9df1bd7c17fffac9c9d71cf32470f6a0dda5eaae
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
722static struct packed_git *alloc_packed_git(int extra)
723{
724 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
725 memset(p, 0, sizeof(*p));
726 p->pack_fd = -1;
727 return p;
728}
729
730static void try_to_free_pack_memory(size_t size)
731{
732 release_pack_memory(size);
733}
734
735struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
736{
737 static int have_set_try_to_free_routine;
738 struct stat st;
739 size_t alloc;
740 struct packed_git *p;
741
742 if (!have_set_try_to_free_routine) {
743 have_set_try_to_free_routine = 1;
744 set_try_to_free_routine(try_to_free_pack_memory);
745 }
746
747 /*
748 * Make sure a corresponding .pack file exists and that
749 * the index looks sane.
750 */
751 if (!strip_suffix_mem(path, &path_len, ".idx"))
752 return NULL;
753
754 /*
755 * ".pack" is long enough to hold any suffix we're adding (and
756 * the use xsnprintf double-checks that)
757 */
758 alloc = st_add3(path_len, strlen(".pack"), 1);
759 p = alloc_packed_git(alloc);
760 memcpy(p->pack_name, path, path_len);
761
762 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
763 if (!access(p->pack_name, F_OK))
764 p->pack_keep = 1;
765
766 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
767 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
768 free(p);
769 return NULL;
770 }
771
772 /* ok, it looks sane as far as we can check without
773 * actually mapping the pack file.
774 */
775 p->pack_size = st.st_size;
776 p->pack_local = local;
777 p->mtime = st.st_mtime;
778 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
779 hashclr(p->sha1);
780 return p;
781}
782
783void install_packed_git(struct packed_git *pack)
784{
785 if (pack->pack_fd != -1)
786 pack_open_fds++;
787
788 pack->next = packed_git;
789 packed_git = pack;
790}
791
792void (*report_garbage)(unsigned seen_bits, const char *path);
793
794static void report_helper(const struct string_list *list,
795 int seen_bits, int first, int last)
796{
797 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
798 return;
799
800 for (; first < last; first++)
801 report_garbage(seen_bits, list->items[first].string);
802}
803
804static void report_pack_garbage(struct string_list *list)
805{
806 int i, baselen = -1, first = 0, seen_bits = 0;
807
808 if (!report_garbage)
809 return;
810
811 string_list_sort(list);
812
813 for (i = 0; i < list->nr; i++) {
814 const char *path = list->items[i].string;
815 if (baselen != -1 &&
816 strncmp(path, list->items[first].string, baselen)) {
817 report_helper(list, seen_bits, first, i);
818 baselen = -1;
819 seen_bits = 0;
820 }
821 if (baselen == -1) {
822 const char *dot = strrchr(path, '.');
823 if (!dot) {
824 report_garbage(PACKDIR_FILE_GARBAGE, path);
825 continue;
826 }
827 baselen = dot - path + 1;
828 first = i;
829 }
830 if (!strcmp(path + baselen, "pack"))
831 seen_bits |= 1;
832 else if (!strcmp(path + baselen, "idx"))
833 seen_bits |= 2;
834 }
835 report_helper(list, seen_bits, first, list->nr);
836}
837
838static void prepare_packed_git_one(char *objdir, int local)
839{
840 struct strbuf path = STRBUF_INIT;
841 size_t dirnamelen;
842 DIR *dir;
843 struct dirent *de;
844 struct string_list garbage = STRING_LIST_INIT_DUP;
845
846 strbuf_addstr(&path, objdir);
847 strbuf_addstr(&path, "/pack");
848 dir = opendir(path.buf);
849 if (!dir) {
850 if (errno != ENOENT)
851 error_errno("unable to open object pack directory: %s",
852 path.buf);
853 strbuf_release(&path);
854 return;
855 }
856 strbuf_addch(&path, '/');
857 dirnamelen = path.len;
858 while ((de = readdir(dir)) != NULL) {
859 struct packed_git *p;
860 size_t base_len;
861
862 if (is_dot_or_dotdot(de->d_name))
863 continue;
864
865 strbuf_setlen(&path, dirnamelen);
866 strbuf_addstr(&path, de->d_name);
867
868 base_len = path.len;
869 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
870 /* Don't reopen a pack we already have. */
871 for (p = packed_git; p; p = p->next) {
872 size_t len;
873 if (strip_suffix(p->pack_name, ".pack", &len) &&
874 len == base_len &&
875 !memcmp(p->pack_name, path.buf, len))
876 break;
877 }
878 if (p == NULL &&
879 /*
880 * See if it really is a valid .idx file with
881 * corresponding .pack file that we can map.
882 */
883 (p = add_packed_git(path.buf, path.len, local)) != NULL)
884 install_packed_git(p);
885 }
886
887 if (!report_garbage)
888 continue;
889
890 if (ends_with(de->d_name, ".idx") ||
891 ends_with(de->d_name, ".pack") ||
892 ends_with(de->d_name, ".bitmap") ||
893 ends_with(de->d_name, ".keep"))
894 string_list_append(&garbage, path.buf);
895 else
896 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
897 }
898 closedir(dir);
899 report_pack_garbage(&garbage);
900 string_list_clear(&garbage, 0);
901 strbuf_release(&path);
902}
903
904static int approximate_object_count_valid;
905
906/*
907 * Give a fast, rough count of the number of objects in the repository. This
908 * ignores loose objects completely. If you have a lot of them, then either
909 * you should repack because your performance will be awful, or they are
910 * all unreachable objects about to be pruned, in which case they're not really
911 * interesting as a measure of repo size in the first place.
912 */
913unsigned long approximate_object_count(void)
914{
915 static unsigned long count;
916 if (!approximate_object_count_valid) {
917 struct packed_git *p;
918
919 prepare_packed_git();
920 count = 0;
921 for (p = packed_git; p; p = p->next) {
922 if (open_pack_index(p))
923 continue;
924 count += p->num_objects;
925 }
926 }
927 return count;
928}
929
930static void *get_next_packed_git(const void *p)
931{
932 return ((const struct packed_git *)p)->next;
933}
934
935static void set_next_packed_git(void *p, void *next)
936{
937 ((struct packed_git *)p)->next = next;
938}
939
940static int sort_pack(const void *a_, const void *b_)
941{
942 const struct packed_git *a = a_;
943 const struct packed_git *b = b_;
944 int st;
945
946 /*
947 * Local packs tend to contain objects specific to our
948 * variant of the project than remote ones. In addition,
949 * remote ones could be on a network mounted filesystem.
950 * Favor local ones for these reasons.
951 */
952 st = a->pack_local - b->pack_local;
953 if (st)
954 return -st;
955
956 /*
957 * Younger packs tend to contain more recent objects,
958 * and more recent objects tend to get accessed more
959 * often.
960 */
961 if (a->mtime < b->mtime)
962 return 1;
963 else if (a->mtime == b->mtime)
964 return 0;
965 return -1;
966}
967
968static void rearrange_packed_git(void)
969{
970 packed_git = llist_mergesort(packed_git, get_next_packed_git,
971 set_next_packed_git, sort_pack);
972}
973
974static void prepare_packed_git_mru(void)
975{
976 struct packed_git *p;
977
978 mru_clear(packed_git_mru);
979 for (p = packed_git; p; p = p->next)
980 mru_append(packed_git_mru, p);
981}
982
983static int prepare_packed_git_run_once = 0;
984void prepare_packed_git(void)
985{
986 struct alternate_object_database *alt;
987
988 if (prepare_packed_git_run_once)
989 return;
990 prepare_packed_git_one(get_object_directory(), 1);
991 prepare_alt_odb();
992 for (alt = alt_odb_list; alt; alt = alt->next)
993 prepare_packed_git_one(alt->path, 0);
994 rearrange_packed_git();
995 prepare_packed_git_mru();
996 prepare_packed_git_run_once = 1;
997}
998
999void reprepare_packed_git(void)
1000{
1001 approximate_object_count_valid = 0;
1002 prepare_packed_git_run_once = 0;
1003 prepare_packed_git();
1004}
1005
1006static void mark_bad_packed_object(struct packed_git *p,
1007 const unsigned char *sha1)
1008{
1009 unsigned i;
1010 for (i = 0; i < p->num_bad_objects; i++)
1011 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1012 return;
1013 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1014 st_mult(GIT_MAX_RAWSZ,
1015 st_add(p->num_bad_objects, 1)));
1016 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1017 p->num_bad_objects++;
1018}
1019
1020static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1021{
1022 struct packed_git *p;
1023 unsigned i;
1024
1025 for (p = packed_git; p; p = p->next)
1026 for (i = 0; i < p->num_bad_objects; i++)
1027 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1028 return p;
1029 return NULL;
1030}
1031
1032/*
1033 * With an in-core object data in "map", rehash it to make sure the
1034 * object name actually matches "sha1" to detect object corruption.
1035 * With "map" == NULL, try reading the object named with "sha1" using
1036 * the streaming interface and rehash it to do the same.
