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 "string-list.h"
11#include "lockfile.h"
12#include "delta.h"
13#include "pack.h"
14#include "blob.h"
15#include "commit.h"
16#include "run-command.h"
17#include "tag.h"
18#include "tree.h"
19#include "tree-walk.h"
20#include "refs.h"
21#include "pack-revindex.h"
22#include "sha1-lookup.h"
23#include "bulk-checkin.h"
24#include "streaming.h"
25#include "dir.h"
26#include "mru.h"
27#include "list.h"
28#include "mergesort.h"
29#include "quote.h"
30
31#define SZ_FMT PRIuMAX
32static inline uintmax_t sz_fmt(size_t s) { return s; }
33
34const unsigned char null_sha1[20];
35const struct object_id null_oid;
36const struct object_id empty_tree_oid = {
37 EMPTY_TREE_SHA1_BIN_LITERAL
38};
39const struct object_id empty_blob_oid = {
40 EMPTY_BLOB_SHA1_BIN_LITERAL
41};
42
43/*
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
48 */
49static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54} *cached_objects;
55static int cached_object_nr, cached_object_alloc;
56
57static struct cached_object empty_tree = {
58 EMPTY_TREE_SHA1_BIN_LITERAL,
59 OBJ_TREE,
60 "",
61 0
62};
63
64static struct cached_object *find_cached_object(const unsigned char *sha1)
65{
66 int i;
67 struct cached_object *co = cached_objects;
68
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
72 }
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
76}
77
78int mkdir_in_gitdir(const char *path)
79{
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
84
85 if (errno != EEXIST)
86 return -1;
87 /*
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
93 */
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
101 }
102 strbuf_release(&sb);
103 }
104 return adjust_shared_perm(path);
105}
106
107enum scld_error safe_create_leading_directories(char *path)
108{
109 char *next_component = path + offset_1st_component(path);
110 enum scld_error ret = SCLD_OK;
111
112 while (ret == SCLD_OK && next_component) {
113 struct stat st;
114 char *slash = next_component, slash_character;
115
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
118
119 if (!*slash)
120 break;
121
122 next_component = slash + 1;
123 while (is_dir_sep(*next_component))
124 next_component++;
125 if (!*next_component)
126 break;
127
128 slash_character = *slash;
129 *slash = '\0';
130 if (!stat(path, &st)) {
131 /* path exists */
132 if (!S_ISDIR(st.st_mode)) {
133 errno = ENOTDIR;
134 ret = SCLD_EXISTS;
135 }
136 } else if (mkdir(path, 0777)) {
137 if (errno == EEXIST &&
138 !stat(path, &st) && S_ISDIR(st.st_mode))
139 ; /* somebody created it since we checked */
140 else if (errno == ENOENT)
141 /*
142 * Either mkdir() failed because
143 * somebody just pruned the containing
144 * directory, or stat() failed because
145 * the file that was in our way was
146 * just removed. Either way, inform
147 * the caller that it might be worth
148 * trying again:
149 */
150 ret = SCLD_VANISHED;
151 else
152 ret = SCLD_FAILED;
153 } else if (adjust_shared_perm(path)) {
154 ret = SCLD_PERMS;
155 }
156 *slash = slash_character;
157 }
158 return ret;
159}
160
161enum scld_error safe_create_leading_directories_const(const char *path)
162{
163 int save_errno;
164 /* path points to cache entries, so xstrdup before messing with it */
165 char *buf = xstrdup(path);
166 enum scld_error result = safe_create_leading_directories(buf);
167
168 save_errno = errno;
169 free(buf);
170 errno = save_errno;
171 return result;
172}
173
174int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
175{
176 /*
177 * The number of times we will try to remove empty directories
178 * in the way of path. This is only 1 because if another
179 * process is racily creating directories that conflict with
180 * us, we don't want to fight against them.
181 */
182 int remove_directories_remaining = 1;
183
184 /*
185 * The number of times that we will try to create the
186 * directories containing path. We are willing to attempt this
187 * more than once, because another process could be trying to
188 * clean up empty directories at the same time as we are
189 * trying to create them.
190 */
191 int create_directories_remaining = 3;
192
193 /* A scratch copy of path, filled lazily if we need it: */
194 struct strbuf path_copy = STRBUF_INIT;
195
196 int ret, save_errno;
197
198 /* Sanity check: */
199 assert(*path);
200
201retry_fn:
202 ret = fn(path, cb);
203 save_errno = errno;
204 if (!ret)
205 goto out;
206
207 if (errno == EISDIR && remove_directories_remaining-- > 0) {
208 /*
209 * A directory is in the way. Maybe it is empty; try
210 * to remove it:
211 */
212 if (!path_copy.len)
213 strbuf_addstr(&path_copy, path);
214
215 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
216 goto retry_fn;
217 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
218 /*
219 * Maybe the containing directory didn't exist, or
220 * maybe it was just deleted by a process that is
221 * racing with us to clean up empty directories. Try
222 * to create it:
223 */
224 enum scld_error scld_result;
225
226 if (!path_copy.len)
227 strbuf_addstr(&path_copy, path);
228
229 do {
230 scld_result = safe_create_leading_directories(path_copy.buf);
231 if (scld_result == SCLD_OK)
232 goto retry_fn;
233 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
234 }
235
236out:
237 strbuf_release(&path_copy);
238 errno = save_errno;
239 return ret;
240}
241
242static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
243{
244 int i;
245 for (i = 0; i < 20; i++) {
246 static char hex[] = "0123456789abcdef";
247 unsigned int val = sha1[i];
248 strbuf_addch(buf, hex[val >> 4]);
249 strbuf_addch(buf, hex[val & 0xf]);
250 if (!i)
251 strbuf_addch(buf, '/');
252 }
253}
254
255const char *sha1_file_name(const unsigned char *sha1)
256{
257 static struct strbuf buf = STRBUF_INIT;
258
259 strbuf_reset(&buf);
260 strbuf_addf(&buf, "%s/", get_object_directory());
261
262 fill_sha1_path(&buf, sha1);
263 return buf.buf;
264}
265
266struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
267{
268 strbuf_setlen(&alt->scratch, alt->base_len);
269 return &alt->scratch;
270}
271
272static const char *alt_sha1_path(struct alternate_object_database *alt,
273 const unsigned char *sha1)
274{
275 struct strbuf *buf = alt_scratch_buf(alt);
276 fill_sha1_path(buf, sha1);
277 return buf->buf;
278}
279
280 char *odb_pack_name(struct strbuf *buf,
281 const unsigned char *sha1,
282 const char *ext)
283{
284 strbuf_reset(buf);
285 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
286 sha1_to_hex(sha1), ext);
287 return buf->buf;
288}
289
290char *sha1_pack_name(const unsigned char *sha1)
291{
292 static struct strbuf buf = STRBUF_INIT;
293 return odb_pack_name(&buf, sha1, "pack");
294}
295
296char *sha1_pack_index_name(const unsigned char *sha1)
297{
298 static struct strbuf buf = STRBUF_INIT;
299 return odb_pack_name(&buf, sha1, "idx");
300}
301
302struct alternate_object_database *alt_odb_list;
303static struct alternate_object_database **alt_odb_tail;
304
305/*
306 * Return non-zero iff the path is usable as an alternate object database.
307 */
308static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
309{
310 struct alternate_object_database *alt;
311
312 /* Detect cases where alternate disappeared */
313 if (!is_directory(path->buf)) {
314 error("object directory %s does not exist; "
315 "check .git/objects/info/alternates.",
316 path->buf);
317 return 0;
318 }
319
320 /*
321 * Prevent the common mistake of listing the same
322 * thing twice, or object directory itself.
323 */
324 for (alt = alt_odb_list; alt; alt = alt->next) {
325 if (!fspathcmp(path->buf, alt->path))
326 return 0;
327 }
328 if (!fspathcmp(path->buf, normalized_objdir))
329 return 0;
330
331 return 1;
332}
333
334/*
335 * Prepare alternate object database registry.
336 *
337 * The variable alt_odb_list points at the list of struct
338 * alternate_object_database. The elements on this list come from
339 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
340 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
341 * whose contents is similar to that environment variable but can be
342 * LF separated. Its base points at a statically allocated buffer that
343 * contains "/the/directory/corresponding/to/.git/objects/...", while
344 * its name points just after the slash at the end of ".git/objects/"
345 * in the example above, and has enough space to hold 40-byte hex
346 * SHA1, an extra slash for the first level indirection, and the
347 * terminating NUL.
348 */
349static int link_alt_odb_entry(const char *entry, const char *relative_base,
350 int depth, const char *normalized_objdir)
351{
352 struct alternate_object_database *ent;
353 struct strbuf pathbuf = STRBUF_INIT;
354
355 if (!is_absolute_path(entry) && relative_base) {
356 strbuf_realpath(&pathbuf, relative_base, 1);
357 strbuf_addch(&pathbuf, '/');
358 }
359 strbuf_addstr(&pathbuf, entry);
360
361 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
362 error("unable to normalize alternate object path: %s",
363 pathbuf.buf);
364 strbuf_release(&pathbuf);
365 return -1;
366 }
367
368 /*
369 * The trailing slash after the directory name is given by
370 * this function at the end. Remove duplicates.
371 */
372 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
373 strbuf_setlen(&pathbuf, pathbuf.len - 1);
374
375 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
376 strbuf_release(&pathbuf);
377 return -1;
378 }
379
380 ent = alloc_alt_odb(pathbuf.buf);
381
382 /* add the alternate entry */
383 *alt_odb_tail = ent;
384 alt_odb_tail = &(ent->next);
385 ent->next = NULL;
386
387 /* recursively add alternates */
388 read_info_alternates(pathbuf.buf, depth + 1);
389
390 strbuf_release(&pathbuf);
391 return 0;
392}
393
394static const char *parse_alt_odb_entry(const char *string,
395 int sep,
396 struct strbuf *out)
397{
398 const char *end;
399
400 strbuf_reset(out);
401
402 if (*string == '#') {
403 /* comment; consume up to next separator */
404 end = strchrnul(string, sep);
405 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
406 /*
407 * quoted path; unquote_c_style has copied the
408 * data for us and set "end". Broken quoting (e.g.,
409 * an entry that doesn't end with a quote) falls
410 * back to the unquoted case below.
411 */
412 } else {
413 /* normal, unquoted path */
414 end = strchrnul(string, sep);
415 strbuf_add(out, string, end - string);
416 }
417
418 if (*end)
419 end++;
420 return end;
421}
422
423static void link_alt_odb_entries(const char *alt, int len, int sep,
424 const char *relative_base, int depth)
425{
426 struct strbuf objdirbuf = STRBUF_INIT;
427 struct strbuf entry = STRBUF_INIT;
428
429 if (depth > 5) {
430 error("%s: ignoring alternate object stores, nesting too deep.",
431 relative_base);
432 return;
433 }
434
435 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
436 if (strbuf_normalize_path(&objdirbuf) < 0)
437 die("unable to normalize object directory: %s",
438 objdirbuf.buf);
439
440 while (*alt) {
441 alt = parse_alt_odb_entry(alt, sep, &entry);
442 if (!entry.len)
443 continue;
444 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
445 }
446 strbuf_release(&entry);
447 strbuf_release(&objdirbuf);
448}
449
450void read_info_alternates(const char * relative_base, int depth)
451{
452 char *map;
453 size_t mapsz;
454 struct stat st;
455 char *path;
456 int fd;
457
458 path = xstrfmt("%s/info/alternates", relative_base);
459 fd = git_open(path);
460 free(path);
461 if (fd < 0)
462 return;
463 if (fstat(fd, &st) || (st.st_size == 0)) {
464 close(fd);
465 return;
466 }
467 mapsz = xsize_t(st.st_size);
468 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
469 close(fd);
470
471 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
472
473 munmap(map, mapsz);
474}
475
476struct alternate_object_database *alloc_alt_odb(const char *dir)
477{
478 struct alternate_object_database *ent;
479
480 FLEX_ALLOC_STR(ent, path, dir);
481 strbuf_init(&ent->scratch, 0);
482 strbuf_addf(&ent->scratch, "%s/", dir);
483 ent->base_len = ent->scratch.len;
484
485 return ent;
486}
487
488void add_to_alternates_file(const char *reference)
489{
490 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
491 char *alts = git_pathdup("objects/info/alternates");
492 FILE *in, *out;
493
494 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
495 out = fdopen_lock_file(lock, "w");
496 if (!out)
497 die_errno("unable to fdopen alternates lockfile");
498
499 in = fopen(alts, "r");
500 if (in) {
501 struct strbuf line = STRBUF_INIT;
502 int found = 0;
503
504 while (strbuf_getline(&line, in) != EOF) {
505 if (!strcmp(reference, line.buf)) {
506 found = 1;
507 break;
508 }
509 fprintf_or_die(out, "%s\n", line.buf);
510 }
511
512 strbuf_release(&line);
513 fclose(in);
514
515 if (found) {
516 rollback_lock_file(lock);
517 lock = NULL;
518 }
519 }
520 else if (errno != ENOENT)
521 die_errno("unable to read alternates file");
522
523 if (lock) {
524 fprintf_or_die(out, "%s\n", reference);
525 if (commit_lock_file(lock))
526 die_errno("unable to move new alternates file into place");
527 if (alt_odb_tail)
528 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
529 }
530 free(alts);
531}
532
533void add_to_alternates_memory(const char *reference)
534{
535 /*
536 * Make sure alternates are initialized, or else our entry may be
537 * overwritten when they are.
