399 } else {
3400 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3401 if (!*contents) {
3402 error("unable to unpack contents of %s", path);
3403 git_inflate_end(&stream);
3404 goto out;
3405 }
3406 if (check_sha1_signature(expected_sha1, *contents,
3407 *size, typename(*type))) {
3408 error("sha1 mismatch for %s (expected %s)", path,
3409 sha1_to_hex(expected_sha1));
3410 free(*contents);
3411 goto out;
3412 }
3413 }
3414
3415 ret = 0; /* everything checks out */
3416
3417out:
3418 if (map)
3419 munmap(map, mapsize);
3420 return ret;
3421}