1037 */
1038int check_sha1_signature(const unsigned char *sha1, void *map,
1039 unsigned long size, const char *type)
1040{
1041 unsigned char real_sha1[20];
1042 enum object_type obj_type;
1043 struct git_istream *st;
1044 git_SHA_CTX c;
1045 char hdr[32];
1046 int hdrlen;
1047
1048 if (map) {
1049 hash_sha1_file(map, size, type, real_sha1);
1050 return hashcmp(sha1, real_sha1) ? -1 : 0;
1051 }
1052
1053 st = open_istream(sha1, &obj_type, &size, NULL);
1054 if (!st)
1055 return -1;
1056
1057 /* Generate the header */
1058 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1059
1060 /* Sha1.. */
1061 git_SHA1_Init(&c);
1062 git_SHA1_Update(&c, hdr, hdrlen);
1063 for (;;) {
1064 char buf[1024 * 16];
1065 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1066
1067 if (readlen < 0) {
1068 close_istream(st);
1069 return -1;
1070 }
1071 if (!readlen)
1072 break;
1073 git_SHA1_Update(&c, buf, readlen);
1074 }
1075 git_SHA1_Final(real_sha1, &c);
1076 close_istream(st);
1077 return hashcmp(sha1, real_sha1) ? -1 : 0;
1078}
1079
1080int git_open_cloexec(const char *name, int flags)
1081{
1082 int fd;
1083 static int o_cloexec = O_CLOEXEC;
1084
1085 fd = open(name, flags | o_cloexec);
1086 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1087 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1088 o_cloexec &= ~O_CLOEXEC;
1089 fd = open(name, flags | o_cloexec);
1090 }
1091
1092#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1093 {
1094 static int fd_cloexec = FD_CLOEXEC;
1095
1096 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1097 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1098 int flags = fcntl(fd, F_GETFD);
1099 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1100 fd_cloexec = 0;
1101 }
1102 }
1103#endif
1104 return fd;
1105}
1106
1107/*
1108 * Find "sha1" as a loose object in the local repository or in an alternate.
1109 * Returns 0 on success, negative on failure.
1110 *
1111 * The "path" out-parameter will give the path of the object we found (if any).
1112 * Note that it may point to static storage and is only valid until another
1113 * call to sha1_file_name(), etc.
1114 */
1115static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1116 const char **path)
1117{
1118 struct alternate_object_database *alt;
1119
1120 *path = sha1_file_name(sha1);
1121 if (!lstat(*path, st))
1122 return 0;
1123
1124 prepare_alt_odb();
1125 errno = ENOENT;
1126 for (alt = alt_odb_list; alt; alt = alt->next) {
1127 *path = alt_sha1_path(alt, sha1);
1128 if (!lstat(*path, st))
1129 return 0;
1130 }
1131
1132 return -1;
1133}
1134
1135/*
1136 * Like stat_sha1_file(), but actually open the object and return the
1137 * descriptor. See the caveats on the "path" parameter above.
1138 */
1139static int open_sha1_file(const unsigned char *sha1, const char **path)
1140{
1141 int fd;
1142 struct alternate_object_database *alt;
1143 int most_interesting_errno;
1144
1145 *path = sha1_file_name(sha1);
1146 fd = git_open(*path);
1147 if (fd >= 0)
1148 return fd;
1149 most_interesting_errno = errno;
1150
1151 prepare_alt_odb();
1152 for (alt = alt_odb_list; alt; alt = alt->next) {
1153 *path = alt_sha1_path(alt, sha1);
1154 fd = git_open(*path);
1155 if (fd >= 0)
1156 return fd;
1157 if (most_interesting_errno == ENOENT)
1158 most_interesting_errno = errno;
1159 }
1160 errno = most_interesting_errno;
1161 return -1;
1162}
1163
1164/*
1165 * Map the loose object at "path" if it is not NULL, or the path found by
1166 * searching for a loose object named "sha1".
1167 */
1168static void *map_sha1_file_1(const char *path,
1169 const unsigned char *sha1,
1170 unsigned long *size)
1171{
1172 void *map;
1173 int fd;
1174
1175 if (path)
1176 fd = git_open(path);
1177 else
1178 fd = open_sha1_file(sha1, &path);
1179 map = NULL;
1180 if (fd >= 0) {
1181 struct stat st;
1182
1183 if (!fstat(fd, &st)) {
1184 *size = xsize_t(st.st_size);
1185 if (!*size) {
1186 /* mmap() is forbidden on empty files */
1187 error("object file %s is empty", path);
1188 return NULL;
1189 }
1190 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1191 }
1192 close(fd);
1193 }
1194 return map;
1195}
1196
1197void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1198{
1199 return map_sha1_file_1(NULL, sha1, size);
1200}
1201
1202unsigned long unpack_object_header_buffer(const unsigned char *buf,
1203 unsigned long len, enum object_type *type, unsigned long *sizep)
1204{
1205 unsigned shift;
1206 unsigned long size, c;
1207 unsigned long used = 0;
1208
1209 c = buf[used++];
1210 *type = (c >> 4) & 7;
1211 size = c & 15;
1212 shift = 4;
1213 while (c & 0x80) {
1214 if (len <= used || bitsizeof(long) <= shift) {
1215 error("bad object header");
1216 size = used = 0;
1217 break;
1218 }
1219 c = buf[used++];
1220 size += (c & 0x7f) << shift;
1221 shift += 7;
1222 }
1223 *sizep = size;
1224 return used;
1225}
1226
1227static int unpack_sha1_short_header(git_zstream *stream,
1228 unsigned char *map, unsigned long mapsize,
1229 void *buffer, unsigned long bufsiz)
1230{
1231 /* Get the data stream */
1232 memset(stream, 0, sizeof(*stream));
1233 stream->next_in = map;
1234 stream->avail_in = mapsize;
1235 stream->next_out = buffer;
1236 stream->avail_out = bufsiz;
1237
1238 git_inflate_init(stream);
1239 return git_inflate(stream, 0);
1240}
1241
1242int unpack_sha1_header(git_zstream *stream,
1243 unsigned char *map, unsigned long mapsize,
1244 void *buffer, unsigned long bufsiz)
1245{
1246 int status = unpack_sha1_short_header(stream, map, mapsize,
1247 buffer, bufsiz);
1248
1249 if (status < Z_OK)
1250 return status;
1251
1252 /* Make sure we have the terminating NUL */
1253 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1254 return -1;
1255 return 0;
1256}
1257
1258static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1259 unsigned long mapsize, void *buffer,
1260 unsigned long bufsiz, struct strbuf *header)
1261{
1262 int status;
1263
1264 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1265 if (status < Z_OK)
1266 return -1;
1267
1268 /*
1269 * Check if entire header is unpacked in the first iteration.
1270 */
1271 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1272 return 0;
1273
1274 /*
1275 * buffer[0..bufsiz] was not large enough. Copy the partial
1276 * result out to header, and then append the result of further
1277 * reading the stream.
1278 */
1279 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1280 stream->next_out = buffer;
1281 stream->avail_out = bufsiz;
1282
1283 do {
1284 status = git_inflate(stream, 0);
1285 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1286 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1287 return 0;
1288 stream->next_out = buffer;
1289 stream->avail_out = bufsiz;
1290 } while (status != Z_STREAM_END);
1291 return -1;
1292}
1293
1294static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1295{
1296 int bytes = strlen(buffer) + 1;
1297 unsigned char *buf = xmallocz(size);
1298 unsigned long n;
1299 int status = Z_OK;
1300
1301 n = stream->total_out - bytes;
1302 if (n > size)
1303 n = size;
1304 memcpy(buf, (char *) buffer + bytes, n);
1305 bytes = n;
1306 if (bytes <= size) {
1307 /*
1308 * The above condition must be (bytes <= size), not
1309 * (bytes < size). In other words, even though we
1310 * expect no more output and set avail_out to zero,
1311 * the input zlib stream may have bytes that express
1312 * "this concludes the stream", and we *do* want to
1313 * eat that input.
1314 *
1315 * Otherwise we would not be able to test that we
1316 * consumed all the input to reach the expected size;
1317 * we also want to check that zlib tells us that all
1318 * went well with status == Z_STREAM_END at the end.
1319 */
1320 stream->next_out = buf + bytes;
1321 stream->avail_out = size - bytes;
1322 while (status == Z_OK)
1323 status = git_inflate(stream, Z_FINISH);
1324 }
1325 if (status == Z_STREAM_END && !stream->avail_in) {
1326 git_inflate_end(stream);
1327 return buf;
1328 }
1329
1330 if (status < 0)
1331 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1332 else if (stream->avail_in)
1333 error("garbage at end of loose object '%s'",
1334 sha1_to_hex(sha1));
1335 free(buf);
1336 return NULL;
1337}
1338
1339/*
1340 * We used to just use "sscanf()", but that's actually way
1341 * too permissive for what we want to check. So do an anal
1342 * object header parse by hand.
1343 */
1344static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1345 unsigned int flags)
1346{
1347 const char *type_buf = hdr;
1348 unsigned long size;
1349 int type, type_len = 0;
1350
1351 /*
1352 * The type can be of any size but is followed by
1353 * a space.
1354 */
1355 for (;;) {
1356 char c = *hdr++;
1357 if (!c)
1358 return -1;
1359 if (c == ' ')
1360 break;
1361 type_len++;
1362 }
1363
1364 type = type_from_string_gently(type_buf, type_len, 1);
1365 if (oi->typename)
1366 strbuf_add(oi->typename, type_buf, type_len);
1367 /*
1368 * Set type to 0 if its an unknown object and
1369 * we're obtaining the type using '--allow-unknown-type'
1370 * option.
1371 */
1372 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1373 type = 0;
1374 else if (type < 0)
1375 die("invalid object type");
1376 if (oi->typep)
1377 *oi->typep = type;
1378
1379 /*
1380 * The length must follow immediately, and be in canonical
1381 * decimal format (ie "010" is not valid).
1382 */
1383 size = *hdr++ - '0';
1384 if (size > 9)
1385 return -1;
1386 if (size) {
1387 for (;;) {
1388 unsigned long c = *hdr - '0';
1389 if (c > 9)
1390 break;
1391 hdr++;
1392 size = size * 10 + c;
1393 }
1394 }
1395
1396 if (oi->sizep)
1397 *oi->sizep = size;
1398
1399 /*
1400 * The length must be followed by a zero byte
1401 */
1402 return *hdr ? -1 : type;
1403}
1404
1405int parse_sha1_header(const char *hdr, unsigned long *sizep)
1406{
1407 struct object_info oi = OBJECT_INFO_INIT;
1408
1409 oi.sizep = sizep;
1410 return parse_sha1_header_extended(hdr, &oi, 0);
1411}
1412
1413unsigned long get_size_from_delta(struct packed_git *p,
1414 struct pack_window **w_curs,
1415 off_t curpos)
1416{
1417 const unsigned char *data;
1418 unsigned char delta_head[20], *in;
1419 git_zstream stream;
1420 int st;
1421
1422 memset(&stream, 0, sizeof(stream));
1423 stream.next_out = delta_head;
1424 stream.avail_out = sizeof(delta_head);
1425
1426 git_inflate_init(&stream);
1427 do {
1428 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1429 stream.next_in = in;
1430 st = git_inflate(&stream, Z_FINISH);
1431 curpos += stream.next_in - in;
1432 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1433 stream.total_out < sizeof(delta_head));
1434 git_inflate_end(&stream);
1435 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1436 error("delta data unpack-initial failed");
1437 return 0;
1438 }
1439
1440 /* Examine the initial part of the delta to figure out
1441 * the result size.