538 */
539 prepare_alt_odb();
540
541 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
542}
543
544/*
545 * Compute the exact path an alternate is at and returns it. In case of
546 * error NULL is returned and the human readable error is added to `err`
547 * `path` may be relative and should point to $GITDIR.
548 * `err` must not be null.
549 */
550char *compute_alternate_path(const char *path, struct strbuf *err)
551{
552 char *ref_git = NULL;
553 const char *repo, *ref_git_s;
554 int seen_error = 0;
555
556 ref_git_s = real_path_if_valid(path);
557 if (!ref_git_s) {
558 seen_error = 1;
559 strbuf_addf(err, _("path '%s' does not exist"), path);
560 goto out;
561 } else
562 /*
563 * Beware: read_gitfile(), real_path() and mkpath()
564 * return static buffer
565 */
566 ref_git = xstrdup(ref_git_s);
567
568 repo = read_gitfile(ref_git);
569 if (!repo)
570 repo = read_gitfile(mkpath("%s/.git", ref_git));
571 if (repo) {
572 free(ref_git);
573 ref_git = xstrdup(repo);
574 }
575
576 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
577 char *ref_git_git = mkpathdup("%s/.git", ref_git);
578 free(ref_git);
579 ref_git = ref_git_git;
580 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
581 struct strbuf sb = STRBUF_INIT;
582 seen_error = 1;
583 if (get_common_dir(&sb, ref_git)) {
584 strbuf_addf(err,
585 _("reference repository '%s' as a linked "
586 "checkout is not supported yet."),
587 path);
588 goto out;
589 }
590
591 strbuf_addf(err, _("reference repository '%s' is not a "
592 "local repository."), path);
593 goto out;
594 }
595
596 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
597 strbuf_addf(err, _("reference repository '%s' is shallow"),
598 path);
599 seen_error = 1;
600 goto out;
601 }
602
603 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
604 strbuf_addf(err,
605 _("reference repository '%s' is grafted"),
606 path);
607 seen_error = 1;
608 goto out;
609 }
610
611out:
612 if (seen_error) {
613 FREE_AND_NULL(ref_git);
614 }
615
616 return ref_git;
617}
618
619int foreach_alt_odb(alt_odb_fn fn, void *cb)
620{
621 struct alternate_object_database *ent;
622 int r = 0;
623
624 prepare_alt_odb();
625 for (ent = alt_odb_list; ent; ent = ent->next) {
626 r = fn(ent, cb);
627 if (r)
628 break;
629 }
630 return r;
631}
632
633void prepare_alt_odb(void)
634{
635 const char *alt;
636
637 if (alt_odb_tail)
638 return;
639
640 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
641 if (!alt) alt = "";
642
643 alt_odb_tail = &alt_odb_list;
644 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
645
646 read_info_alternates(get_object_directory(), 0);
647}
648
649/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
650static int freshen_file(const char *fn)
651{
652 struct utimbuf t;
653 t.actime = t.modtime = time(NULL);
654 return !utime(fn, &t);
655}
656
657/*
658 * All of the check_and_freshen functions return 1 if the file exists and was
659 * freshened (if freshening was requested), 0 otherwise. If they return
660 * 0, you should not assume that it is safe to skip a write of the object (it
661 * either does not exist on disk, or has a stale mtime and may be subject to
662 * pruning).
663 */
664int check_and_freshen_file(const char *fn, int freshen)
665{
666 if (access(fn, F_OK))
667 return 0;
668 if (freshen && !freshen_file(fn))
669 return 0;
670 return 1;
671}
672
673static int check_and_freshen_local(const unsigned char *sha1, int freshen)
674{
675 return check_and_freshen_file(sha1_file_name(sha1), freshen);
676}
677
678static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
679{
680 struct alternate_object_database *alt;
681 prepare_alt_odb();
682 for (alt = alt_odb_list; alt; alt = alt->next) {
683 const char *path = alt_sha1_path(alt, sha1);
684 if (check_and_freshen_file(path, freshen))
685 return 1;
686 }
687 return 0;
688}
689
690static int check_and_freshen(const unsigned char *sha1, int freshen)
691{
692 return check_and_freshen_local(sha1, freshen) ||
693 check_and_freshen_nonlocal(sha1, freshen);
694}
695
696int has_loose_object_nonlocal(const unsigned char *sha1)
697{
698 return check_and_freshen_nonlocal(sha1, 0);
699}
700
701static int has_loose_object(const unsigned char *sha1)
702{
703 return check_and_freshen(sha1, 0);
704}
705
706static unsigned int pack_used_ctr;
707static unsigned int pack_mmap_calls;
708static unsigned int peak_pack_open_windows;
709static unsigned int pack_open_windows;
710static unsigned int pack_open_fds;
711static unsigned int pack_max_fds;
712static size_t peak_pack_mapped;
713static size_t pack_mapped;
714struct packed_git *packed_git;
715
716static struct mru packed_git_mru_storage;
717struct mru *packed_git_mru = &packed_git_mru_storage;
718
719void pack_report(void)
720{
721 fprintf(stderr,
722 "pack_report: getpagesize() = %10" SZ_FMT "\n"
723 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
724 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
725 sz_fmt(getpagesize()),
726 sz_fmt(packed_git_window_size),
727 sz_fmt(packed_git_limit));
728 fprintf(stderr,
729 "pack_report: pack_used_ctr = %10u\n"
730 "pack_report: pack_mmap_calls = %10u\n"
731 "pack_report: pack_open_windows = %10u / %10u\n"
732 "pack_report: pack_mapped = "
733 "%10" SZ_FMT " / %10" SZ_FMT "\n",
734 pack_used_ctr,
735 pack_mmap_calls,
736 pack_open_windows, peak_pack_open_windows,
737 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
738}
739
740/*
741 * Open and mmap the index file at path, perform a couple of
742 * consistency checks, then record its information to p. Return 0 on
743 * success.
744 */
745static int check_packed_git_idx(const char *path, struct packed_git *p)
746{
747 void *idx_map;
748 struct pack_idx_header *hdr;
749 size_t idx_size;
750 uint32_t version, nr, i, *index;
751 int fd = git_open(path);
752 struct stat st;
753
754 if (fd < 0)
755 return -1;
756 if (fstat(fd, &st)) {
757 close(fd);
758 return -1;
759 }
760 idx_size = xsize_t(st.st_size);
761 if (idx_size < 4 * 256 + 20 + 20) {
762 close(fd);
763 return error("index file %s is too small", path);
764 }
765 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
766 close(fd);
767
768 hdr = idx_map;
769 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
770 version = ntohl(hdr->idx_version);
771 if (version < 2 || version > 2) {
772 munmap(idx_map, idx_size);
773 return error("index file %s is version %"PRIu32
774 " and is not supported by this binary"
775 " (try upgrading GIT to a newer version)",
776 path, version);
777 }
778 } else
779 version = 1;
780
781 nr = 0;
782 index = idx_map;
783 if (version > 1)
784 index += 2; /* skip index header */
785 for (i = 0; i < 256; i++) {
786 uint32_t n = ntohl(index[i]);
787 if (n < nr) {
788 munmap(idx_map, idx_size);
789 return error("non-monotonic index %s", path);
790 }
791 nr = n;
792 }
793
794 if (version == 1) {
795 /*
796 * Total size:
797 * - 256 index entries 4 bytes each
798 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
799 * - 20-byte SHA1 of the packfile
800 * - 20-byte SHA1 file checksum
801 */
802 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
803 munmap(idx_map, idx_size);
804 return error("wrong index v1 file size in %s", path);
805 }
806 } else if (version == 2) {
807 /*
808 * Minimum size:
809 * - 8 bytes of header
810 * - 256 index entries 4 bytes each
811 * - 20-byte sha1 entry * nr
812 * - 4-byte crc entry * nr
813 * - 4-byte offset entry * nr
814 * - 20-byte SHA1 of the packfile
815 * - 20-byte SHA1 file checksum
816 * And after the 4-byte offset table might be a
817 * variable sized table containing 8-byte entries
818 * for offsets larger than 2^31.
819 */
820 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
821 unsigned long max_size = min_size;
822 if (nr)
823 max_size += (nr - 1)*8;
824 if (idx_size < min_size || idx_size > max_size) {
825 munmap(idx_map, idx_size);
826 return error("wrong index v2 file size in %s", path);
827 }
828 if (idx_size != min_size &&
829 /*
830 * make sure we can deal with large pack offsets.
831 * 31-bit signed offset won't be enough, neither
832 * 32-bit unsigned one will be.
833 */
834 (sizeof(off_t) <= 4)) {
835 munmap(idx_map, idx_size);
836 return error("pack too large for current definition of off_t in %s", path);
837 }
838 }
839
840 p->index_version = version;
841 p->index_data = idx_map;
842 p->index_size = idx_size;
843 p->num_objects = nr;
844 return 0;
845}
846
847int open_pack_index(struct packed_git *p)
848{
849 char *idx_name;
850 size_t len;
851 int ret;
852
853 if (p->index_data)
854 return 0;
855
856 if (!strip_suffix(p->pack_name, ".pack", &len))
857 die("BUG: pack_name does not end in .pack");
858 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
859 ret = check_packed_git_idx(idx_name, p);
860 free(idx_name);
861 return ret;
862}
863
864static void scan_windows(struct packed_git *p,
865 struct packed_git **lru_p,
866 struct pack_window **lru_w,
867 struct pack_window **lru_l)
868{
869 struct pack_window *w, *w_l;
870
871 for (w_l = NULL, w = p->windows; w; w = w->next) {
872 if (!w->inuse_cnt) {
873 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
874 *lru_p = p;
875 *lru_w = w;
876 *lru_l = w_l;
877 }
878 }
879 w_l = w;
880 }
881}
882
883static int unuse_one_window(struct packed_git *current)
884{
885 struct packed_git *p, *lru_p = NULL;
886 struct pack_window *lru_w = NULL, *lru_l = NULL;
887
888 if (current)
889 scan_windows(current, &lru_p, &lru_w, &lru_l);
890 for (p = packed_git; p; p = p->next)
891 scan_windows(p, &lru_p, &lru_w, &lru_l);
892 if (lru_p) {
893 munmap(lru_w->base, lru_w->len);
894 pack_mapped -= lru_w->len;
895 if (lru_l)
896 lru_l->next = lru_w->next;
897 else
898 lru_p->windows = lru_w->next;
899 free(lru_w);
900 pack_open_windows--;
901 return 1;
902 }
903 return 0;
904}
905
906void release_pack_memory(size_t need)
907{
908 size_t cur = pack_mapped;
909 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
910 ; /* nothing */
911}
912
913static void mmap_limit_check(size_t length)
914{
915 static size_t limit = 0;
916 if (!limit) {
917 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
918 if (!limit)
919 limit = SIZE_MAX;
920 }
921 if (length > limit)
922 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
923 (uintmax_t)length, (uintmax_t)limit);
924}
925
926void *xmmap_gently(void *start, size_t length,
927 int prot, int flags, int fd, off_t offset)
928{
929 void *ret;
930
931 mmap_limit_check(length);
932 ret = mmap(start, length, prot, flags, fd, offset);
933 if (ret == MAP_FAILED) {
934 if (!length)
935 return NULL;
936 release_pack_memory(length);
937 ret = mmap(start, length, prot, flags, fd, offset);
938 }
939 return ret;
940}
941
942void *xmmap(void *start, size_t length,
943 int prot, int flags, int fd, off_t offset)
944{
945 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
946 if (ret == MAP_FAILED)
947 die_errno("mmap failed");
948 return ret;
949}
950
951void close_pack_windows(struct packed_git *p)
952{
953 while (p->windows) {
954 struct pack_window *w = p->windows;
955
956 if (w->inuse_cnt)
957 die("pack '%s' still has open windows to it",
958 p->pack_name);
959 munmap(w->base, w->len);
960 pack_mapped -= w->len;
961 pack_open_windows--;
962 p->windows = w->next;
963 free(w);
964 }
965}
966
967static int close_pack_fd(struct packed_git *p)
968{
969 if (p->pack_fd < 0)
970 return 0;
971
972 close(p->pack_fd);
973 pack_open_fds--;
974 p->pack_fd = -1;
975
976 return 1;
977}
978
979static void close_pack(struct packed_git *p)
980{
981 close_pack_windows(p);
982 close_pack_fd(p);
983 close_pack_index(p);
984}
985
986void close_all_packs(void)
987{
988 struct packed_git *p;
989
990 for (p = packed_git; p; p = p->next)
991 if (p->do_not_close)
992 die("BUG: want to close pack marked 'do-not-close'");
993 else
994 close_pack(p);
995}
996
997
998/*
999 * The LRU pack is the one with the oldest MRU window, preferring packs
1000 * with no used windows, or the oldest mtime if it has no windows allocated.
1001 */
1002static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
1003{
1004 struct pack_window *w, *this_mru_w;
1005 int has_windows_inuse = 0;
1006
1007 /*
1008 * Reject this pack if it has windows and the previously selected
1009 * one does not. If this pack does not have windows, reject
1010 * it if the pack file is newer than the previously selected one.