1442 */
1443 data = delta_head;
1444
1445 /* ignore base size */
1446 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1447
1448 /* Read the result size */
1449 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1450}
1451
1452static off_t get_delta_base(struct packed_git *p,
1453 struct pack_window **w_curs,
1454 off_t *curpos,
1455 enum object_type type,
1456 off_t delta_obj_offset)
1457{
1458 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1459 off_t base_offset;
1460
1461 /* use_pack() assured us we have [base_info, base_info + 20)
1462 * as a range that we can look at without walking off the
1463 * end of the mapped window. Its actually the hash size
1464 * that is assured. An OFS_DELTA longer than the hash size
1465 * is stupid, as then a REF_DELTA would be smaller to store.
1466 */
1467 if (type == OBJ_OFS_DELTA) {
1468 unsigned used = 0;
1469 unsigned char c = base_info[used++];
1470 base_offset = c & 127;
1471 while (c & 128) {
1472 base_offset += 1;
1473 if (!base_offset || MSB(base_offset, 7))
1474 return 0; /* overflow */
1475 c = base_info[used++];
1476 base_offset = (base_offset << 7) + (c & 127);
1477 }
1478 base_offset = delta_obj_offset - base_offset;
1479 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1480 return 0; /* out of bound */
1481 *curpos += used;
1482 } else if (type == OBJ_REF_DELTA) {
1483 /* The base entry _must_ be in the same pack */
1484 base_offset = find_pack_entry_one(base_info, p);
1485 *curpos += 20;
1486 } else
1487 die("I am totally screwed");
1488 return base_offset;
1489}
1490
1491/*
1492 * Like get_delta_base above, but we return the sha1 instead of the pack
1493 * offset. This means it is cheaper for REF deltas (we do not have to do
1494 * the final object lookup), but more expensive for OFS deltas (we
1495 * have to load the revidx to convert the offset back into a sha1).
1496 */
1497static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1498 struct pack_window **w_curs,
1499 off_t curpos,
1500 enum object_type type,
1501 off_t delta_obj_offset)
1502{
1503 if (type == OBJ_REF_DELTA) {
1504 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1505 return base;
1506 } else if (type == OBJ_OFS_DELTA) {
1507 struct revindex_entry *revidx;
1508 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1509 type, delta_obj_offset);
1510
1511 if (!base_offset)
1512 return NULL;
1513
1514 revidx = find_pack_revindex(p, base_offset);
1515 if (!revidx)
1516 return NULL;
1517
1518 return nth_packed_object_sha1(p, revidx->nr);
1519 } else
1520 return NULL;
1521}
1522
1523int unpack_object_header(struct packed_git *p,
1524 struct pack_window **w_curs,
1525 off_t *curpos,
1526 unsigned long *sizep)
1527{
1528 unsigned char *base;
1529 unsigned long left;
1530 unsigned long used;
1531 enum object_type type;
1532
1533 /* use_pack() assures us we have [base, base + 20) available
1534 * as a range that we can look at. (Its actually the hash
1535 * size that is assured.) With our object header encoding
1536 * the maximum deflated object size is 2^137, which is just
1537 * insane, so we know won't exceed what we have been given.
1538 */
1539 base = use_pack(p, w_curs, *curpos, &left);
1540 used = unpack_object_header_buffer(base, left, &type, sizep);
1541 if (!used) {
1542 type = OBJ_BAD;
1543 } else
1544 *curpos += used;
1545
1546 return type;
1547}
1548
1549static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1550{
1551 int type;
1552 struct revindex_entry *revidx;
1553 const unsigned char *sha1;
1554 revidx = find_pack_revindex(p, obj_offset);
1555 if (!revidx)
1556 return OBJ_BAD;
1557 sha1 = nth_packed_object_sha1(p, revidx->nr);
1558 mark_bad_packed_object(p, sha1);
1559 type = sha1_object_info(sha1, NULL);
1560 if (type <= OBJ_NONE)
1561 return OBJ_BAD;
1562 return type;
1563}
1564
1565#define POI_STACK_PREALLOC 64
1566
1567static enum object_type packed_to_object_type(struct packed_git *p,
1568 off_t obj_offset,
1569 enum object_type type,
1570 struct pack_window **w_curs,
1571 off_t curpos)
1572{
1573 off_t small_poi_stack[POI_STACK_PREALLOC];
1574 off_t *poi_stack = small_poi_stack;
1575 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1576
1577 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1578 off_t base_offset;
1579 unsigned long size;
1580 /* Push the object we're going to leave behind */
1581 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1582 poi_stack_alloc = alloc_nr(poi_stack_nr);
1583 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1584 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1585 } else {
1586 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1587 }
1588 poi_stack[poi_stack_nr++] = obj_offset;
1589 /* If parsing the base offset fails, just unwind */
1590 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1591 if (!base_offset)
1592 goto unwind;
1593 curpos = obj_offset = base_offset;
1594 type = unpack_object_header(p, w_curs, &curpos, &size);
1595 if (type <= OBJ_NONE) {
1596 /* If getting the base itself fails, we first
1597 * retry the base, otherwise unwind */
1598 type = retry_bad_packed_offset(p, base_offset);
1599 if (type > OBJ_NONE)
1600 goto out;
1601 goto unwind;
1602 }
1603 }
1604
1605 switch (type) {
1606 case OBJ_BAD:
1607 case OBJ_COMMIT:
1608 case OBJ_TREE:
1609 case OBJ_BLOB:
1610 case OBJ_TAG:
1611 break;
1612 default:
1613 error("unknown object type %i at offset %"PRIuMAX" in %s",
1614 type, (uintmax_t)obj_offset, p->pack_name);
1615 type = OBJ_BAD;
1616 }
1617
1618out:
1619 if (poi_stack != small_poi_stack)
1620 free(poi_stack);
1621 return type;
1622
1623unwind:
1624 while (poi_stack_nr) {
1625 obj_offset = poi_stack[--poi_stack_nr];
1626 type = retry_bad_packed_offset(p, obj_offset);
1627 if (type > OBJ_NONE)
1628 goto out;
1629 }
1630 type = OBJ_BAD;
1631 goto out;
1632}
1633
1634static struct hashmap delta_base_cache;
1635static size_t delta_base_cached;
1636
1637static LIST_HEAD(delta_base_cache_lru);
1638
1639struct delta_base_cache_key {
1640 struct packed_git *p;
1641 off_t base_offset;
1642};
1643
1644struct delta_base_cache_entry {
1645 struct hashmap hash;
1646 struct delta_base_cache_key key;
1647 struct list_head lru;
1648 void *data;
1649 unsigned long size;
1650 enum object_type type;
1651};
1652
1653static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1654{
1655 unsigned int hash;
1656
1657 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1658 hash += (hash >> 8) + (hash >> 16);
1659 return hash;
1660}
1661
1662static struct delta_base_cache_entry *
1663get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1664{
1665 struct hashmap_entry entry;
1666 struct delta_base_cache_key key;
1667
1668 if (!delta_base_cache.cmpfn)
1669 return NULL;
1670
1671 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1672 key.p = p;
1673 key.base_offset = base_offset;
1674 return hashmap_get(&delta_base_cache, &entry, &key);
1675}
1676
1677static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1678 const struct delta_base_cache_key *b)
1679{
1680 return a->p == b->p && a->base_offset == b->base_offset;
1681}
1682
1683static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1684 const void *va, const void *vb,
1685 const void *vkey)
1686{
1687 const struct delta_base_cache_entry *a = va, *b = vb;
1688 const struct delta_base_cache_key *key = vkey;
1689 if (key)
1690 return !delta_base_cache_key_eq(&a->key, key);
1691 else
1692 return !delta_base_cache_key_eq(&a->key, &b->key);
1693}
1694
1695static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1696{
1697 return !!get_delta_base_cache_entry(p, base_offset);
1698}
1699
1700/*
1701 * Remove the entry from the cache, but do _not_ free the associated
1702 * entry data. The caller takes ownership of the "data" buffer, and
1703 * should copy out any fields it wants before detaching.
1704 */
1705static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1706{
1707 hashmap_remove(&delta_base_cache, ent, &ent->key);
1708 list_del(&ent->lru);
1709 delta_base_cached -= ent->size;
1710 free(ent);
1711}
1712
1713static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1714 unsigned long *base_size, enum object_type *type)
1715{
1716 struct delta_base_cache_entry *ent;
1717
1718 ent = get_delta_base_cache_entry(p, base_offset);
1719 if (!ent)
1720 return unpack_entry(p, base_offset, type, base_size);
1721
1722 if (type)
1723 *type = ent->type;
1724 if (base_size)
1725 *base_size = ent->size;
1726 return xmemdupz(ent->data, ent->size);
1727}
1728
1729static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1730{
1731 free(ent->data);
1732 detach_delta_base_cache_entry(ent);
1733}
1734
1735void clear_delta_base_cache(void)
1736{
1737 struct list_head *lru, *tmp;
1738 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1739 struct delta_base_cache_entry *entry =
1740 list_entry(lru, struct delta_base_cache_entry, lru);
1741 release_delta_base_cache(entry);
1742 }
1743}
1744
1745static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1746 void *base, unsigned long base_size, enum object_type type)
1747{
1748 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1749 struct list_head *lru, *tmp;
1750
1751 delta_base_cached += base_size;
1752
1753 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1754 struct delta_base_cache_entry *f =
1755 list_entry(lru, struct delta_base_cache_entry, lru);
1756 if (delta_base_cached <= delta_base_cache_limit)
1757 break;
1758 release_delta_base_cache(f);
1759 }
1760
1761 ent->key.p = p;
1762 ent->key.base_offset = base_offset;
1763 ent->type = type;
1764 ent->data = base;
1765 ent->size = base_size;
1766 list_add_tail(&ent->lru, &delta_base_cache_lru);
1767
1768 if (!delta_base_cache.cmpfn)
1769 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1770 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1771 hashmap_add(&delta_base_cache, ent);
1772}
1773
1774int packed_object_info(struct packed_git *p, off_t obj_offset,
1775 struct object_info *oi)
1776{
1777 struct pack_window *w_curs = NULL;
1778 unsigned long size;
1779 off_t curpos = obj_offset;
1780 enum object_type type;
1781
1782 /*
1783 * We always get the representation type, but only convert it to
1784 * a "real" type later if the caller is interested.
1785 */
1786 if (oi->contentp) {
1787 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1788 &type);
1789 if (!*oi->contentp)
1790 type = OBJ_BAD;
1791 } else {
1792 type = unpack_object_header(p, &w_curs, &curpos, &size);
1793 }
1794
1795 if (!oi->contentp && oi->sizep) {
1796 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1797 off_t tmp_pos = curpos;
1798 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1799 type, obj_offset);
1800 if (!base_offset) {
1801 type = OBJ_BAD;
1802 goto out;
1803 }
1804 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1805 if (*oi->sizep == 0) {
1806 type = OBJ_BAD;
1807 goto out;
1808 }
1809 } else {
1810 *oi->sizep = size;
1811 }
1812 }
1813
1814 if (oi->disk_sizep) {
1815 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1816 *oi->disk_sizep = revidx[1].offset - obj_offset;
1817 }
1818
1819 if (oi->typep || oi->typename) {
1820 enum object_type ptot;
1821 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1822 curpos);
1823 if (oi->typep)
1824 *oi->typep = ptot;
1825 if (oi->typename) {
1826 const char *tn = typename(ptot);
1827 if (tn)
1828 strbuf_addstr(oi->typename, tn);
1829 }
1830 if (ptot < 0) {
1831 type = OBJ_BAD;
1832 goto out;
1833 }
1834 }
1835
1836 if (oi->delta_base_sha1) {
1837 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1838 const unsigned char *base;
1839
1840 base = get_delta_base_sha1(p, &w_curs, curpos,
1841 type, obj_offset);
1842 if (!base) {
1843 type = OBJ_BAD;
1844 goto out;
1845 }
1846
1847 hashcpy(oi->delta_base_sha1, base);
1848 } else
1849 hashclr(oi->delta_base_sha1);
1850 }
1851
1852 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1853 OI_PACKED;
1854
1855out:
1856 unuse_pack(&w_curs);
1857 return type;
1858}
1859
1860static void *unpack_compressed_entry(struct packed_git *p,
1861 struct pack_window **w_curs,
1862 off_t curpos,
1863 unsigned long size)
1864{
1865 int st;
1866 git_zstream stream;
1867 unsigned char *buffer, *in;
1868
1869 buffer = xmallocz_gently(size);
1870 if (!buffer)
1871 return NULL;
1872 memset(&stream, 0, sizeof(stream));
1873 stream.next_out = buffer;
1874 stream.avail_out = size + 1;
1875
1876 git_inflate_init(&stream);
1877 do {
1878 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1879 stream.next_in = in;
1880 st = git_inflate(&stream, Z_FINISH);
1881 if (!stream.avail_out)
1882 break; /* the payload is larger than it should be */
1883 curpos += stream.next_in - in;
1884 } while (st == Z_OK || st == Z_BUF_ERROR);
1885 git_inflate_end(&stream);
1886 if ((st != Z_STREAM_END) || stream.total_out != size) {
1887 free(buffer);
1888 return NULL;
1889 }
1890
1891 return buffer;
1892}
1893
1894static void *read_object(const unsigned char *sha1, enum object_type *type,
1895 unsigned long *size);
1896
1897static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1898{
1899 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1900 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1901 p->pack_name, (uintmax_t)obj_offset);
1902}
1903
1904int do_check_packed_object_crc;
1905
1906#define UNPACK_ENTRY_STACK_PREALLOC 64
1907struct unpack_entry_stack_ent {
1908 off_t obj_offset;
1909 off_t curpos;
1910 unsigned long size;
1911};
1912
1913void *unpack_entry(struct packed_git *p, off_t obj_offset,
1914 enum object_type *final_type, unsigned long *final_size)
1915{
1916 struct pack_window *w_curs = NULL;
1917 off_t curpos = obj_offset;
1918 void *data = NULL;
1919 unsigned long size;
1920 enum object_type type;
1921 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1922 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1923 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1924 int base_from_cache = 0;
1925
1926 write_pack_access_log(p, obj_offset);
1927
1928 /* PHASE 1: drill down to the innermost base object */
1929 for (;;) {
1930 off_t base_offset;
1931 int i;
1932 struct delta_base_cache_entry *ent;
1933
1934 ent = get_delta_base_cache_entry(p, curpos);
1935 if (ent) {
1936 type = ent->type;
1937 data = ent->data;
1938 size = ent->size;
1939 detach_delta_base_cache_entry(ent);
1940 base_from_cache = 1;
1941 break;
1942 }
1943
1944 if (do_check_packed_object_crc && p->index_version > 1) {
1945 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1946 off_t len = revidx[1].offset - obj_offset;
1947 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1948 const unsigned char *sha1 =
1949 nth_packed_object_sha1(p, revidx->nr);
1950 error("bad packed object CRC for %s",
1951 sha1_to_hex(sha1));
1952 mark_bad_packed_object(p, sha1);
1953 data = NULL;
1954 goto out;
1955 }
1956 }
1957
1958 type = unpack_object_header(p, &w_curs, &curpos, &size);
1959 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1960 break;
1961
1962 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1963 if (!base_offset) {
1964 error("failed to validate delta base reference "
1965 "at offset %"PRIuMAX" from %s",
1966 (uintmax_t)curpos, p->pack_name);
1967 /* bail to phase 2, in hopes of recovery */
1968 data = NULL;
1969 break;
1970 }
1971
1972 /* push object, proceed to base */
1973 if (delta_stack_nr >= delta_stack_alloc
1974 && delta_stack == small_delta_stack) {
1975 delta_stack_alloc = alloc_nr(delta_stack_nr);
1976 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1977 memcpy(delta_stack, small_delta_stack,
1978 sizeof(*delta_stack)*delta_stack_nr);
1979 } else {
1980 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1981 }
1982 i = delta_stack_nr++;
1983 delta_stack[i].obj_offset = obj_offset;
1984 delta_stack[i].curpos = curpos;
1985 delta_stack[i].size = size;
1986
1987 curpos = obj_offset = base_offset;
1988 }
1989
1990 /* PHASE 2: handle the base */
1991 switch (type) {
1992 case OBJ_OFS_DELTA:
1993 case OBJ_REF_DELTA:
1994 if (data)
1995 die("BUG: unpack_entry: left loop at a valid delta");
1996 break;
1997 case OBJ_COMMIT:
1998 case OBJ_TREE:
1999 case OBJ_BLOB:
2000 case OBJ_TAG:
2001 if (!base_from_cache)
2002 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2003 break;
2004 default:
2005 data = NULL;
2006 error("unknown object type %i at offset %"PRIuMAX" in %s",
2007 type, (uintmax_t)obj_offset, p->pack_name);
2008 }
2009
2010 /* PHASE 3: apply deltas in order */
2011
2012 /* invariants:
2013 * 'data' holds the base data, or NULL if there was corruption
2014 */
2015 while (delta_stack_nr) {
2016 void *delta_data;
2017 void *base = data;
2018 void *external_base = NULL;
2019 unsigned long delta_size, base_size = size;
2020 int i;
2021
2022 data = NULL;
2023
2024 if (base)
2025 add_delta_base_cache(p, obj_offset, base, base_size, type);
2026
2027 if (!base) {
2028 /*
2029 * We're probably in deep shit, but let's try to fetch
2030 * the required base anyway from another pack or loose.
2031 * This is costly but should happen only in the presence
2032 * of a corrupted pack, and is better than failing outright.
2033 */
2034 struct revindex_entry *revidx;
2035 const unsigned char *base_sha1;
2036 revidx = find_pack_revindex(p, obj_offset);
2037 if (revidx) {
2038 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2039 error("failed to read delta base object %s"
2040 " at offset %"PRIuMAX" from %s",
2041 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2042 p->pack_name);
2043 mark_bad_packed_object(p, base_sha1);
2044 base = read_object(base_sha1, &type, &base_size);
2045 external_base = base;
2046 }
2047 }
2048
2049 i = --delta_stack_nr;
2050 obj_offset = delta_stack[i].obj_offset;
2051 curpos = delta_stack[i].curpos;
2052 delta_size = delta_stack[i].size;
2053
2054 if (!base)
2055 continue;
2056
2057 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2058
2059 if (!delta_data) {
2060 error("failed to unpack compressed delta "
2061 "at offset %"PRIuMAX" from %s",
2062 (uintmax_t)curpos, p->pack_name);
2063 data = NULL;
2064 free(external_base);
2065 continue;
2066 }
2067
2068 data = patch_delta(base, base_size,
2069 delta_data, delta_size,
2070 &size);
2071
2072 /*
2073 * We could not apply the delta; warn the user, but keep going.
2074 * Our failure will be noticed either in the next iteration of
2075 * the loop, or if this is the final delta, in the caller when
2076 * we return NULL. Those code paths will take care of making
2077 * a more explicit warning and retrying with another copy of
2078 * the object.