1011 */
1012 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
1013 return;
1014
1015 for (w = this_mru_w = p->windows; w; w = w->next) {
1016 /*
1017 * Reject this pack if any of its windows are in use,
1018 * but the previously selected pack did not have any
1019 * inuse windows. Otherwise, record that this pack
1020 * has windows in use.
1021 */
1022 if (w->inuse_cnt) {
1023 if (*accept_windows_inuse)
1024 has_windows_inuse = 1;
1025 else
1026 return;
1027 }
1028
1029 if (w->last_used > this_mru_w->last_used)
1030 this_mru_w = w;
1031
1032 /*
1033 * Reject this pack if it has windows that have been
1034 * used more recently than the previously selected pack.
1035 * If the previously selected pack had windows inuse and
1036 * we have not encountered a window in this pack that is
1037 * inuse, skip this check since we prefer a pack with no
1038 * inuse windows to one that has inuse windows.
1039 */
1040 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
1041 this_mru_w->last_used > (*mru_w)->last_used)
1042 return;
1043 }
1044
1045 /*
1046 * Select this pack.
1047 */
1048 *mru_w = this_mru_w;
1049 *lru_p = p;
1050 *accept_windows_inuse = has_windows_inuse;
1051}
1052
1053static int close_one_pack(void)
1054{
1055 struct packed_git *p, *lru_p = NULL;
1056 struct pack_window *mru_w = NULL;
1057 int accept_windows_inuse = 1;
1058
1059 for (p = packed_git; p; p = p->next) {
1060 if (p->pack_fd == -1)
1061 continue;
1062 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1063 }
1064
1065 if (lru_p)
1066 return close_pack_fd(lru_p);
1067
1068 return 0;
1069}
1070
1071void unuse_pack(struct pack_window **w_cursor)
1072{
1073 struct pack_window *w = *w_cursor;
1074 if (w) {
1075 w->inuse_cnt--;
1076 *w_cursor = NULL;
1077 }
1078}
1079
1080void close_pack_index(struct packed_git *p)
1081{
1082 if (p->index_data) {
1083 munmap((void *)p->index_data, p->index_size);
1084 p->index_data = NULL;
1085 }
1086}
1087
1088static unsigned int get_max_fd_limit(void)
1089{
1090#ifdef RLIMIT_NOFILE
1091 {
1092 struct rlimit lim;
1093
1094 if (!getrlimit(RLIMIT_NOFILE, &lim))
1095 return lim.rlim_cur;
1096 }
1097#endif
1098
1099#ifdef _SC_OPEN_MAX
1100 {
1101 long open_max = sysconf(_SC_OPEN_MAX);
1102 if (0 < open_max)
1103 return open_max;
1104 /*
1105 * Otherwise, we got -1 for one of the two
1106 * reasons:
1107 *
1108 * (1) sysconf() did not understand _SC_OPEN_MAX
1109 * and signaled an error with -1; or
1110 * (2) sysconf() said there is no limit.
1111 *
1112 * We _could_ clear errno before calling sysconf() to
1113 * tell these two cases apart and return a huge number
1114 * in the latter case to let the caller cap it to a
1115 * value that is not so selfish, but letting the
1116 * fallback OPEN_MAX codepath take care of these cases
1117 * is a lot simpler.
1118 */
1119 }
1120#endif
1121
1122#ifdef OPEN_MAX
1123 return OPEN_MAX;
1124#else
1125 return 1; /* see the caller ;-) */
1126#endif
1127}
1128
1129/*
1130 * Do not call this directly as this leaks p->pack_fd on error return;
1131 * call open_packed_git() instead.
1132 */
1133static int open_packed_git_1(struct packed_git *p)
1134{
1135 struct stat st;
1136 struct pack_header hdr;
1137 unsigned char sha1[20];
1138 unsigned char *idx_sha1;
1139 long fd_flag;
1140
1141 if (!p->index_data && open_pack_index(p))
1142 return error("packfile %s index unavailable", p->pack_name);
1143
1144 if (!pack_max_fds) {
1145 unsigned int max_fds = get_max_fd_limit();
1146
1147 /* Save 3 for stdin/stdout/stderr, 22 for work */
1148 if (25 < max_fds)
1149 pack_max_fds = max_fds - 25;
1150 else
1151 pack_max_fds = 1;
1152 }
1153
1154 while (pack_max_fds <= pack_open_fds && close_one_pack())
1155 ; /* nothing */
1156
1157 p->pack_fd = git_open(p->pack_name);
1158 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1159 return -1;
1160 pack_open_fds++;
1161
1162 /* If we created the struct before we had the pack we lack size. */
1163 if (!p->pack_size) {
1164 if (!S_ISREG(st.st_mode))
1165 return error("packfile %s not a regular file", p->pack_name);
1166 p->pack_size = st.st_size;
1167 } else if (p->pack_size != st.st_size)
1168 return error("packfile %s size changed", p->pack_name);
1169
1170 /* We leave these file descriptors open with sliding mmap;
1171 * there is no point keeping them open across exec(), though.
1172 */
1173 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1174 if (fd_flag < 0)
1175 return error("cannot determine file descriptor flags");
1176 fd_flag |= FD_CLOEXEC;
1177 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1178 return error("cannot set FD_CLOEXEC");
1179
1180 /* Verify we recognize this pack file format. */
1181 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1182 return error("file %s is far too short to be a packfile", p->pack_name);
1183 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1184 return error("file %s is not a GIT packfile", p->pack_name);
1185 if (!pack_version_ok(hdr.hdr_version))
1186 return error("packfile %s is version %"PRIu32" and not"
1187 " supported (try upgrading GIT to a newer version)",
1188 p->pack_name, ntohl(hdr.hdr_version));
1189
1190 /* Verify the pack matches its index. */
1191 if (p->num_objects != ntohl(hdr.hdr_entries))
1192 return error("packfile %s claims to have %"PRIu32" objects"
1193 " while index indicates %"PRIu32" objects",
1194 p->pack_name, ntohl(hdr.hdr_entries),
1195 p->num_objects);
1196 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1197 return error("end of packfile %s is unavailable", p->pack_name);
1198 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1199 return error("packfile %s signature is unavailable", p->pack_name);
1200 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1201 if (hashcmp(sha1, idx_sha1))
1202 return error("packfile %s does not match index", p->pack_name);
1203 return 0;
1204}
1205
1206static int open_packed_git(struct packed_git *p)
1207{
1208 if (!open_packed_git_1(p))
1209 return 0;
1210 close_pack_fd(p);
1211 return -1;
1212}
1213
1214static int in_window(struct pack_window *win, off_t offset)
1215{
1216 /* We must promise at least 20 bytes (one hash) after the
1217 * offset is available from this window, otherwise the offset
1218 * is not actually in this window and a different window (which
1219 * has that one hash excess) must be used. This is to support
1220 * the object header and delta base parsing routines below.
1221 */
1222 off_t win_off = win->offset;
1223 return win_off <= offset
1224 && (offset + 20) <= (win_off + win->len);
1225}
1226
1227unsigned char *use_pack(struct packed_git *p,
1228 struct pack_window **w_cursor,
1229 off_t offset,
1230 unsigned long *left)
1231{
1232 struct pack_window *win = *w_cursor;
1233
1234 /* Since packfiles end in a hash of their content and it's
1235 * pointless to ask for an offset into the middle of that
1236 * hash, and the in_window function above wouldn't match
1237 * don't allow an offset too close to the end of the file.
1238 */
1239 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1240 die("packfile %s cannot be accessed", p->pack_name);
1241 if (offset > (p->pack_size - 20))
1242 die("offset beyond end of packfile (truncated pack?)");
1243 if (offset < 0)
1244 die(_("offset before end of packfile (broken .idx?)"));
1245
1246 if (!win || !in_window(win, offset)) {
1247 if (win)
1248 win->inuse_cnt--;
1249 for (win = p->windows; win; win = win->next) {
1250 if (in_window(win, offset))
1251 break;
1252 }
1253 if (!win) {
1254 size_t window_align = packed_git_window_size / 2;
1255 off_t len;
1256
1257 if (p->pack_fd == -1 && open_packed_git(p))
1258 die("packfile %s cannot be accessed", p->pack_name);
1259
1260 win = xcalloc(1, sizeof(*win));
1261 win->offset = (offset / window_align) * window_align;
1262 len = p->pack_size - win->offset;
1263 if (len > packed_git_window_size)
1264 len = packed_git_window_size;
1265 win->len = (size_t)len;
1266 pack_mapped += win->len;
1267 while (packed_git_limit < pack_mapped
1268 && unuse_one_window(p))
1269 ; /* nothing */
1270 win->base = xmmap(NULL, win->len,
1271 PROT_READ, MAP_PRIVATE,
1272 p->pack_fd, win->offset);
1273 if (win->base == MAP_FAILED)
1274 die_errno("packfile %s cannot be mapped",
1275 p->pack_name);
1276 if (!win->offset && win->len == p->pack_size
1277 && !p->do_not_close)
1278 close_pack_fd(p);
1279 pack_mmap_calls++;
1280 pack_open_windows++;
1281 if (pack_mapped > peak_pack_mapped)
1282 peak_pack_mapped = pack_mapped;
1283 if (pack_open_windows > peak_pack_open_windows)
1284 peak_pack_open_windows = pack_open_windows;
1285 win->next = p->windows;
1286 p->windows = win;
1287 }
1288 }
1289 if (win != *w_cursor) {
1290 win->last_used = pack_used_ctr++;
1291 win->inuse_cnt++;
1292 *w_cursor = win;
1293 }
1294 offset -= win->offset;
1295 if (left)
1296 *left = win->len - xsize_t(offset);
1297 return win->base + offset;
1298}
1299
1300static struct packed_git *alloc_packed_git(int extra)
1301{
1302 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1303 memset(p, 0, sizeof(*p));
1304 p->pack_fd = -1;
1305 return p;
1306}
1307
1308static void try_to_free_pack_memory(size_t size)
1309{
1310 release_pack_memory(size);
1311}
1312
1313struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1314{
1315 static int have_set_try_to_free_routine;
1316 struct stat st;
1317 size_t alloc;
1318 struct packed_git *p;
1319
1320 if (!have_set_try_to_free_routine) {
1321 have_set_try_to_free_routine = 1;
1322 set_try_to_free_routine(try_to_free_pack_memory);
1323 }
1324
1325 /*
1326 * Make sure a corresponding .pack file exists and that
1327 * the index looks sane.
1328 */
1329 if (!strip_suffix_mem(path, &path_len, ".idx"))
1330 return NULL;
1331
1332 /*
1333 * ".pack" is long enough to hold any suffix we're adding (and
1334 * the use xsnprintf double-checks that)
1335 */
1336 alloc = st_add3(path_len, strlen(".pack"), 1);
1337 p = alloc_packed_git(alloc);
1338 memcpy(p->pack_name, path, path_len);
1339
1340 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1341 if (!access(p->pack_name, F_OK))
1342 p->pack_keep = 1;
1343
1344 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1345 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1346 free(p);
1347 return NULL;
1348 }
1349
1350 /* ok, it looks sane as far as we can check without
1351 * actually mapping the pack file.
1352 */
1353 p->pack_size = st.st_size;
1354 p->pack_local = local;
1355 p->mtime = st.st_mtime;
1356 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1357 hashclr(p->sha1);
1358 return p;
1359}
1360
1361struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1362{
1363 const char *path = sha1_pack_name(sha1);
1364 size_t alloc = st_add(strlen(path), 1);
1365 struct packed_git *p = alloc_packed_git(alloc);
1366
1367 memcpy(p->pack_name, path, alloc); /* includes NUL */
1368 hashcpy(p->sha1, sha1);
1369 if (check_packed_git_idx(idx_path, p)) {
1370 free(p);
1371 return NULL;
1372 }
1373
1374 return p;
1375}
1376
1377void install_packed_git(struct packed_git *pack)
1378{
1379 if (pack->pack_fd != -1)
1380 pack_open_fds++;
1381
1382 pack->next = packed_git;
1383 packed_git = pack;
1384}
1385
1386void (*report_garbage)(unsigned seen_bits, const char *path);
1387
1388static void report_helper(const struct string_list *list,
1389 int seen_bits, int first, int last)
1390{
1391 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1392 return;
1393
1394 for (; first < last; first++)
1395 report_garbage(seen_bits, list->items[first].string);
1396}
1397
1398static void report_pack_garbage(struct string_list *list)
1399{
1400 int i, baselen = -1, first = 0, seen_bits = 0;
1401
1402 if (!report_garbage)
1403 return;
1404
1405 string_list_sort(list);
1406
1407 for (i = 0; i < list->nr; i++) {
1408 const char *path = list->items[i].string;
1409 if (baselen != -1 &&
1410 strncmp(path, list->items[first].string, baselen)) {
1411 report_helper(list, seen_bits, first, i);
1412 baselen = -1;
1413 seen_bits = 0;
1414 }
1415 if (baselen == -1) {
1416 const char *dot = strrchr(path, '.');
1417 if (!dot) {
1418 report_garbage(PACKDIR_FILE_GARBAGE, path);
1419 continue;
1420 }
1421 baselen = dot - path + 1;
1422 first = i;
1423 }
1424 if (!strcmp(path + baselen, "pack"))
1425 seen_bits |= 1;
1426 else if (!strcmp(path + baselen, "idx"))
1427 seen_bits |= 2;
1428 }
1429 report_helper(list, seen_bits, first, list->nr);
1430}
1431
1432static void prepare_packed_git_one(char *objdir, int local)
1433{
1434 struct strbuf path = STRBUF_INIT;
1435 size_t dirnamelen;
1436 DIR *dir;
1437 struct dirent *de;
1438 struct string_list garbage = STRING_LIST_INIT_DUP;
1439
1440 strbuf_addstr(&path, objdir);
1441 strbuf_addstr(&path, "/pack");
1442 dir = opendir(path.buf);
1443 if (!dir) {
1444 if (errno != ENOENT)
1445 error_errno("unable to open object pack directory: %s",
1446 path.buf);
1447 strbuf_release(&path);
1448 return;
1449 }
1450 strbuf_addch(&path, '/');
1451 dirnamelen = path.len;
1452 while ((de = readdir(dir)) != NULL) {
1453 struct packed_git *p;
1454 size_t base_len;
1455
1456 if (is_dot_or_dotdot(de->d_name))
1457 continue;
1458
1459 strbuf_setlen(&path, dirnamelen);
1460 strbuf_addstr(&path, de->d_name);
1461
1462 base_len = path.len;
1463 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1464 /* Don't reopen a pack we already have. */
1465 for (p = packed_git; p; p = p->next) {
1466 size_t len;
1467 if (strip_suffix(p->pack_name, ".pack", &len) &&
1468 len == base_len &&
1469 !memcmp(p->pack_name, path.buf, len))
1470 break;
1471 }
1472 if (p == NULL &&
1473 /*
1474 * See if it really is a valid .idx file with
1475 * corresponding .pack file that we can map.
1476 */
1477 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1478 install_packed_git(p);
1479 }
1480
1481 if (!report_garbage)
1482 continue;
1483
1484 if (ends_with(de->d_name, ".idx") ||
1485 ends_with(de->d_name, ".pack") ||
1486 ends_with(de->d_name, ".bitmap") ||
1487 ends_with(de->d_name, ".keep"))
1488 string_list_append(&garbage, path.buf);
1489 else
1490 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1491 }
1492 closedir(dir);
1493 report_pack_garbage(&garbage);
1494 string_list_clear(&garbage, 0);
1495 strbuf_release(&path);
1496}
1497
1498static int approximate_object_count_valid;
1499
1500/*
1501 * Give a fast, rough count of the number of objects in the repository. This
1502 * ignores loose objects completely. If you have a lot of them, then either
1503 * you should repack because your performance will be awful, or they are
1504 * all unreachable objects about to be pruned, in which case they're not really
1505 * interesting as a measure of repo size in the first place.
1506 */
1507unsigned long approximate_object_count(void)
1508{
1509 static unsigned long count;
1510 if (!approximate_object_count_valid) {
1511 struct packed_git *p;
1512
1513 prepare_packed_git();
1514 count = 0;
1515 for (p = packed_git; p; p = p->next) {
1516 if (open_pack_index(p))
1517 continue;
1518 count += p->num_objects;
1519 }
1520 }
1521 return count;
1522}
1523
1524static void *get_next_packed_git(const void *p)
1525{
1526 return ((const struct packed_git *)p)->next;
1527}
1528
1529static void set_next_packed_git(void *p, void *next)
1530{
1531 ((struct packed_git *)p)->next = next;
1532}
1533
1534static int sort_pack(const void *a_, const void *b_)
1535{
1536 const struct packed_git *a = a_;
1537 const struct packed_git *b = b_;
1538 int st;
1539
1540 /*
1541 * Local packs tend to contain objects specific to our
1542 * variant of the project than remote ones. In addition,
1543 * remote ones could be on a network mounted filesystem.
1544 * Favor local ones for these reasons.
1545 */
1546 st = a->pack_local - b->pack_local;
1547 if (st)
1548 return -st;
1549
1550 /*
1551 * Younger packs tend to contain more recent objects,
1552 * and more recent objects tend to get accessed more
1553 * often.
1554 */
1555 if (a->mtime < b->mtime)
1556 return 1;
1557 else if (a->mtime == b->mtime)
1558 return 0;
1559 return -1;
1560}
1561
1562static void rearrange_packed_git(void)
1563{
1564 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1565 set_next_packed_git, sort_pack);
1566}
1567
1568static void prepare_packed_git_mru(void)
1569{
1570 struct packed_git *p;
1571
1572 mru_clear(packed_git_mru);
1573 for (p = packed_git; p; p = p->next)
1574 mru_append(packed_git_mru, p);
1575}
1576
1577static int prepare_packed_git_run_once = 0;
1578void prepare_packed_git(void)
1579{
1580 struct alternate_object_database *alt;
1581
1582 if (prepare_packed_git_run_once)
1583 return;
1584 prepare_packed_git_one(get_object_directory(), 1);
1585 prepare_alt_odb();
1586 for (alt = alt_odb_list; alt; alt = alt->next)
1587 prepare_packed_git_one(alt->path, 0);
1588 rearrange_packed_git();
1589 prepare_packed_git_mru();
1590 prepare_packed_git_run_once = 1;
1591}
1592
1593void reprepare_packed_git(void)
1594{
1595 approximate_object_count_valid = 0;
1596 prepare_packed_git_run_once = 0;
1597 prepare_packed_git();
1598}
1599
1600static void mark_bad_packed_object(struct packed_git *p,
1601 const unsigned char *sha1)
1602{
1603 unsigned i;
1604 for (i = 0; i < p->num_bad_objects; i++)
1605 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1606 return;
1607 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1608 st_mult(GIT_MAX_RAWSZ,
1609 st_add(p->num_bad_objects, 1)));
1610 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1611 p->num_bad_objects++;
1612}
1613
1614static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1615{
1616 struct packed_git *p;
1617 unsigned i;
1618
1619 for (p = packed_git; p; p = p->next)
1620 for (i = 0; i < p->num_bad_objects; i++)
1621 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1622 return p;
1623 return NULL;
1624}
1625
1626/*
1627 * With an in-core object data in "map", rehash it to make sure the
1628 * object name actually matches "sha1" to detect object corruption.
1629 * With "map" == NULL, try reading the object named with "sha1" using
1630 * the streaming interface and rehash it to do the same.
1631 */
1632int check_sha1_signature(const unsigned char *sha1, void *map,
1633 unsigned long size, const char *type)
1634{
1635 unsigned char real_sha1[20];
1636 enum object_type obj_type;
1637 struct git_istream *st;
1638 git_SHA_CTX c;
1639 char hdr[32];
1640 int hdrlen;
1641
1642 if (map) {
1643 hash_sha1_file(map, size, type, real_sha1);
1644 return hashcmp(sha1, real_sha1) ? -1 : 0;
1645 }
1646
1647 st = open_istream(sha1, &obj_type, &size, NULL);
1648 if (!st)
1649 return -1;
1650
1651 /* Generate the header */
1652 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1653
1654 /* Sha1.. */
1655 git_SHA1_Init(&c);
1656 git_SHA1_Update(&c, hdr, hdrlen);
1657 for (;;) {
1658 char buf[1024 * 16];
1659 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1660
1661 if (readlen < 0) {
1662 close_istream(st);
1663 return -1;
1664 }
1665 if (!readlen)
1666 break;
1667 git_SHA1_Update(&c, buf, readlen);
1668 }
1669 git_SHA1_Final(real_sha1, &c);
1670 close_istream(st);
1671 return hashcmp(sha1, real_sha1) ? -1 : 0;
1672}
1673
1674int git_open_cloexec(const char *name, int flags)
1675{
1676 int fd;
1677 static int o_cloexec = O_CLOEXEC;
1678
1679 fd = open(name, flags | o_cloexec);
1680 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1681 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1682 o_cloexec &= ~O_CLOEXEC;
1683 fd = open(name, flags | o_cloexec);
1684 }
1685
1686#if defined(F_GETFL) && defined(F_SETFL) && defined(FD_CLOEXEC)
1687 {
1688 static int fd_cloexec = FD_CLOEXEC;
1689
1690 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1691 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1692 int flags = fcntl(fd, F_GETFL);
1693 if (fcntl(fd, F_SETFL, flags | fd_cloexec))
1694 fd_cloexec = 0;
1695 }
1696 }
1697#endif
1698 return fd;
1699}
1700
1701/*
1702 * Find "sha1" as a loose object in the local repository or in an alternate.
1703 * Returns 0 on success, negative on failure.
1704 *
1705 * The "path" out-parameter will give the path of the object we found (if any).
1706 * Note that it may point to static storage and is only valid until another
1707 * call to sha1_file_name(), etc.
1708 */
1709static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1710 const char **path)
1711{
1712 struct alternate_object_database *alt;
1713
1714 *path = sha1_file_name(sha1);
1715 if (!lstat(*path, st))
1716 return 0;
1717
1718 prepare_alt_odb();
1719 errno = ENOENT;
1720 for (alt = alt_odb_list; alt; alt = alt->next) {
1721 *path = alt_sha1_path(alt, sha1);
1722 if (!lstat(*path, st))
1723 return 0;
1724 }
1725
1726 return -1;
1727}
1728
1729/*
1730 * Like stat_sha1_file(), but actually open the object and return the
1731 * descriptor. See the caveats on the "path" parameter above.
1732 */
1733static int open_sha1_file(const unsigned char *sha1, const char **path)
1734{
1735 int fd;
1736 struct alternate_object_database *alt;
1737 int most_interesting_errno;
1738
1739 *path = sha1_file_name(sha1);
1740 fd = git_open(*path);
1741 if (fd >= 0)
1742 return fd;
1743 most_interesting_errno = errno;
1744
1745 prepare_alt_odb();
1746 for (alt = alt_odb_list; alt; alt = alt->next) {
1747 *path = alt_sha1_path(alt, sha1);
1748 fd = git_open(*path);
1749 if (fd >= 0)
1750 return fd;
1751 if (most_interesting_errno == ENOENT)
1752 most_interesting_errno = errno;
1753 }
1754 errno = most_interesting_errno;
1755 return -1;
1756}
1757
1758/*
1759 * Map the loose object at "path" if it is not NULL, or the path found by
1760 * searching for a loose object named "sha1".
1761 */
1762static void *map_sha1_file_1(const char *path,
1763 const unsigned char *sha1,
1764 unsigned long *size)
1765{
1766 void *map;
1767 int fd;
1768
1769 if (path)
1770 fd = git_open(path);
1771 else
1772 fd = open_sha1_file(sha1, &path);
1773 map = NULL;
1774 if (fd >= 0) {
1775 struct stat st;
1776
1777 if (!fstat(fd, &st)) {
1778 *size = xsize_t(st.st_size);
1779 if (!*size) {
1780 /* mmap() is forbidden on empty files */
1781 error("object file %s is empty", path);
1782 return NULL;
1783 }
1784 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1785 }
1786 close(fd);
1787 }
1788 return map;
1789}
1790
1791void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1792{
1793 return map_sha1_file_1(NULL, sha1, size);
1794}
1795
1796unsigned long unpack_object_header_buffer(const unsigned char *buf,
1797 unsigned long len, enum object_type *type, unsigned long *sizep)
1798{
1799 unsigned shift;
1800 unsigned long size, c;
1801 unsigned long used = 0;
1802
1803 c = buf[used++];
1804 *type = (c >> 4) & 7;
1805 size = c & 15;
1806 shift = 4;
1807 while (c & 0x80) {
1808 if (len <= used || bitsizeof(long) <= shift) {
1809 error("bad object header");
1810 size = used = 0;
1811 break;
1812 }
1813 c = buf[used++];
1814 size += (c & 0x7f) << shift;
1815 shift += 7;
1816 }
1817 *sizep = size;
1818 return used;
1819}
1820
1821static int unpack_sha1_short_header(git_zstream *stream,
1822 unsigned char *map, unsigned long mapsize,
1823 void *buffer, unsigned long bufsiz)
1824{
1825 /* Get the data stream */
1826 memset(stream, 0, sizeof(*stream));
1827 stream->next_in = map;
1828 stream->avail_in = mapsize;
1829 stream->next_out = buffer;
1830 stream->avail_out = bufsiz;
1831
1832 git_inflate_init(stream);
1833 return git_inflate(stream, 0);
1834}
1835
1836int unpack_sha1_header(git_zstream *stream,
1837 unsigned char *map, unsigned long mapsize,
1838 void *buffer, unsigned long bufsiz)
1839{
1840 int status = unpack_sha1_short_header(stream, map, mapsize,
1841 buffer, bufsiz);
1842
1843 if (status < Z_OK)
1844 return status;
1845
1846 /* Make sure we have the terminating NUL */
1847 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1848 return -1;
1849 return 0;
1850}
1851
1852static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1853 unsigned long mapsize, void *buffer,
1854 unsigned long bufsiz, struct strbuf *header)
1855{
1856 int status;
1857
1858 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1859 if (status < Z_OK)
1860 return -1;
1861
1862 /*
1863 * Check if entire header is unpacked in the first iteration.
1864 */
1865 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1866 return 0;
1867
1868 /*
1869 * buffer[0..bufsiz] was not large enough. Copy the partial
1870 * result out to header, and then append the result of further
1871 * reading the stream.