2079 */
2080 if (!data)
2081 error("failed to apply delta");
2082
2083 free(delta_data);
2084 free(external_base);
2085 }
2086
2087 if (final_type)
2088 *final_type = type;
2089 if (final_size)
2090 *final_size = size;
2091
2092out:
2093 unuse_pack(&w_curs);
2094
2095 if (delta_stack != small_delta_stack)
2096 free(delta_stack);
2097
2098 return data;
2099}
2100
2101const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2102 uint32_t n)
2103{
2104 const unsigned char *index = p->index_data;
2105 if (!index) {
2106 if (open_pack_index(p))
2107 return NULL;
2108 index = p->index_data;
2109 }
2110 if (n >= p->num_objects)
2111 return NULL;
2112 index += 4 * 256;
2113 if (p->index_version == 1) {
2114 return index + 24 * n + 4;
2115 } else {
2116 index += 8;
2117 return index + 20 * n;
2118 }
2119}
2120
2121const struct object_id *nth_packed_object_oid(struct object_id *oid,
2122 struct packed_git *p,
2123 uint32_t n)
2124{
2125 const unsigned char *hash = nth_packed_object_sha1(p, n);
2126 if (!hash)
2127 return NULL;
2128 hashcpy(oid->hash, hash);
2129 return oid;
2130}
2131
2132void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2133{
2134 const unsigned char *ptr = vptr;
2135 const unsigned char *start = p->index_data;
2136 const unsigned char *end = start + p->index_size;
2137 if (ptr < start)
2138 die(_("offset before start of pack index for %s (corrupt index?)"),
2139 p->pack_name);
2140 /* No need to check for underflow; .idx files must be at least 8 bytes */
2141 if (ptr >= end - 8)
2142 die(_("offset beyond end of pack index for %s (truncated index?)"),
2143 p->pack_name);
2144}
2145
2146off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2147{
2148 const unsigned char *index = p->index_data;
2149 index += 4 * 256;
2150 if (p->index_version == 1) {
2151 return ntohl(*((uint32_t *)(index + 24 * n)));
2152 } else {
2153 uint32_t off;
2154 index += 8 + p->num_objects * (20 + 4);
2155 off = ntohl(*((uint32_t *)(index + 4 * n)));
2156 if (!(off & 0x80000000))
2157 return off;
2158 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2159 check_pack_index_ptr(p, index);
2160 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2161 ntohl(*((uint32_t *)(index + 4)));
2162 }
2163}
2164
2165off_t find_pack_entry_one(const unsigned char *sha1,
2166 struct packed_git *p)
2167{
2168 const uint32_t *level1_ofs = p->index_data;
2169 const unsigned char *index = p->index_data;
2170 unsigned hi, lo, stride;
2171 static int debug_lookup = -1;
2172
2173 if (debug_lookup < 0)
2174 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2175
2176 if (!index) {
2177 if (open_pack_index(p))
2178 return 0;
2179 level1_ofs = p->index_data;
2180 index = p->index_data;
2181 }
2182 if (p->index_version > 1) {
2183 level1_ofs += 2;
2184 index += 8;
2185 }
2186 index += 4 * 256;
2187 hi = ntohl(level1_ofs[*sha1]);
2188 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2189 if (p->index_version > 1) {
2190 stride = 20;
2191 } else {
2192 stride = 24;
2193 index += 4;
2194 }
2195
2196 if (debug_lookup)
2197 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2198 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2199
2200 while (lo < hi) {
2201 unsigned mi = (lo + hi) / 2;
2202 int cmp = hashcmp(index + mi * stride, sha1);
2203
2204 if (debug_lookup)
2205 printf("lo %u hi %u rg %u mi %u\n",
2206 lo, hi, hi - lo, mi);
2207 if (!cmp)
2208 return nth_packed_object_offset(p, mi);
2209 if (cmp > 0)
2210 hi = mi;
2211 else
2212 lo = mi+1;
2213 }
2214 return 0;
2215}
2216
2217int is_pack_valid(struct packed_git *p)
2218{
2219 /* An already open pack is known to be valid. */
2220 if (p->pack_fd != -1)
2221 return 1;
2222
2223 /* If the pack has one window completely covering the
2224 * file size, the pack is known to be valid even if
2225 * the descriptor is not currently open.
2226 */
2227 if (p->windows) {
2228 struct pack_window *w = p->windows;
2229
2230 if (!w->offset && w->len == p->pack_size)
2231 return 1;
2232 }
2233
2234 /* Force the pack to open to prove its valid. */
2235 return !open_packed_git(p);
2236}
2237
2238static int fill_pack_entry(const unsigned char *sha1,
2239 struct pack_entry *e,
2240 struct packed_git *p)
2241{
2242 off_t offset;
2243
2244 if (p->num_bad_objects) {
2245 unsigned i;
2246 for (i = 0; i < p->num_bad_objects; i++)
2247 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2248 return 0;
2249 }
2250
2251 offset = find_pack_entry_one(sha1, p);
2252 if (!offset)
2253 return 0;
2254
2255 /*
2256 * We are about to tell the caller where they can locate the
2257 * requested object. We better make sure the packfile is
2258 * still here and can be accessed before supplying that
2259 * answer, as it may have been deleted since the index was
2260 * loaded!
2261 */
2262 if (!is_pack_valid(p))
2263 return 0;
2264 e->offset = offset;
2265 e->p = p;
2266 hashcpy(e->sha1, sha1);
2267 return 1;
2268}
2269
2270/*
2271 * Iff a pack file contains the object named by sha1, return true and
2272 * store its location to e.
2273 */
2274static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2275{
2276 struct mru_entry *p;
2277
2278 prepare_packed_git();
2279 if (!packed_git)
2280 return 0;
2281
2282 for (p = packed_git_mru->head; p; p = p->next) {
2283 if (fill_pack_entry(sha1, e, p->item)) {
2284 mru_mark(packed_git_mru, p);
2285 return 1;
2286 }
2287 }
2288 return 0;
2289}
2290
2291struct packed_git *find_sha1_pack(const unsigned char *sha1,
2292 struct packed_git *packs)
2293{
2294 struct packed_git *p;
2295
2296 for (p = packs; p; p = p->next) {
2297 if (find_pack_entry_one(sha1, p))
2298 return p;
2299 }
2300 return NULL;
2301
2302}
2303
2304static int sha1_loose_object_info(const unsigned char *sha1,
2305 struct object_info *oi,
2306 int flags)
2307{
2308 int status = 0;
2309 unsigned long mapsize;
2310 void *map;
2311 git_zstream stream;
2312 char hdr[32];
2313 struct strbuf hdrbuf = STRBUF_INIT;
2314 unsigned long size_scratch;
2315
2316 if (oi->delta_base_sha1)
2317 hashclr(oi->delta_base_sha1);
2318
2319 /*
2320 * If we don't care about type or size, then we don't
2321 * need to look inside the object at all. Note that we
2322 * do not optimize out the stat call, even if the
2323 * caller doesn't care about the disk-size, since our
2324 * return value implicitly indicates whether the
2325 * object even exists.
2326 */
2327 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2328 const char *path;
2329 struct stat st;
2330 if (stat_sha1_file(sha1, &st, &path) < 0)
2331 return -1;
2332 if (oi->disk_sizep)
2333 *oi->disk_sizep = st.st_size;
2334 return 0;
2335 }
2336
2337 map = map_sha1_file(sha1, &mapsize);
2338 if (!map)
2339 return -1;
2340
2341 if (!oi->sizep)
2342 oi->sizep = &size_scratch;
2343
2344 if (oi->disk_sizep)
2345 *oi->disk_sizep = mapsize;
2346 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2347 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2348 status = error("unable to unpack %s header with --allow-unknown-type",
2349 sha1_to_hex(sha1));
2350 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2351 status = error("unable to unpack %s header",
2352 sha1_to_hex(sha1));
2353 if (status < 0)
2354 ; /* Do nothing */
2355 else if (hdrbuf.len) {
2356 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2357 status = error("unable to parse %s header with --allow-unknown-type",
2358 sha1_to_hex(sha1));
2359 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2360 status = error("unable to parse %s header", sha1_to_hex(sha1));
2361
2362 if (status >= 0 && oi->contentp)
2363 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2364 *oi->sizep, sha1);
2365 else
2366 git_inflate_end(&stream);
2367
2368 munmap(map, mapsize);
2369 if (status && oi->typep)
2370 *oi->typep = status;
2371 if (oi->sizep == &size_scratch)
2372 oi->sizep = NULL;
2373 strbuf_release(&hdrbuf);
2374 oi->whence = OI_LOOSE;
2375 return (status < 0) ? status : 0;
2376}
2377
2378int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2379{
2380 static struct object_info blank_oi = OBJECT_INFO_INIT;
2381 struct pack_entry e;
2382 int rtype;
2383 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2384 lookup_replace_object(sha1) :
2385 sha1;
2386
2387 if (!oi)
2388 oi = &blank_oi;
2389
2390 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2391 struct cached_object *co = find_cached_object(real);
2392 if (co) {
2393 if (oi->typep)
2394 *(oi->typep) = co->type;
2395 if (oi->sizep)
2396 *(oi->sizep) = co->size;
2397 if (oi->disk_sizep)
2398 *(oi->disk_sizep) = 0;
2399 if (oi->delta_base_sha1)
2400 hashclr(oi->delta_base_sha1);
2401 if (oi->typename)
2402 strbuf_addstr(oi->typename, typename(co->type));
2403 if (oi->contentp)
2404 *oi->contentp = xmemdupz(co->buf, co->size);
2405 oi->whence = OI_CACHED;
2406 return 0;
2407 }
2408 }
2409
2410 if (!find_pack_entry(real, &e)) {
2411 /* Most likely it's a loose object. */
2412 if (!sha1_loose_object_info(real, oi, flags))
2413 return 0;
2414
2415 /* Not a loose object; someone else may have just packed it. */
2416 if (flags & OBJECT_INFO_QUICK) {
2417 return -1;
2418 } else {
2419 reprepare_packed_git();
2420 if (!find_pack_entry(real, &e))
2421 return -1;
2422 }
2423 }
2424
2425 if (oi == &blank_oi)
2426 /*
2427 * We know that the caller doesn't actually need the
2428 * information below, so return early.