1872 */
1873 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1874 stream->next_out = buffer;
1875 stream->avail_out = bufsiz;
1876
1877 do {
1878 status = git_inflate(stream, 0);
1879 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1880 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1881 return 0;
1882 stream->next_out = buffer;
1883 stream->avail_out = bufsiz;
1884 } while (status != Z_STREAM_END);
1885 return -1;
1886}
1887
1888static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1889{
1890 int bytes = strlen(buffer) + 1;
1891 unsigned char *buf = xmallocz(size);
1892 unsigned long n;
1893 int status = Z_OK;
1894
1895 n = stream->total_out - bytes;
1896 if (n > size)
1897 n = size;
1898 memcpy(buf, (char *) buffer + bytes, n);
1899 bytes = n;
1900 if (bytes <= size) {
1901 /*
1902 * The above condition must be (bytes <= size), not
1903 * (bytes < size). In other words, even though we
1904 * expect no more output and set avail_out to zero,
1905 * the input zlib stream may have bytes that express
1906 * "this concludes the stream", and we *do* want to
1907 * eat that input.
1908 *
1909 * Otherwise we would not be able to test that we
1910 * consumed all the input to reach the expected size;
1911 * we also want to check that zlib tells us that all
1912 * went well with status == Z_STREAM_END at the end.
1913 */
1914 stream->next_out = buf + bytes;
1915 stream->avail_out = size - bytes;
1916 while (status == Z_OK)
1917 status = git_inflate(stream, Z_FINISH);
1918 }
1919 if (status == Z_STREAM_END && !stream->avail_in) {
1920 git_inflate_end(stream);
1921 return buf;
1922 }
1923
1924 if (status < 0)
1925 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1926 else if (stream->avail_in)
1927 error("garbage at end of loose object '%s'",
1928 sha1_to_hex(sha1));
1929 free(buf);
1930 return NULL;
1931}
1932
1933/*
1934 * We used to just use "sscanf()", but that's actually way
1935 * too permissive for what we want to check. So do an anal
1936 * object header parse by hand.
1937 */
1938static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1939 unsigned int flags)
1940{
1941 const char *type_buf = hdr;
1942 unsigned long size;
1943 int type, type_len = 0;
1944
1945 /*
1946 * The type can be of any size but is followed by
1947 * a space.
1948 */
1949 for (;;) {
1950 char c = *hdr++;
1951 if (!c)
1952 return -1;
1953 if (c == ' ')
1954 break;
1955 type_len++;
1956 }
1957
1958 type = type_from_string_gently(type_buf, type_len, 1);
1959 if (oi->typename)
1960 strbuf_add(oi->typename, type_buf, type_len);
1961 /*
1962 * Set type to 0 if its an unknown object and
1963 * we're obtaining the type using '--allow-unknown-type'
1964 * option.
1965 */
1966 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1967 type = 0;
1968 else if (type < 0)
1969 die("invalid object type");
1970 if (oi->typep)
1971 *oi->typep = type;
1972
1973 /*
1974 * The length must follow immediately, and be in canonical
1975 * decimal format (ie "010" is not valid).
1976 */
1977 size = *hdr++ - '0';
1978 if (size > 9)
1979 return -1;
1980 if (size) {
1981 for (;;) {
1982 unsigned long c = *hdr - '0';
1983 if (c > 9)
1984 break;
1985 hdr++;
1986 size = size * 10 + c;
1987 }
1988 }
1989
1990 if (oi->sizep)
1991 *oi->sizep = size;
1992
1993 /*
1994 * The length must be followed by a zero byte
1995 */
1996 return *hdr ? -1 : type;
1997}
1998
1999int parse_sha1_header(const char *hdr, unsigned long *sizep)
2000{
2001 struct object_info oi = OBJECT_INFO_INIT;
2002
2003 oi.sizep = sizep;
2004 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
2005}
2006
2007static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
2008{
2009 int ret;
2010 git_zstream stream;
2011 char hdr[8192];
2012
2013 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
2014 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
2015 return NULL;
2016
2017 return unpack_sha1_rest(&stream, hdr, *size, sha1);
2018}
2019
2020unsigned long get_size_from_delta(struct packed_git *p,
2021 struct pack_window **w_curs,
2022 off_t curpos)
2023{
2024 const unsigned char *data;
2025 unsigned char delta_head[20], *in;
2026 git_zstream stream;
2027 int st;
2028
2029 memset(&stream, 0, sizeof(stream));
2030 stream.next_out = delta_head;
2031 stream.avail_out = sizeof(delta_head);
2032
2033 git_inflate_init(&stream);
2034 do {
2035 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2036 stream.next_in = in;
2037 st = git_inflate(&stream, Z_FINISH);
2038 curpos += stream.next_in - in;
2039 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
2040 stream.total_out < sizeof(delta_head));
2041 git_inflate_end(&stream);
2042 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
2043 error("delta data unpack-initial failed");
2044 return 0;
2045 }
2046
2047 /* Examine the initial part of the delta to figure out
2048 * the result size.
2049 */
2050 data = delta_head;
2051
2052 /* ignore base size */
2053 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2054
2055 /* Read the result size */
2056 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2057}
2058
2059static off_t get_delta_base(struct packed_git *p,
2060 struct pack_window **w_curs,
2061 off_t *curpos,
2062 enum object_type type,
2063 off_t delta_obj_offset)
2064{
2065 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
2066 off_t base_offset;
2067
2068 /* use_pack() assured us we have [base_info, base_info + 20)
2069 * as a range that we can look at without walking off the
2070 * end of the mapped window. Its actually the hash size
2071 * that is assured. An OFS_DELTA longer than the hash size
2072 * is stupid, as then a REF_DELTA would be smaller to store.
2073 */
2074 if (type == OBJ_OFS_DELTA) {
2075 unsigned used = 0;
2076 unsigned char c = base_info[used++];
2077 base_offset = c & 127;
2078 while (c & 128) {
2079 base_offset += 1;
2080 if (!base_offset || MSB(base_offset, 7))
2081 return 0; /* overflow */
2082 c = base_info[used++];
2083 base_offset = (base_offset << 7) + (c & 127);
2084 }
2085 base_offset = delta_obj_offset - base_offset;
2086 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2087 return 0; /* out of bound */
2088 *curpos += used;
2089 } else if (type == OBJ_REF_DELTA) {
2090 /* The base entry _must_ be in the same pack */
2091 base_offset = find_pack_entry_one(base_info, p);
2092 *curpos += 20;
2093 } else
2094 die("I am totally screwed");
2095 return base_offset;
2096}
2097
2098/*
2099 * Like get_delta_base above, but we return the sha1 instead of the pack
2100 * offset. This means it is cheaper for REF deltas (we do not have to do
2101 * the final object lookup), but more expensive for OFS deltas (we
2102 * have to load the revidx to convert the offset back into a sha1).
2103 */
2104static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2105 struct pack_window **w_curs,
2106 off_t curpos,
2107 enum object_type type,
2108 off_t delta_obj_offset)
2109{
2110 if (type == OBJ_REF_DELTA) {
2111 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2112 return base;
2113 } else if (type == OBJ_OFS_DELTA) {
2114 struct revindex_entry *revidx;
2115 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2116 type, delta_obj_offset);
2117
2118 if (!base_offset)
2119 return NULL;
2120
2121 revidx = find_pack_revindex(p, base_offset);
2122 if (!revidx)
2123 return NULL;
2124
2125 return nth_packed_object_sha1(p, revidx->nr);
2126 } else
2127 return NULL;
2128}
2129
2130int unpack_object_header(struct packed_git *p,
2131 struct pack_window **w_curs,
2132 off_t *curpos,
2133 unsigned long *sizep)
2134{
2135 unsigned char *base;
2136 unsigned long left;
2137 unsigned long used;
2138 enum object_type type;
2139
2140 /* use_pack() assures us we have [base, base + 20) available
2141 * as a range that we can look at. (Its actually the hash
2142 * size that is assured.) With our object header encoding
2143 * the maximum deflated object size is 2^137, which is just
2144 * insane, so we know won't exceed what we have been given.
2145 */
2146 base = use_pack(p, w_curs, *curpos, &left);
2147 used = unpack_object_header_buffer(base, left, &type, sizep);
2148 if (!used) {
2149 type = OBJ_BAD;
2150 } else
2151 *curpos += used;
2152
2153 return type;
2154}
2155
2156static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2157{
2158 int type;
2159 struct revindex_entry *revidx;
2160 const unsigned char *sha1;
2161 revidx = find_pack_revindex(p, obj_offset);
2162 if (!revidx)
2163 return OBJ_BAD;
2164 sha1 = nth_packed_object_sha1(p, revidx->nr);
2165 mark_bad_packed_object(p, sha1);
2166 type = sha1_object_info(sha1, NULL);
2167 if (type <= OBJ_NONE)
2168 return OBJ_BAD;
2169 return type;
2170}
2171
2172#define POI_STACK_PREALLOC 64
2173
2174static enum object_type packed_to_object_type(struct packed_git *p,
2175 off_t obj_offset,
2176 enum object_type type,
2177 struct pack_window **w_curs,
2178 off_t curpos)
2179{
2180 off_t small_poi_stack[POI_STACK_PREALLOC];
2181 off_t *poi_stack = small_poi_stack;
2182 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2183
2184 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2185 off_t base_offset;
2186 unsigned long size;
2187 /* Push the object we're going to leave behind */
2188 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2189 poi_stack_alloc = alloc_nr(poi_stack_nr);
2190 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2191 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2192 } else {
2193 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2194 }
2195 poi_stack[poi_stack_nr++] = obj_offset;
2196 /* If parsing the base offset fails, just unwind */
2197 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2198 if (!base_offset)
2199 goto unwind;
2200 curpos = obj_offset = base_offset;
2201 type = unpack_object_header(p, w_curs, &curpos, &size);
2202 if (type <= OBJ_NONE) {
2203 /* If getting the base itself fails, we first
2204 * retry the base, otherwise unwind */
2205 type = retry_bad_packed_offset(p, base_offset);
2206 if (type > OBJ_NONE)
2207 goto out;
2208 goto unwind;
2209 }
2210 }
2211
2212 switch (type) {
2213 case OBJ_BAD:
2214 case OBJ_COMMIT:
2215 case OBJ_TREE:
2216 case OBJ_BLOB:
2217 case OBJ_TAG:
2218 break;
2219 default:
2220 error("unknown object type %i at offset %"PRIuMAX" in %s",
2221 type, (uintmax_t)obj_offset, p->pack_name);
2222 type = OBJ_BAD;
2223 }
2224
2225out:
2226 if (poi_stack != small_poi_stack)
2227 free(poi_stack);
2228 return type;
2229
2230unwind:
2231 while (poi_stack_nr) {
2232 obj_offset = poi_stack[--poi_stack_nr];
2233 type = retry_bad_packed_offset(p, obj_offset);
2234 if (type > OBJ_NONE)
2235 goto out;
2236 }
2237 type = OBJ_BAD;
2238 goto out;
2239}
2240
2241int packed_object_info(struct packed_git *p, off_t obj_offset,
2242 struct object_info *oi)
2243{
2244 struct pack_window *w_curs = NULL;
2245 unsigned long size;
2246 off_t curpos = obj_offset;
2247 enum object_type type;
2248
2249 /*
2250 * We always get the representation type, but only convert it to
2251 * a "real" type later if the caller is interested.