2429 */
2430 return 0;
2431
2432 rtype = packed_object_info(e.p, e.offset, oi);
2433 if (rtype < 0) {
2434 mark_bad_packed_object(e.p, real);
2435 return sha1_object_info_extended(real, oi, 0);
2436 } else if (oi->whence == OI_PACKED) {
2437 oi->u.packed.offset = e.offset;
2438 oi->u.packed.pack = e.p;
2439 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2440 rtype == OBJ_OFS_DELTA);
2441 }
2442
2443 return 0;
2444}
2445
2446/* returns enum object_type or negative */
2447int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2448{
2449 enum object_type type;
2450 struct object_info oi = OBJECT_INFO_INIT;
2451
2452 oi.typep = &type;
2453 oi.sizep = sizep;
2454 if (sha1_object_info_extended(sha1, &oi,
2455 OBJECT_INFO_LOOKUP_REPLACE) < 0)
2456 return -1;
2457 return type;
2458}
2459
2460int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2461 unsigned char *sha1)
2462{
2463 struct cached_object *co;
2464
2465 hash_sha1_file(buf, len, typename(type), sha1);
2466 if (has_sha1_file(sha1) || find_cached_object(sha1))
2467 return 0;
2468 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2469 co = &cached_objects[cached_object_nr++];
2470 co->size = len;
2471 co->type = type;
2472 co->buf = xmalloc(len);
2473 memcpy(co->buf, buf, len);
2474 hashcpy(co->sha1, sha1);
2475 return 0;
2476}
2477
2478static void *read_object(const unsigned char *sha1, enum object_type *type,
2479 unsigned long *size)
2480{
2481 struct object_info oi = OBJECT_INFO_INIT;
2482 void *content;
2483 oi.typep = type;
2484 oi.sizep = size;
2485 oi.contentp = &content;
2486
2487 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
2488 return NULL;
2489 return content;
2490}
2491
2492/*
2493 * This function dies on corrupt objects; the callers who want to
2494 * deal with them should arrange to call read_object() and give error
2495 * messages themselves.
2496 */
2497void *read_sha1_file_extended(const unsigned char *sha1,
2498 enum object_type *type,
2499 unsigned long *size,
2500 int lookup_replace)
2501{
2502 void *data;
2503 const struct packed_git *p;
2504 const char *path;
2505 struct stat st;
2506 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
2507 : sha1;
2508
2509 errno = 0;
2510 data = read_object(repl, type, size);
2511 if (data)
2512 return data;
2513
2514 if (errno && errno != ENOENT)
2515 die_errno("failed to read object %s", sha1_to_hex(sha1));
2516
2517 /* die if we replaced an object with one that does not exist */
2518 if (repl != sha1)
2519 die("replacement %s not found for %s",
2520 sha1_to_hex(repl), sha1_to_hex(sha1));
2521
2522 if (!stat_sha1_file(repl, &st, &path))
2523 die("loose object %s (stored in %s) is corrupt",
2524 sha1_to_hex(repl), path);
2525
2526 if ((p = has_packed_and_bad(repl)) != NULL)
2527 die("packed object %s (stored in %s) is corrupt",
2528 sha1_to_hex(repl), p->pack_name);
2529
2530 return NULL;
2531}
2532
2533void *read_object_with_reference(const unsigned char *sha1,
2534 const char *required_type_name,
2535 unsigned long *size,
2536 unsigned char *actual_sha1_return)
2537{
2538 enum object_type type, required_type;
2539 void *buffer;
2540 unsigned long isize;
2541 unsigned char actual_sha1[20];
2542
2543 required_type = type_from_string(required_type_name);
2544 hashcpy(actual_sha1, sha1);
2545 while (1) {
2546 int ref_length = -1;
2547 const char *ref_type = NULL;
2548
2549 buffer = read_sha1_file(actual_sha1, &type, &isize);
2550 if (!buffer)
2551 return NULL;
2552 if (type == required_type) {
2553 *size = isize;
2554 if (actual_sha1_return)
2555 hashcpy(actual_sha1_return, actual_sha1);
2556 return buffer;
2557 }
2558 /* Handle references */
2559 else if (type == OBJ_COMMIT)
2560 ref_type = "tree ";
2561 else if (type == OBJ_TAG)
2562 ref_type = "object ";
2563 else {
2564 free(buffer);
2565 return NULL;
2566 }
2567 ref_length = strlen(ref_type);
2568
2569 if (ref_length + 40 > isize ||
2570 memcmp(buffer, ref_type, ref_length) ||
2571 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2572 free(buffer);
2573 return NULL;
2574 }
2575 free(buffer);
2576 /* Now we have the ID of the referred-to object in
2577 * actual_sha1. Check again. */
2578 }
2579}
2580
2581static void write_sha1_file_prepare(const void *buf, unsigned long len,
2582 const char *type, unsigned char *sha1,
2583 char *hdr, int *hdrlen)
2584{
2585 git_SHA_CTX c;
2586
2587 /* Generate the header */
2588 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2589
2590 /* Sha1.. */
2591 git_SHA1_Init(&c);
2592 git_SHA1_Update(&c, hdr, *hdrlen);
2593 git_SHA1_Update(&c, buf, len);
2594 git_SHA1_Final(sha1, &c);
2595}
2596
2597/*
2598 * Move the just written object into its final resting place.
2599 */
2600int finalize_object_file(const char *tmpfile, const char *filename)
2601{
2602 int ret = 0;
2603
2604 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2605 goto try_rename;
2606 else if (link(tmpfile, filename))
2607 ret = errno;
2608
2609 /*
2610 * Coda hack - coda doesn't like cross-directory links,
2611 * so we fall back to a rename, which will mean that it
2612 * won't be able to check collisions, but that's not a
2613 * big deal.
2614 *
2615 * The same holds for FAT formatted media.
2616 *
2617 * When this succeeds, we just return. We have nothing
2618 * left to unlink.
2619 */
2620 if (ret && ret != EEXIST) {
2621 try_rename:
2622 if (!rename(tmpfile, filename))
2623 goto out;
2624 ret = errno;
2625 }
2626 unlink_or_warn(tmpfile);
2627 if (ret) {
2628 if (ret != EEXIST) {
2629 return error_errno("unable to write sha1 filename %s", filename);
2630 }
2631 /* FIXME!!! Collision check here ? */
2632 }
2633
2634out:
2635 if (adjust_shared_perm(filename))
2636 return error("unable to set permission to '%s'", filename);
2637 return 0;
2638}
2639
2640static int write_buffer(int fd, const void *buf, size_t len)
2641{
2642 if (write_in_full(fd, buf, len) < 0)
2643 return error_errno("file write error");
2644 return 0;
2645}
2646
2647int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2648 unsigned char *sha1)
2649{
2650 char hdr[32];
2651 int hdrlen = sizeof(hdr);
2652 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2653 return 0;
2654}
2655
2656/* Finalize a file on disk, and close it. */
2657static void close_sha1_file(int fd)
2658{
2659 if (fsync_object_files)
2660 fsync_or_die(fd, "sha1 file");
2661 if (close(fd) != 0)
2662 die_errno("error when closing sha1 file");
2663}
2664
2665/* Size of directory component, including the ending '/' */
2666static inline int directory_size(const char *filename)
2667{
2668 const char *s = strrchr(filename, '/');
2669 if (!s)
2670 return 0;
2671 return s - filename + 1;
2672}
2673
2674/*
2675 * This creates a temporary file in the same directory as the final
2676 * 'filename'
2677 *
2678 * We want to avoid cross-directory filename renames, because those
2679 * can have problems on various filesystems (FAT, NFS, Coda).
2680 */
2681static int create_tmpfile(struct strbuf *tmp, const char *filename)
2682{
2683 int fd, dirlen = directory_size(filename);
2684
2685 strbuf_reset(tmp);
2686 strbuf_add(tmp, filename, dirlen);
2687 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2688 fd = git_mkstemp_mode(tmp->buf, 0444);
2689 if (fd < 0 && dirlen && errno == ENOENT) {
2690 /*
2691 * Make sure the directory exists; note that the contents
2692 * of the buffer are undefined after mkstemp returns an
2693 * error, so we have to rewrite the whole buffer from
2694 * scratch.
2695 */
2696 strbuf_reset(tmp);
2697 strbuf_add(tmp, filename, dirlen - 1);
2698 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
2699 return -1;
2700 if (adjust_shared_perm(tmp->buf))
2701 return -1;
2702
2703 /* Try again */
2704 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
2705 fd = git_mkstemp_mode(tmp->buf, 0444);
2706 }
2707 return fd;
2708}
2709
2710static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2711 const void *buf, unsigned long len, time_t mtime)
2712{
2713 int fd, ret;
2714 unsigned char compressed[4096];
2715 git_zstream stream;
2716 git_SHA_CTX c;
2717 unsigned char parano_sha1[20];
2718 static struct strbuf tmp_file = STRBUF_INIT;
2719 const char *filename = sha1_file_name(sha1);
2720
2721 fd = create_tmpfile(&tmp_file, filename);
2722 if (fd < 0) {
2723 if (errno == EACCES)
2724 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2725 else
2726 return error_errno("unable to create temporary file");
2727 }
2728
2729 /* Set it up */
2730 git_deflate_init(&stream, zlib_compression_level);
2731 stream.next_out = compressed;
2732 stream.avail_out = sizeof(compressed);
2733 git_SHA1_Init(&c);
2734
2735 /* First header.. */
2736 stream.next_in = (unsigned char *)hdr;
2737 stream.avail_in = hdrlen;
2738 while (git_deflate(&stream, 0) == Z_OK)
2739 ; /* nothing */
2740 git_SHA1_Update(&c, hdr, hdrlen);
2741
2742 /* Then the data itself.. */
2743 stream.next_in = (void *)buf;
2744 stream.avail_in = len;
2745 do {
2746 unsigned char *in0 = stream.next_in;
2747 ret = git_deflate(&stream, Z_FINISH);
2748 git_SHA1_Update(&c, in0, stream.next_in - in0);
2749 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2750 die("unable to write sha1 file");
2751 stream.next_out = compressed;
2752 stream.avail_out = sizeof(compressed);
2753 } while (ret == Z_OK);
2754
2755 if (ret != Z_STREAM_END)
2756 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2757 ret = git_deflate_end_gently(&stream);
2758 if (ret != Z_OK)
2759 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2760 git_SHA1_Final(parano_sha1, &c);
2761 if (hashcmp(sha1, parano_sha1) != 0)
2762 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2763
2764 close_sha1_file(fd);
2765
2766 if (mtime) {
2767 struct utimbuf utb;
2768 utb.actime = mtime;
2769 utb.modtime = mtime;
2770 if (utime(tmp_file.buf, &utb) < 0)
2771 warning_errno("failed utime() on %s", tmp_file.buf);
2772 }
2773
2774 return finalize_object_file(tmp_file.buf, filename);
2775}
2776
2777static int freshen_loose_object(const unsigned char *sha1)
2778{
2779 return check_and_freshen(sha1, 1);
2780}
2781
2782static int freshen_packed_object(const unsigned char *sha1)
2783{
2784 struct pack_entry e;
2785 if (!find_pack_entry(sha1, &e))
2786 return 0;
2787 if (e.p->freshened)
2788 return 1;
2789 if (!freshen_file(e.p->pack_name))
2790 return 0;
2791 e.p->freshened = 1;
2792 return 1;
2793}
2794
2795int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
2796{
2797 char hdr[32];
2798 int hdrlen = sizeof(hdr);
2799
2800 /* Normally if we have it in the pack then we do not bother writing
2801 * it out into .git/objects/??/?{38} file.
2802 */
2803 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2804 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2805 return 0;
2806 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2807}
2808
2809int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
2810 unsigned char *sha1, unsigned flags)
2811{
2812 char *header;
2813 int hdrlen, status = 0;
2814
2815 /* type string, SP, %lu of the length plus NUL must fit this */
2816 hdrlen = strlen(type) + 32;
2817 header = xmalloc(hdrlen);
2818 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
2819
2820 if (!(flags & HASH_WRITE_OBJECT))
2821 goto cleanup;
2822 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2823 goto cleanup;
2824 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
2825
2826cleanup:
2827 free(header);
2828 return status;
2829}
2830
2831int force_object_loose(const unsigned char *sha1, time_t mtime)
2832{
2833 void *buf;
2834 unsigned long len;
2835 enum object_type type;
2836 char hdr[32];
2837 int hdrlen;
2838 int ret;
2839
2840 if (has_loose_object(sha1))
2841 return 0;
2842 buf = read_object(sha1, &type, &len);
2843 if (!buf)
2844 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2845 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
2846 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2847 free(buf);
2848
2849 return ret;
2850}
2851
2852int has_pack_index(const unsigned char *sha1)
2853{
2854 struct stat st;
2855 if (stat(sha1_pack_index_name(sha1), &st))
2856 return 0;
2857 return 1;
2858}
2859
2860int has_sha1_pack(const unsigned char *sha1)
2861{
2862 struct pack_entry e;
2863 return find_pack_entry(sha1, &e);
2864}
2865
2866int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
2867{
2868 if (!startup_info->have_repository)
2869 return 0;
2870 return sha1_object_info_extended(sha1, NULL,
2871 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
2872}
2873
2874int has_object_file(const struct object_id *oid)
2875{
2876 return has_sha1_file(oid->hash);
2877}
2878
2879int has_object_file_with_flags(const struct object_id *oid, int flags)
2880{
2881 return has_sha1_file_with_flags(oid->hash, flags);
2882}
2883
2884static void check_tree(const void *buf, size_t size)
2885{
2886 struct tree_desc desc;
2887 struct name_entry entry;
2888
2889 init_tree_desc(&desc, buf, size);
2890 while (tree_entry(&desc, &entry))
2891 /* do nothing
2892 * tree_entry() will die() on malformed entries */
2893 ;
2894}
2895
2896static void check_commit(const void *buf, size_t size)
2897{
2898 struct commit c;
2899 memset(&c, 0, sizeof(c));
2900 if (parse_commit_buffer(&c, buf, size))
2901 die("corrupt commit");
2902}
2903
2904static void check_tag(const void *buf, size_t size)
2905{
2906 struct tag t;
2907 memset(&t, 0, sizeof(t));
2908 if (parse_tag_buffer(&t, buf, size))
2909 die("corrupt tag");
2910}
2911
2912static int index_mem(unsigned char *sha1, void *buf, size_t size,
2913 enum object_type type,
2914 const char *path, unsigned flags)
2915{
2916 int ret, re_allocated = 0;
2917 int write_object = flags & HASH_WRITE_OBJECT;
2918
2919 if (!type)
2920 type = OBJ_BLOB;
2921
2922 /*
2923 * Convert blobs to git internal format
2924 */
2925 if ((type == OBJ_BLOB) && path) {
2926 struct strbuf nbuf = STRBUF_INIT;
2927 if (convert_to_git(&the_index, path, buf, size, &nbuf,
2928 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2929 buf = strbuf_detach(&nbuf, &size);
2930 re_allocated = 1;
2931 }
2932 }
2933 if (flags & HASH_FORMAT_CHECK) {
2934 if (type == OBJ_TREE)
2935 check_tree(buf, size);
2936 if (type == OBJ_COMMIT)
2937 check_commit(buf, size);
2938 if (type == OBJ_TAG)
2939 check_tag(buf, size);
2940 }
2941
2942 if (write_object)
2943 ret = write_sha1_file(buf, size, typename(type), sha1);
2944 else
2945 ret = hash_sha1_file(buf, size, typename(type), sha1);
2946 if (re_allocated)
2947 free(buf);
2948 return ret;
2949}
2950
2951static int index_stream_convert_blob(unsigned char *sha1, int fd,
2952 const char *path, unsigned flags)
2953{
2954 int ret;
2955 const int write_object = flags & HASH_WRITE_OBJECT;
2956 struct strbuf sbuf = STRBUF_INIT;
2957
2958 assert(path);
2959 assert(would_convert_to_git_filter_fd(path));
2960
2961 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
2962 write_object ? safe_crlf : SAFE_CRLF_FALSE);
2963
2964 if (write_object)
2965 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2966 sha1);
2967 else
2968 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2969 sha1);
2970 strbuf_release(&sbuf);
2971 return ret;
2972}
2973
2974static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2975 const char *path, unsigned flags)
2976{
2977 struct strbuf sbuf = STRBUF_INIT;
2978 int ret;
2979
2980 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2981 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2982 else
2983 ret = -1;
2984 strbuf_release(&sbuf);
2985 return ret;
2986}
2987
2988#define SMALL_FILE_SIZE (32*1024)
2989
2990static int index_core(unsigned char *sha1, int fd, size_t size,
2991 enum object_type type, const char *path,
2992 unsigned flags)
2993{
2994 int ret;
2995
2996 if (!size) {
2997 ret = index_mem(sha1, "", size, type, path, flags);
2998 } else if (size <= SMALL_FILE_SIZE) {
2999 char *buf = xmalloc(size);
3000 if (size == read_in_full(fd, buf, size))
3001 ret = index_mem(sha1, buf, size, type, path, flags);
3002 else
3003 ret = error_errno("short read");
3004 free(buf);
3005 } else {
3006 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3007 ret = index_mem(sha1, buf, size, type, path, flags);
3008 munmap(buf, size);
3009 }
3010 return ret;
3011}
3012
3013/*
3014 * This creates one packfile per large blob unless bulk-checkin
3015 * machinery is "plugged".
3016 *
3017 * This also bypasses the usual "convert-to-git" dance, and that is on
3018 * purpose. We could write a streaming version of the converting
3019 * functions and insert that before feeding the data to fast-import
3020 * (or equivalent in-core API described above). However, that is
3021 * somewhat complicated, as we do not know the size of the filter
3022 * result, which we need to know beforehand when writing a git object.
3023 * Since the primary motivation for trying to stream from the working
3024 * tree file and to avoid mmaping it in core is to deal with large
3025 * binary blobs, they generally do not want to get any conversion, and
3026 * callers should avoid this code path when filters are requested.
3027 */
3028static int index_stream(unsigned char *sha1, int fd, size_t size,
3029 enum object_type type, const char *path,
3030 unsigned flags)
3031{
3032 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3033}
3034
3035int index_fd(unsigned char *sha1, int fd, struct stat *st,
3036 enum object_type type, const char *path, unsigned flags)
3037{
3038 int ret;
3039
3040 /*
3041 * Call xsize_t() only when needed to avoid potentially unnecessary
3042 * die() for large files.