2252 */
2253 type = unpack_object_header(p, &w_curs, &curpos, &size);
2254
2255 if (oi->sizep) {
2256 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2257 off_t tmp_pos = curpos;
2258 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2259 type, obj_offset);
2260 if (!base_offset) {
2261 type = OBJ_BAD;
2262 goto out;
2263 }
2264 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2265 if (*oi->sizep == 0) {
2266 type = OBJ_BAD;
2267 goto out;
2268 }
2269 } else {
2270 *oi->sizep = size;
2271 }
2272 }
2273
2274 if (oi->disk_sizep) {
2275 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2276 *oi->disk_sizep = revidx[1].offset - obj_offset;
2277 }
2278
2279 if (oi->typep) {
2280 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2281 if (*oi->typep < 0) {
2282 type = OBJ_BAD;
2283 goto out;
2284 }
2285 }
2286
2287 if (oi->delta_base_sha1) {
2288 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2289 const unsigned char *base;
2290
2291 base = get_delta_base_sha1(p, &w_curs, curpos,
2292 type, obj_offset);
2293 if (!base) {
2294 type = OBJ_BAD;
2295 goto out;
2296 }
2297
2298 hashcpy(oi->delta_base_sha1, base);
2299 } else
2300 hashclr(oi->delta_base_sha1);
2301 }
2302
2303out:
2304 unuse_pack(&w_curs);
2305 return type;
2306}
2307
2308static void *unpack_compressed_entry(struct packed_git *p,
2309 struct pack_window **w_curs,
2310 off_t curpos,
2311 unsigned long size)
2312{
2313 int st;
2314 git_zstream stream;
2315 unsigned char *buffer, *in;
2316
2317 buffer = xmallocz_gently(size);
2318 if (!buffer)
2319 return NULL;
2320 memset(&stream, 0, sizeof(stream));
2321 stream.next_out = buffer;
2322 stream.avail_out = size + 1;
2323
2324 git_inflate_init(&stream);
2325 do {
2326 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2327 stream.next_in = in;
2328 st = git_inflate(&stream, Z_FINISH);
2329 if (!stream.avail_out)
2330 break; /* the payload is larger than it should be */
2331 curpos += stream.next_in - in;
2332 } while (st == Z_OK || st == Z_BUF_ERROR);
2333 git_inflate_end(&stream);
2334 if ((st != Z_STREAM_END) || stream.total_out != size) {
2335 free(buffer);
2336 return NULL;
2337 }
2338
2339 return buffer;
2340}
2341
2342static struct hashmap delta_base_cache;
2343static size_t delta_base_cached;
2344
2345static LIST_HEAD(delta_base_cache_lru);
2346
2347struct delta_base_cache_key {
2348 struct packed_git *p;
2349 off_t base_offset;
2350};
2351
2352struct delta_base_cache_entry {
2353 struct hashmap hash;
2354 struct delta_base_cache_key key;
2355 struct list_head lru;
2356 void *data;
2357 unsigned long size;
2358 enum object_type type;
2359};
2360
2361static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2362{
2363 unsigned int hash;
2364
2365 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2366 hash += (hash >> 8) + (hash >> 16);
2367 return hash;
2368}
2369
2370static struct delta_base_cache_entry *
2371get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2372{
2373 struct hashmap_entry entry;
2374 struct delta_base_cache_key key;
2375
2376 if (!delta_base_cache.cmpfn)
2377 return NULL;
2378
2379 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2380 key.p = p;
2381 key.base_offset = base_offset;
2382 return hashmap_get(&delta_base_cache, &entry, &key);
2383}
2384
2385static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2386 const struct delta_base_cache_key *b)
2387{
2388 return a->p == b->p && a->base_offset == b->base_offset;
2389}
2390
2391static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2392 const void *vkey)
2393{
2394 const struct delta_base_cache_entry *a = va, *b = vb;
2395 const struct delta_base_cache_key *key = vkey;
2396 if (key)
2397 return !delta_base_cache_key_eq(&a->key, key);
2398 else
2399 return !delta_base_cache_key_eq(&a->key, &b->key);
2400}
2401
2402static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2403{
2404 return !!get_delta_base_cache_entry(p, base_offset);
2405}
2406
2407/*
2408 * Remove the entry from the cache, but do _not_ free the associated
2409 * entry data. The caller takes ownership of the "data" buffer, and
2410 * should copy out any fields it wants before detaching.
2411 */
2412static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2413{
2414 hashmap_remove(&delta_base_cache, ent, &ent->key);
2415 list_del(&ent->lru);
2416 delta_base_cached -= ent->size;
2417 free(ent);
2418}
2419
2420static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2421 unsigned long *base_size, enum object_type *type)
2422{
2423 struct delta_base_cache_entry *ent;
2424
2425 ent = get_delta_base_cache_entry(p, base_offset);
2426 if (!ent)
2427 return unpack_entry(p, base_offset, type, base_size);
2428
2429 *type = ent->type;
2430 *base_size = ent->size;
2431 return xmemdupz(ent->data, ent->size);
2432}
2433
2434static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2435{
2436 free(ent->data);
2437 detach_delta_base_cache_entry(ent);
2438}
2439
2440void clear_delta_base_cache(void)
2441{
2442 struct list_head *lru, *tmp;
2443 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2444 struct delta_base_cache_entry *entry =
2445 list_entry(lru, struct delta_base_cache_entry, lru);
2446 release_delta_base_cache(entry);
2447 }
2448}
2449
2450static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2451 void *base, unsigned long base_size, enum object_type type)
2452{
2453 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2454 struct list_head *lru, *tmp;
2455
2456 delta_base_cached += base_size;
2457
2458 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2459 struct delta_base_cache_entry *f =
2460 list_entry(lru, struct delta_base_cache_entry, lru);
2461 if (delta_base_cached <= delta_base_cache_limit)
2462 break;
2463 release_delta_base_cache(f);
2464 }
2465
2466 ent->key.p = p;
2467 ent->key.base_offset = base_offset;
2468 ent->type = type;
2469 ent->data = base;
2470 ent->size = base_size;
2471 list_add_tail(&ent->lru, &delta_base_cache_lru);
2472
2473 if (!delta_base_cache.cmpfn)
2474 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2475 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2476 hashmap_add(&delta_base_cache, ent);
2477}
2478
2479static void *read_object(const unsigned char *sha1, enum object_type *type,
2480 unsigned long *size);
2481
2482static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2483{
2484 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2485 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2486 p->pack_name, (uintmax_t)obj_offset);
2487}
2488
2489int do_check_packed_object_crc;
2490
2491#define UNPACK_ENTRY_STACK_PREALLOC 64
2492struct unpack_entry_stack_ent {
2493 off_t obj_offset;
2494 off_t curpos;
2495 unsigned long size;
2496};
2497
2498void *unpack_entry(struct packed_git *p, off_t obj_offset,
2499 enum object_type *final_type, unsigned long *final_size)
2500{
2501 struct pack_window *w_curs = NULL;
2502 off_t curpos = obj_offset;
2503 void *data = NULL;
2504 unsigned long size;
2505 enum object_type type;
2506 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2507 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2508 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2509 int base_from_cache = 0;
2510
2511 write_pack_access_log(p, obj_offset);
2512
2513 /* PHASE 1: drill down to the innermost base object */
2514 for (;;) {
2515 off_t base_offset;
2516 int i;
2517 struct delta_base_cache_entry *ent;
2518
2519 ent = get_delta_base_cache_entry(p, curpos);
2520 if (ent) {
2521 type = ent->type;
2522 data = ent->data;
2523 size = ent->size;
2524 detach_delta_base_cache_entry(ent);
2525 base_from_cache = 1;
2526 break;
2527 }
2528
2529 if (do_check_packed_object_crc && p->index_version > 1) {
2530 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2531 off_t len = revidx[1].offset - obj_offset;
2532 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2533 const unsigned char *sha1 =
2534 nth_packed_object_sha1(p, revidx->nr);
2535 error("bad packed object CRC for %s",
2536 sha1_to_hex(sha1));
2537 mark_bad_packed_object(p, sha1);
2538 unuse_pack(&w_curs);
2539 return NULL;
2540 }
2541 }
2542
2543 type = unpack_object_header(p, &w_curs, &curpos, &size);
2544 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2545 break;
2546
2547 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2548 if (!base_offset) {
2549 error("failed to validate delta base reference "
2550 "at offset %"PRIuMAX" from %s",
2551 (uintmax_t)curpos, p->pack_name);
2552 /* bail to phase 2, in hopes of recovery */
2553 data = NULL;
2554 break;
2555 }
2556
2557 /* push object, proceed to base */
2558 if (delta_stack_nr >= delta_stack_alloc
2559 && delta_stack == small_delta_stack) {
2560 delta_stack_alloc = alloc_nr(delta_stack_nr);
2561 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2562 memcpy(delta_stack, small_delta_stack,
2563 sizeof(*delta_stack)*delta_stack_nr);
2564 } else {
2565 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2566 }
2567 i = delta_stack_nr++;
2568 delta_stack[i].obj_offset = obj_offset;
2569 delta_stack[i].curpos = curpos;
2570 delta_stack[i].size = size;
2571
2572 curpos = obj_offset = base_offset;
2573 }
2574
2575 /* PHASE 2: handle the base */
2576 switch (type) {
2577 case OBJ_OFS_DELTA:
2578 case OBJ_REF_DELTA:
2579 if (data)
2580 die("BUG: unpack_entry: left loop at a valid delta");
2581 break;
2582 case OBJ_COMMIT:
2583 case OBJ_TREE:
2584 case OBJ_BLOB:
2585 case OBJ_TAG:
2586 if (!base_from_cache)
2587 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2588 break;
2589 default:
2590 data = NULL;
2591 error("unknown object type %i at offset %"PRIuMAX" in %s",
2592 type, (uintmax_t)obj_offset, p->pack_name);
2593 }
2594
2595 /* PHASE 3: apply deltas in order */
2596
2597 /* invariants:
2598 * 'data' holds the base data, or NULL if there was corruption
2599 */
2600 while (delta_stack_nr) {
2601 void *delta_data;
2602 void *base = data;
2603 void *external_base = NULL;
2604 unsigned long delta_size, base_size = size;
2605 int i;
2606
2607 data = NULL;
2608
2609 if (base)
2610 add_delta_base_cache(p, obj_offset, base, base_size, type);
2611
2612 if (!base) {
2613 /*
2614 * We're probably in deep shit, but let's try to fetch
2615 * the required base anyway from another pack or loose.
2616 * This is costly but should happen only in the presence
2617 * of a corrupted pack, and is better than failing outright.
2618 */
2619 struct revindex_entry *revidx;
2620 const unsigned char *base_sha1;
2621 revidx = find_pack_revindex(p, obj_offset);
2622 if (revidx) {
2623 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2624 error("failed to read delta base object %s"
2625 " at offset %"PRIuMAX" from %s",
2626 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2627 p->pack_name);
2628 mark_bad_packed_object(p, base_sha1);
2629 base = read_object(base_sha1, &type, &base_size);
2630 external_base = base;
2631 }
2632 }
2633
2634 i = --delta_stack_nr;
2635 obj_offset = delta_stack[i].obj_offset;
2636 curpos = delta_stack[i].curpos;
2637 delta_size = delta_stack[i].size;
2638
2639 if (!base)
2640 continue;
2641
2642 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2643
2644 if (!delta_data) {
2645 error("failed to unpack compressed delta "
2646 "at offset %"PRIuMAX" from %s",
2647 (uintmax_t)curpos, p->pack_name);
2648 data = NULL;
2649 free(external_base);
2650 continue;
2651 }
2652
2653 data = patch_delta(base, base_size,
2654 delta_data, delta_size,
2655 &size);
2656
2657 /*
2658 * We could not apply the delta; warn the user, but keep going.
2659 * Our failure will be noticed either in the next iteration of
2660 * the loop, or if this is the final delta, in the caller when
2661 * we return NULL. Those code paths will take care of making
2662 * a more explicit warning and retrying with another copy of
2663 * the object.
2664 */
2665 if (!data)
2666 error("failed to apply delta");
2667
2668 free(delta_data);
2669 free(external_base);
2670 }
2671
2672 *final_type = type;
2673 *final_size = size;
2674
2675 unuse_pack(&w_curs);
2676
2677 if (delta_stack != small_delta_stack)
2678 free(delta_stack);
2679
2680 return data;
2681}
2682
2683const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2684 uint32_t n)
2685{
2686 const unsigned char *index = p->index_data;
2687 if (!index) {
2688 if (open_pack_index(p))
2689 return NULL;
2690 index = p->index_data;
2691 }
2692 if (n >= p->num_objects)
2693 return NULL;
2694 index += 4 * 256;
2695 if (p->index_version == 1) {
2696 return index + 24 * n + 4;
2697 } else {
2698 index += 8;
2699 return index + 20 * n;
2700 }
2701}
2702
2703const struct object_id *nth_packed_object_oid(struct object_id *oid,
2704 struct packed_git *p,
2705 uint32_t n)
2706{
2707 const unsigned char *hash = nth_packed_object_sha1(p, n);
2708 if (!hash)
2709 return NULL;
2710 hashcpy(oid->hash, hash);
2711 return oid;
2712}
2713
2714void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2715{
2716 const unsigned char *ptr = vptr;
2717 const unsigned char *start = p->index_data;
2718 const unsigned char *end = start + p->index_size;
2719 if (ptr < start)
2720 die(_("offset before start of pack index for %s (corrupt index?)"),
2721 p->pack_name);
2722 /* No need to check for underflow; .idx files must be at least 8 bytes */
2723 if (ptr >= end - 8)
2724 die(_("offset beyond end of pack index for %s (truncated index?)"),
2725 p->pack_name);
2726}
2727
2728off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2729{
2730 const unsigned char *index = p->index_data;
2731 index += 4 * 256;
2732 if (p->index_version == 1) {
2733 return ntohl(*((uint32_t *)(index + 24 * n)));
2734 } else {
2735 uint32_t off;
2736 index += 8 + p->num_objects * (20 + 4);
2737 off = ntohl(*((uint32_t *)(index + 4 * n)));
2738 if (!(off & 0x80000000))
2739 return off;
2740 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2741 check_pack_index_ptr(p, index);
2742 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2743 ntohl(*((uint32_t *)(index + 4)));
2744 }
2745}
2746
2747off_t find_pack_entry_one(const unsigned char *sha1,
2748 struct packed_git *p)
2749{
2750 const uint32_t *level1_ofs = p->index_data;
2751 const unsigned char *index = p->index_data;
2752 unsigned hi, lo, stride;
2753 static int use_lookup = -1;
2754 static int debug_lookup = -1;
2755
2756 if (debug_lookup < 0)
2757 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2758
2759 if (!index) {
2760 if (open_pack_index(p))
2761 return 0;
2762 level1_ofs = p->index_data;
2763 index = p->index_data;
2764 }
2765 if (p->index_version > 1) {
2766 level1_ofs += 2;
2767 index += 8;
2768 }
2769 index += 4 * 256;
2770 hi = ntohl(level1_ofs[*sha1]);
2771 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2772 if (p->index_version > 1) {
2773 stride = 20;
2774 } else {
2775 stride = 24;
2776 index += 4;
2777 }
2778
2779 if (debug_lookup)
2780 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2781 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2782
2783 if (use_lookup < 0)
2784 use_lookup = !!getenv("GIT_USE_LOOKUP");
2785 if (use_lookup) {
2786 int pos = sha1_entry_pos(index, stride, 0,
2787 lo, hi, p->num_objects, sha1);
2788 if (pos < 0)
2789 return 0;
2790 return nth_packed_object_offset(p, pos);
2791 }
2792
2793 do {
2794 unsigned mi = (lo + hi) / 2;
2795 int cmp = hashcmp(index + mi * stride, sha1);
2796
2797 if (debug_lookup)
2798 printf("lo %u hi %u rg %u mi %u\n",
2799 lo, hi, hi - lo, mi);
2800 if (!cmp)
2801 return nth_packed_object_offset(p, mi);
2802 if (cmp > 0)
2803 hi = mi;
2804 else
2805 lo = mi+1;
2806 } while (lo < hi);
2807 return 0;
2808}
2809
2810int is_pack_valid(struct packed_git *p)
2811{
2812 /* An already open pack is known to be valid. */
2813 if (p->pack_fd != -1)
2814 return 1;
2815
2816 /* If the pack has one window completely covering the
2817 * file size, the pack is known to be valid even if
2818 * the descriptor is not currently open.
2819 */
2820 if (p->windows) {
2821 struct pack_window *w = p->windows;
2822
2823 if (!w->offset && w->len == p->pack_size)
2824 return 1;
2825 }
2826
2827 /* Force the pack to open to prove its valid. */
2828 return !open_packed_git(p);
2829}
2830
2831static int fill_pack_entry(const unsigned char *sha1,
2832 struct pack_entry *e,
2833 struct packed_git *p)
2834{
2835 off_t offset;
2836
2837 if (p->num_bad_objects) {
2838 unsigned i;
2839 for (i = 0; i < p->num_bad_objects; i++)
2840 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2841 return 0;
2842 }
2843
2844 offset = find_pack_entry_one(sha1, p);
2845 if (!offset)
2846 return 0;
2847
2848 /*
2849 * We are about to tell the caller where they can locate the
2850 * requested object. We better make sure the packfile is
2851 * still here and can be accessed before supplying that
2852 * answer, as it may have been deleted since the index was
2853 * loaded!
2854 */
2855 if (!is_pack_valid(p))
2856 return 0;
2857 e->offset = offset;
2858 e->p = p;
2859 hashcpy(e->sha1, sha1);
2860 return 1;
2861}
2862
2863/*
2864 * Iff a pack file contains the object named by sha1, return true and
2865 * store its location to e.
2866 */
2867static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2868{
2869 struct mru_entry *p;
2870
2871 prepare_packed_git();
2872 if (!packed_git)
2873 return 0;
2874
2875 for (p = packed_git_mru->head; p; p = p->next) {
2876 if (fill_pack_entry(sha1, e, p->item)) {
2877 mru_mark(packed_git_mru, p);
2878 return 1;
2879 }
2880 }
2881 return 0;
2882}
2883
2884struct packed_git *find_sha1_pack(const unsigned char *sha1,
2885 struct packed_git *packs)
2886{
2887 struct packed_git *p;
2888
2889 for (p = packs; p; p = p->next) {
2890 if (find_pack_entry_one(sha1, p))
2891 return p;
2892 }
2893 return NULL;
2894
2895}
2896
2897static int sha1_loose_object_info(const unsigned char *sha1,
2898 struct object_info *oi,
2899 int flags)
2900{
2901 int status = 0;
2902 unsigned long mapsize;
2903 void *map;
2904 git_zstream stream;
2905 char hdr[32];
2906 struct strbuf hdrbuf = STRBUF_INIT;
2907
2908 if (oi->delta_base_sha1)
2909 hashclr(oi->delta_base_sha1);
2910
2911 /*
2912 * If we don't care about type or size, then we don't
2913 * need to look inside the object at all. Note that we
2914 * do not optimize out the stat call, even if the
2915 * caller doesn't care about the disk-size, since our
2916 * return value implicitly indicates whether the
2917 * object even exists.
2918 */
2919 if (!oi->typep && !oi->typename && !oi->sizep) {
2920 const char *path;
2921 struct stat st;
2922 if (stat_sha1_file(sha1, &st, &path) < 0)
2923 return -1;
2924 if (oi->disk_sizep)
2925 *oi->disk_sizep = st.st_size;
2926 return 0;
2927 }
2928
2929 map = map_sha1_file(sha1, &mapsize);
2930 if (!map)
2931 return -1;
2932 if (oi->disk_sizep)
2933 *oi->disk_sizep = mapsize;
2934 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2935 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2936 status = error("unable to unpack %s header with --allow-unknown-type",
2937 sha1_to_hex(sha1));
2938 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2939 status = error("unable to unpack %s header",
2940 sha1_to_hex(sha1));
2941 if (status < 0)
2942 ; /* Do nothing */
2943 else if (hdrbuf.len) {
2944 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2945 status = error("unable to parse %s header with --allow-unknown-type",
2946 sha1_to_hex(sha1));
2947 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2948 status = error("unable to parse %s header", sha1_to_hex(sha1));
2949 git_inflate_end(&stream);
2950 munmap(map, mapsize);
2951 if (status && oi->typep)
2952 *oi->typep = status;
2953 strbuf_release(&hdrbuf);
2954 return (status < 0) ? status : 0;
2955}
2956
2957int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2958{
2959 struct cached_object *co;
2960 struct pack_entry e;
2961 int rtype;
2962 enum object_type real_type;
2963 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2964
2965 co = find_cached_object(real);
2966 if (co) {
2967 if (oi->typep)
2968 *(oi->typep) = co->type;
2969 if (oi->sizep)
2970 *(oi->sizep) = co->size;
2971 if (oi->disk_sizep)
2972 *(oi->disk_sizep) = 0;
2973 if (oi->delta_base_sha1)
2974 hashclr(oi->delta_base_sha1);
2975 if (oi->typename)
2976 strbuf_addstr(oi->typename, typename(co->type));
2977 oi->whence = OI_CACHED;
2978 return 0;
2979 }
2980
2981 if (!find_pack_entry(real, &e)) {
2982 /* Most likely it's a loose object. */
2983 if (!sha1_loose_object_info(real, oi, flags)) {
2984 oi->whence = OI_LOOSE;
2985 return 0;
2986 }
2987
2988 /* Not a loose object; someone else may have just packed it. */
2989 reprepare_packed_git();
2990 if (!find_pack_entry(real, &e))
2991 return -1;
2992 }
2993
2994 /*
2995 * packed_object_info() does not follow the delta chain to
2996 * find out the real type, unless it is given oi->typep.
2997 */
2998 if (oi->typename && !oi->typep)
2999 oi->typep = &real_type;
3000
3001 rtype = packed_object_info(e.p, e.offset, oi);
3002 if (rtype < 0) {
3003 mark_bad_packed_object(e.p, real);
3004 if (oi->typep == &real_type)
3005 oi->typep = NULL;
3006 return sha1_object_info_extended(real, oi, 0);
3007 } else if (in_delta_base_cache(e.p, e.offset)) {
3008 oi->whence = OI_DBCACHED;
3009 } else {
3010 oi->whence = OI_PACKED;
3011 oi->u.packed.offset = e.offset;
3012 oi->u.packed.pack = e.p;
3013 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
3014 rtype == OBJ_OFS_DELTA);
3015 }
3016 if (oi->typename)
3017 strbuf_addstr(oi->typename, typename(*oi->typep));
3018 if (oi->typep == &real_type)
3019 oi->typep = NULL;
3020
3021 return 0;
3022}
3023
3024/* returns enum object_type or negative */
3025int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
3026{
3027 enum object_type type;
3028 struct object_info oi = OBJECT_INFO_INIT;
3029
3030 oi.typep = &type;
3031 oi.sizep = sizep;
3032 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
3033 return -1;
3034 return type;
3035}
3036
3037static void *read_packed_sha1(const unsigned char *sha1,
3038 enum object_type *type, unsigned long *size)
3039{
3040 struct pack_entry e;
3041 void *data;
3042
3043 if (!find_pack_entry(sha1, &e))
3044 return NULL;
3045 data = cache_or_unpack_entry(e.p, e.offset, size, type);
3046 if (!data) {
3047 /*
3048 * We're probably in deep shit, but let's try to fetch
3049 * the required object anyway from another pack or loose.
3050 * This should happen only in the presence of a corrupted
3051 * pack, and is better than failing outright.
3052 */
3053 error("failed to read object %s at offset %"PRIuMAX" from %s",
3054 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
3055 mark_bad_packed_object(e.p, sha1);
3056 data = read_object(sha1, type, size);
3057 }
3058 return data;
3059}
3060
3061int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
3062 unsigned char *sha1)
3063{
3064 struct cached_object *co;
3065
3066 hash_sha1_file(buf, len, typename(type), sha1);
3067 if (has_sha1_file(sha1) || find_cached_object(sha1))
3068 return 0;
3069 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
3070 co = &cached_objects[cached_object_nr++];
3071 co->size = len;
3072 co->type = type;
3073 co->buf = xmalloc(len);
3074 memcpy(co->buf, buf, len);
3075 hashcpy(co->sha1, sha1);
3076 return 0;
3077}
3078
3079static void *read_object(const unsigned char *sha1, enum object_type *type,
3080 unsigned long *size)
3081{
3082 unsigned long mapsize;
3083 void *map, *buf;
3084 struct cached_object *co;
3085
3086 co = find_cached_object(sha1);
3087 if (co) {
3088 *type = co->type;
3089 *size = co->size;
3090 return xmemdupz(co->buf, co->size);
3091 }
3092
3093 buf = read_packed_sha1(sha1, type, size);
3094 if (buf)
3095 return buf;
3096 map = map_sha1_file(sha1, &mapsize);
3097 if (map) {
3098 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
3099 munmap(map, mapsize);
3100 return buf;
3101 }
3102 reprepare_packed_git();
3103 return read_packed_sha1(sha1, type, size);
3104}
3105
3106/*
3107 * This function dies on corrupt objects; the callers who want to
3108 * deal with them should arrange to call read_object() and give error
3109 * messages themselves.
3110 */
3111void *read_sha1_file_extended(const unsigned char *sha1,
3112 enum object_type *type,
3113 unsigned long *size,
3114 unsigned flag)
3115{
3116 void *data;
3117 const struct packed_git *p;
3118 const char *path;
3119 struct stat st;
3120 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3121
3122 errno = 0;
3123 data = read_object(repl, type, size);
3124 if (data)
3125 return data;
3126
3127 if (errno && errno != ENOENT)
3128 die_errno("failed to read object %s", sha1_to_hex(sha1));
3129
3130 /* die if we replaced an object with one that does not exist */
3131 if (repl != sha1)
3132 die("replacement %s not found for %s",
3133 sha1_to_hex(repl), sha1_to_hex(sha1));
3134
3135 if (!stat_sha1_file(repl, &st, &path))
3136 die("loose object %s (stored in %s) is corrupt",
3137 sha1_to_hex(repl), path);
3138
3139 if ((p = has_packed_and_bad(repl)) != NULL)
3140 die("packed object %s (stored in %s) is corrupt",
3141 sha1_to_hex(repl), p->pack_name);
3142
3143 return NULL;
3144}
3145
3146void *read_object_with_reference(const unsigned char *sha1,
3147 const char *required_type_name,
3148 unsigned long *size,
3149 unsigned char *actual_sha1_return)
3150{
3151 enum object_type type, required_type;
3152 void *buffer;
3153 unsigned long isize;
3154 unsigned char actual_sha1[20];
3155
3156 required_type = type_from_string(required_type_name);
3157 hashcpy(actual_sha1, sha1);
3158 while (1) {
3159 int ref_length = -1;
3160 const char *ref_type = NULL;
3161
3162 buffer = read_sha1_file(actual_sha1, &type, &isize);
3163 if (!buffer)
3164 return NULL;
3165 if (type == required_type) {
3166 *size = isize;
3167 if (actual_sha1_return)
3168 hashcpy(actual_sha1_return, actual_sha1);
3169 return buffer;
3170 }
3171 /* Handle references */
3172 else if (type == OBJ_COMMIT)
3173 ref_type = "tree ";
3174 else if (type == OBJ_TAG)
3175 ref_type = "object ";
3176 else {
3177 free(buffer);
3178 return NULL;
3179 }
3180 ref_length = strlen(ref_type);
3181
3182 if (ref_length + 40 > isize ||
3183 memcmp(buffer, ref_type, ref_length) ||
3184 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3185 free(buffer);
3186 return NULL;
3187 }
3188 free(buffer);
3189 /* Now we have the ID of the referred-to object in
3190 * actual_sha1. Check again. */
3191 }
3192}
3193
3194static void write_sha1_file_prepare(const void *buf, unsigned long len,
3195 const char *type, unsigned char *sha1,
3196 char *hdr, int *hdrlen)
3197{
3198 git_SHA_CTX c;
3199
3200 /* Generate the header */
3201 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3202
3203 /* Sha1.. */
3204 git_SHA1_Init(&c);
3205 git_SHA1_Update(&c, hdr, *hdrlen);
3206 git_SHA1_Update(&c, buf, len);
3207 git_SHA1_Final(sha1, &c);
3208}
3209
3210/*
3211 * Move the just written object into its final resting place.