3043 */
3044 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3045 ret = index_stream_convert_blob(sha1, fd, path, flags);
3046 else if (!S_ISREG(st->st_mode))
3047 ret = index_pipe(sha1, fd, type, path, flags);
3048 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3049 (path && would_convert_to_git(&the_index, path)))
3050 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3051 flags);
3052 else
3053 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3054 flags);
3055 close(fd);
3056 return ret;
3057}
3058
3059int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3060{
3061 int fd;
3062 struct strbuf sb = STRBUF_INIT;
3063
3064 switch (st->st_mode & S_IFMT) {
3065 case S_IFREG:
3066 fd = open(path, O_RDONLY);
3067 if (fd < 0)
3068 return error_errno("open(\"%s\")", path);
3069 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3070 return error("%s: failed to insert into database",
3071 path);
3072 break;
3073 case S_IFLNK:
3074 if (strbuf_readlink(&sb, path, st->st_size))
3075 return error_errno("readlink(\"%s\")", path);
3076 if (!(flags & HASH_WRITE_OBJECT))
3077 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3078 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3079 return error("%s: failed to insert into database",
3080 path);
3081 strbuf_release(&sb);
3082 break;
3083 case S_IFDIR:
3084 return resolve_gitlink_ref(path, "HEAD", sha1);
3085 default:
3086 return error("%s: unsupported file type", path);
3087 }
3088 return 0;
3089}
3090
3091int read_pack_header(int fd, struct pack_header *header)
3092{
3093 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3094 /* "eof before pack header was fully read" */
3095 return PH_ERROR_EOF;
3096
3097 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3098 /* "protocol error (pack signature mismatch detected)" */
3099 return PH_ERROR_PACK_SIGNATURE;
3100 if (!pack_version_ok(header->hdr_version))
3101 /* "protocol error (pack version unsupported)" */
3102 return PH_ERROR_PROTOCOL;
3103 return 0;
3104}
3105
3106void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3107{
3108 enum object_type type = sha1_object_info(sha1, NULL);
3109 if (type < 0)
3110 die("%s is not a valid object", sha1_to_hex(sha1));
3111 if (type != expect)
3112 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3113 typename(expect));
3114}
3115
3116int for_each_file_in_obj_subdir(unsigned int subdir_nr,
3117 struct strbuf *path,
3118 each_loose_object_fn obj_cb,
3119 each_loose_cruft_fn cruft_cb,
3120 each_loose_subdir_fn subdir_cb,
3121 void *data)
3122{
3123 size_t origlen, baselen;
3124 DIR *dir;
3125 struct dirent *de;
3126 int r = 0;
3127
3128 if (subdir_nr > 0xff)
3129 BUG("invalid loose object subdirectory: %x", subdir_nr);
3130
3131 origlen = path->len;
3132 strbuf_complete(path, '/');
3133 strbuf_addf(path, "%02x", subdir_nr);
3134 baselen = path->len;
3135
3136 dir = opendir(path->buf);
3137 if (!dir) {
3138 if (errno != ENOENT)
3139 r = error_errno("unable to open %s", path->buf);
3140 strbuf_setlen(path, origlen);
3141 return r;
3142 }
3143
3144 while ((de = readdir(dir))) {
3145 if (is_dot_or_dotdot(de->d_name))
3146 continue;
3147
3148 strbuf_setlen(path, baselen);
3149 strbuf_addf(path, "/%s", de->d_name);
3150
3151 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3152 char hex[GIT_MAX_HEXSZ+1];
3153 struct object_id oid;
3154
3155 xsnprintf(hex, sizeof(hex), "%02x%s",
3156 subdir_nr, de->d_name);
3157 if (!get_oid_hex(hex, &oid)) {
3158 if (obj_cb) {
3159 r = obj_cb(&oid, path->buf, data);
3160 if (r)
3161 break;
3162 }
3163 continue;
3164 }
3165 }
3166
3167 if (cruft_cb) {
3168 r = cruft_cb(de->d_name, path->buf, data);
3169 if (r)
3170 break;
3171 }
3172 }
3173 closedir(dir);
3174
3175 strbuf_setlen(path, baselen);
3176 if (!r && subdir_cb)
3177 r = subdir_cb(subdir_nr, path->buf, data);
3178
3179 strbuf_setlen(path, origlen);
3180
3181 return r;
3182}
3183
3184int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3185 each_loose_object_fn obj_cb,
3186 each_loose_cruft_fn cruft_cb,
3187 each_loose_subdir_fn subdir_cb,
3188 void *data)
3189{
3190 int r = 0;
3191 int i;
3192
3193 for (i = 0; i < 256; i++) {
3194 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3195 subdir_cb, data);
3196 if (r)
3197 break;
3198 }
3199
3200 return r;
3201}
3202
3203int for_each_loose_file_in_objdir(const char *path,
3204 each_loose_object_fn obj_cb,
3205 each_loose_cruft_fn cruft_cb,
3206 each_loose_subdir_fn subdir_cb,
3207 void *data)
3208{
3209 struct strbuf buf = STRBUF_INIT;
3210 int r;
3211
3212 strbuf_addstr(&buf, path);
3213 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3214 subdir_cb, data);
3215 strbuf_release(&buf);
3216
3217 return r;
3218}
3219
3220struct loose_alt_odb_data {
3221 each_loose_object_fn *cb;
3222 void *data;
3223};
3224
3225static int loose_from_alt_odb(struct alternate_object_database *alt,
3226 void *vdata)
3227{
3228 struct loose_alt_odb_data *data = vdata;
3229 struct strbuf buf = STRBUF_INIT;
3230 int r;
3231
3232 strbuf_addstr(&buf, alt->path);
3233 r = for_each_loose_file_in_objdir_buf(&buf,
3234 data->cb, NULL, NULL,
3235 data->data);
3236 strbuf_release(&buf);
3237 return r;
3238}
3239
3240int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3241{
3242 struct loose_alt_odb_data alt;
3243 int r;
3244
3245 r = for_each_loose_file_in_objdir(get_object_directory(),
3246 cb, NULL, NULL, data);
3247 if (r)
3248 return r;
3249
3250 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3251 return 0;
3252
3253 alt.cb = cb;
3254 alt.data = data;
3255 return foreach_alt_odb(loose_from_alt_odb, &alt);
3256}
3257
3258static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3259{
3260 uint32_t i;
3261 int r = 0;
3262
3263 for (i = 0; i < p->num_objects; i++) {
3264 struct object_id oid;
3265
3266 if (!nth_packed_object_oid(&oid, p, i))
3267 return error("unable to get sha1 of object %u in %s",
3268 i, p->pack_name);
3269
3270 r = cb(&oid, p, i, data);
3271 if (r)
3272 break;
3273 }
3274 return r;
3275}
3276
3277int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3278{
3279 struct packed_git *p;
3280 int r = 0;
3281 int pack_errors = 0;
3282
3283 prepare_packed_git();
3284 for (p = packed_git; p; p = p->next) {
3285 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3286 continue;
3287 if (open_pack_index(p)) {
3288 pack_errors = 1;
3289 continue;
3290 }
3291 r = for_each_object_in_pack(p, cb, data);
3292 if (r)
3293 break;
3294 }
3295 return r ? r : pack_errors;
3296}
3297
3298static int check_stream_sha1(git_zstream *stream,
3299 const char *hdr,
3300 unsigned long size,
3301 const char *path,
3302 const unsigned char *expected_sha1)
3303{
3304 git_SHA_CTX c;
3305 unsigned char real_sha1[GIT_MAX_RAWSZ];
3306 unsigned char buf[4096];
3307 unsigned long total_read;
3308 int status = Z_OK;
3309
3310 git_SHA1_Init(&c);
3311 git_SHA1_Update(&c, hdr, stream->total_out);
3312
3313 /*
3314 * We already read some bytes into hdr, but the ones up to the NUL
3315 * do not count against the object's content size.
3316 */
3317 total_read = stream->total_out - strlen(hdr) - 1;
3318
3319 /*
3320 * This size comparison must be "<=" to read the final zlib packets;
3321 * see the comment in unpack_sha1_rest for details.
3322 */
3323 while (total_read <= size &&
3324 (status == Z_OK || status == Z_BUF_ERROR)) {
3325 stream->next_out = buf;
3326 stream->avail_out = sizeof(buf);
3327 if (size - total_read < stream->avail_out)
3328 stream->avail_out = size - total_read;
3329 status = git_inflate(stream, Z_FINISH);
3330 git_SHA1_Update(&c, buf, stream->next_out - buf);
3331 total_read += stream->next_out - buf;
3332 }
3333 git_inflate_end(stream);
3334
3335 if (status != Z_STREAM_END) {
3336 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3337 return -1;
3338 }
3339 if (stream->avail_in) {
3340 error("garbage at end of loose object '%s'",
3341 sha1_to_hex(expected_sha1));
3342 return -1;
3343 }
3344
3345 git_SHA1_Final(real_sha1, &c);
3346 if (hashcmp(expected_sha1, real_sha1)) {
3347 error("sha1 mismatch for %s (expected %s)", path,
3348 sha1_to_hex(expected_sha1));
3349 return -1;
3350 }
3351
3352 return 0;
3353}
3354
3355int read_loose_object(const char *path,
3356 const unsigned char *expected_sha1,
3357 enum object_type *type,
3358 unsigned long *size,
3359 void **contents)
3360{
3361 int ret = -1;
3362 void *map = NULL;
3363 unsigned long mapsize;
3364 git_zstream stream;
3365 char hdr[32];
3366
3367 *contents = NULL;
3368
3369 map = map_sha1_file_1(path, NULL, &mapsize);
3370 if (!map) {
3371 error_errno("unable to mmap %s", path);
3372 goto out;
3373 }
3374
3375 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3376 error("unable to unpack header of %s", path);
3377 goto out;
3378 }
3379
3380 *type = parse_sha1_header(hdr, size);
3381 if (*type < 0) {
3382 error("unable to parse header of %s", path);
3383 git_inflate_end(&stream);
3384 goto out;
3385 }
3386
3387 if (*type == OBJ_BLOB) {
3388 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3389 goto out;
3390 } else {
3391 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3392 if (!*contents) {
3393 error("unable to unpack contents of %s", path);
3394 git_inflate_end(&stream);
3395 goto out;
3396 }
3397 if (check_sha1_signature(expected_sha1, *contents,
3398 *size, typename(*type))) {
3399 error("sha1 mismatch for %s (expected %s)", path,
3400 sha1_to_hex(expected_sha1));
3401 free(*contents);
3402 goto out;
3403 }
3404 }
3405
3406 ret = 0; /* everything checks out */
3407
3408out:
3409 if (map)
3410 munmap(map, mapsize);
3411 return ret;
3412}