3212 */
3213int finalize_object_file(const char *tmpfile, const char *filename)
3214{
3215 int ret = 0;
3216
3217 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3218 goto try_rename;
3219 else if (link(tmpfile, filename))
3220 ret = errno;
3221
3222 /*
3223 * Coda hack - coda doesn't like cross-directory links,
3224 * so we fall back to a rename, which will mean that it
3225 * won't be able to check collisions, but that's not a
3226 * big deal.
3227 *
3228 * The same holds for FAT formatted media.
3229 *
3230 * When this succeeds, we just return. We have nothing
3231 * left to unlink.
3232 */
3233 if (ret && ret != EEXIST) {
3234 try_rename:
3235 if (!rename(tmpfile, filename))
3236 goto out;
3237 ret = errno;
3238 }
3239 unlink_or_warn(tmpfile);
3240 if (ret) {
3241 if (ret != EEXIST) {
3242 return error_errno("unable to write sha1 filename %s", filename);
3243 }
3244 /* FIXME!!! Collision check here ? */
3245 }
3246
3247out:
3248 if (adjust_shared_perm(filename))
3249 return error("unable to set permission to '%s'", filename);
3250 return 0;
3251}
3252
3253static int write_buffer(int fd, const void *buf, size_t len)
3254{
3255 if (write_in_full(fd, buf, len) < 0)
3256 return error_errno("file write error");
3257 return 0;
3258}
3259
3260int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3261 unsigned char *sha1)
3262{
3263 char hdr[32];
3264 int hdrlen = sizeof(hdr);
3265 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3266 return 0;
3267}
3268
3269/* Finalize a file on disk, and close it. */
3270static void close_sha1_file(int fd)
3271{
3272 if (fsync_object_files)
3273 fsync_or_die(fd, "sha1 file");
3274 if (close(fd) != 0)
3275 die_errno("error when closing sha1 file");
3276}
3277
3278/* Size of directory component, including the ending '/' */
3279static inline int directory_size(const char *filename)
3280{
3281 const char *s = strrchr(filename, '/');
3282 if (!s)
3283 return 0;
3284 return s - filename + 1;
3285}
3286
3287/*
3288 * This creates a temporary file in the same directory as the final
3289 * 'filename'
3290 *
3291 * We want to avoid cross-directory filename renames, because those
3292 * can have problems on various filesystems (FAT, NFS, Coda).
3293 */
3294static int create_tmpfile(struct strbuf *tmp, const char *filename)
3295{
3296 int fd, dirlen = directory_size(filename);
3297
3298 strbuf_reset(tmp);
3299 strbuf_add(tmp, filename, dirlen);
3300 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3301 fd = git_mkstemp_mode(tmp->buf, 0444);
3302 if (fd < 0 && dirlen && errno == ENOENT) {
3303 /*
3304 * Make sure the directory exists; note that the contents
3305 * of the buffer are undefined after mkstemp returns an
3306 * error, so we have to rewrite the whole buffer from
3307 * scratch.
3308 */
3309 strbuf_reset(tmp);
3310 strbuf_add(tmp, filename, dirlen - 1);
3311 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3312 return -1;
3313 if (adjust_shared_perm(tmp->buf))
3314 return -1;
3315
3316 /* Try again */
3317 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3318 fd = git_mkstemp_mode(tmp->buf, 0444);
3319 }
3320 return fd;
3321}
3322
3323static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3324 const void *buf, unsigned long len, time_t mtime)
3325{
3326 int fd, ret;
3327 unsigned char compressed[4096];
3328 git_zstream stream;
3329 git_SHA_CTX c;
3330 unsigned char parano_sha1[20];
3331 static struct strbuf tmp_file = STRBUF_INIT;
3332 const char *filename = sha1_file_name(sha1);
3333
3334 fd = create_tmpfile(&tmp_file, filename);
3335 if (fd < 0) {
3336 if (errno == EACCES)
3337 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3338 else
3339 return error_errno("unable to create temporary file");
3340 }
3341
3342 /* Set it up */
3343 git_deflate_init(&stream, zlib_compression_level);
3344 stream.next_out = compressed;
3345 stream.avail_out = sizeof(compressed);
3346 git_SHA1_Init(&c);
3347
3348 /* First header.. */
3349 stream.next_in = (unsigned char *)hdr;
3350 stream.avail_in = hdrlen;
3351 while (git_deflate(&stream, 0) == Z_OK)
3352 ; /* nothing */
3353 git_SHA1_Update(&c, hdr, hdrlen);
3354
3355 /* Then the data itself.. */
3356 stream.next_in = (void *)buf;
3357 stream.avail_in = len;
3358 do {
3359 unsigned char *in0 = stream.next_in;
3360 ret = git_deflate(&stream, Z_FINISH);
3361 git_SHA1_Update(&c, in0, stream.next_in - in0);
3362 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3363 die("unable to write sha1 file");
3364 stream.next_out = compressed;
3365 stream.avail_out = sizeof(compressed);
3366 } while (ret == Z_OK);
3367
3368 if (ret != Z_STREAM_END)
3369 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3370 ret = git_deflate_end_gently(&stream);
3371 if (ret != Z_OK)
3372 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3373 git_SHA1_Final(parano_sha1, &c);
3374 if (hashcmp(sha1, parano_sha1) != 0)
3375 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3376
3377 close_sha1_file(fd);
3378
3379 if (mtime) {
3380 struct utimbuf utb;
3381 utb.actime = mtime;
3382 utb.modtime = mtime;
3383 if (utime(tmp_file.buf, &utb) < 0)
3384 warning_errno("failed utime() on %s", tmp_file.buf);
3385 }
3386
3387 return finalize_object_file(tmp_file.buf, filename);
3388}
3389
3390static int freshen_loose_object(const unsigned char *sha1)
3391{
3392 return check_and_freshen(sha1, 1);
3393}
3394
3395static int freshen_packed_object(const unsigned char *sha1)
3396{
3397 struct pack_entry e;
3398 if (!find_pack_entry(sha1, &e))
3399 return 0;
3400 if (e.p->freshened)
3401 return 1;
3402 if (!freshen_file(e.p->pack_name))
3403 return 0;
3404 e.p->freshened = 1;
3405 return 1;
3406}
3407
3408int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3409{
3410 char hdr[32];
3411 int hdrlen = sizeof(hdr);
3412
3413 /* Normally if we have it in the pack then we do not bother writing
3414 * it out into .git/objects/??/?{38} file.
3415 */
3416 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3417 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3418 return 0;
3419 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3420}
3421
3422int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3423 unsigned char *sha1, unsigned flags)
3424{
3425 char *header;
3426 int hdrlen, status = 0;
3427
3428 /* type string, SP, %lu of the length plus NUL must fit this */
3429 hdrlen = strlen(type) + 32;
3430 header = xmalloc(hdrlen);
3431 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3432
3433 if (!(flags & HASH_WRITE_OBJECT))
3434 goto cleanup;
3435 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3436 goto cleanup;
3437 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3438
3439cleanup:
3440 free(header);
3441 return status;
3442}
3443
3444int force_object_loose(const unsigned char *sha1, time_t mtime)
3445{
3446 void *buf;
3447 unsigned long len;
3448 enum object_type type;
3449 char hdr[32];
3450 int hdrlen;
3451 int ret;
3452
3453 if (has_loose_object(sha1))
3454 return 0;
3455 buf = read_packed_sha1(sha1, &type, &len);
3456 if (!buf)
3457 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3458 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3459 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3460 free(buf);
3461
3462 return ret;
3463}
3464
3465int has_pack_index(const unsigned char *sha1)
3466{
3467 struct stat st;
3468 if (stat(sha1_pack_index_name(sha1), &st))
3469 return 0;
3470 return 1;
3471}
3472
3473int has_sha1_pack(const unsigned char *sha1)
3474{
3475 struct pack_entry e;
3476 return find_pack_entry(sha1, &e);
3477}
3478
3479int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3480{
3481 struct pack_entry e;
3482
3483 if (!startup_info->have_repository)
3484 return 0;
3485 if (find_pack_entry(sha1, &e))
3486 return 1;
3487 if (has_loose_object(sha1))
3488 return 1;
3489 if (flags & HAS_SHA1_QUICK)
3490 return 0;
3491 reprepare_packed_git();
3492 return find_pack_entry(sha1, &e);
3493}
3494
3495int has_object_file(const struct object_id *oid)
3496{
3497 return has_sha1_file(oid->hash);
3498}
3499
3500int has_object_file_with_flags(const struct object_id *oid, int flags)
3501{
3502 return has_sha1_file_with_flags(oid->hash, flags);
3503}
3504
3505static void check_tree(const void *buf, size_t size)
3506{
3507 struct tree_desc desc;
3508 struct name_entry entry;
3509
3510 init_tree_desc(&desc, buf, size);
3511 while (tree_entry(&desc, &entry))
3512 /* do nothing
3513 * tree_entry() will die() on malformed entries */
3514 ;
3515}
3516
3517static void check_commit(const void *buf, size_t size)
3518{
3519 struct commit c;
3520 memset(&c, 0, sizeof(c));
3521 if (parse_commit_buffer(&c, buf, size))
3522 die("corrupt commit");
3523}
3524
3525static void check_tag(const void *buf, size_t size)
3526{
3527 struct tag t;
3528 memset(&t, 0, sizeof(t));
3529 if (parse_tag_buffer(&t, buf, size))
3530 die("corrupt tag");
3531}
3532
3533static int index_mem(unsigned char *sha1, void *buf, size_t size,
3534 enum object_type type,
3535 const char *path, unsigned flags)
3536{
3537 int ret, re_allocated = 0;
3538 int write_object = flags & HASH_WRITE_OBJECT;
3539
3540 if (!type)
3541 type = OBJ_BLOB;
3542
3543 /*
3544 * Convert blobs to git internal format
3545 */
3546 if ((type == OBJ_BLOB) && path) {
3547 struct strbuf nbuf = STRBUF_INIT;
3548 if (convert_to_git(path, buf, size, &nbuf,
3549 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3550 buf = strbuf_detach(&nbuf, &size);
3551 re_allocated = 1;
3552 }
3553 }
3554 if (flags & HASH_FORMAT_CHECK) {
3555 if (type == OBJ_TREE)
3556 check_tree(buf, size);
3557 if (type == OBJ_COMMIT)
3558 check_commit(buf, size);
3559 if (type == OBJ_TAG)
3560 check_tag(buf, size);
3561 }
3562
3563 if (write_object)
3564 ret = write_sha1_file(buf, size, typename(type), sha1);
3565 else
3566 ret = hash_sha1_file(buf, size, typename(type), sha1);
3567 if (re_allocated)
3568 free(buf);
3569 return ret;
3570}
3571
3572static int index_stream_convert_blob(unsigned char *sha1, int fd,
3573 const char *path, unsigned flags)
3574{
3575 int ret;
3576 const int write_object = flags & HASH_WRITE_OBJECT;
3577 struct strbuf sbuf = STRBUF_INIT;
3578
3579 assert(path);
3580 assert(would_convert_to_git_filter_fd(path));
3581
3582 convert_to_git_filter_fd(path, fd, &sbuf,
3583 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3584
3585 if (write_object)
3586 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3587 sha1);
3588 else
3589 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3590 sha1);
3591 strbuf_release(&sbuf);
3592 return ret;
3593}
3594
3595static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3596 const char *path, unsigned flags)
3597{
3598 struct strbuf sbuf = STRBUF_INIT;
3599 int ret;
3600
3601 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3602 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3603 else
3604 ret = -1;
3605 strbuf_release(&sbuf);
3606 return ret;
3607}
3608
3609#define SMALL_FILE_SIZE (32*1024)
3610
3611static int index_core(unsigned char *sha1, int fd, size_t size,
3612 enum object_type type, const char *path,
3613 unsigned flags)
3614{
3615 int ret;
3616
3617 if (!size) {
3618 ret = index_mem(sha1, "", size, type, path, flags);
3619 } else if (size <= SMALL_FILE_SIZE) {
3620 char *buf = xmalloc(size);
3621 if (size == read_in_full(fd, buf, size))
3622 ret = index_mem(sha1, buf, size, type, path, flags);
3623 else
3624 ret = error_errno("short read");
3625 free(buf);
3626 } else {
3627 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3628 ret = index_mem(sha1, buf, size, type, path, flags);
3629 munmap(buf, size);
3630 }
3631 return ret;
3632}
3633
3634/*
3635 * This creates one packfile per large blob unless bulk-checkin
3636 * machinery is "plugged".
3637 *
3638 * This also bypasses the usual "convert-to-git" dance, and that is on
3639 * purpose. We could write a streaming version of the converting
3640 * functions and insert that before feeding the data to fast-import
3641 * (or equivalent in-core API described above). However, that is
3642 * somewhat complicated, as we do not know the size of the filter
3643 * result, which we need to know beforehand when writing a git object.
3644 * Since the primary motivation for trying to stream from the working
3645 * tree file and to avoid mmaping it in core is to deal with large
3646 * binary blobs, they generally do not want to get any conversion, and
3647 * callers should avoid this code path when filters are requested.
3648 */
3649static int index_stream(unsigned char *sha1, int fd, size_t size,
3650 enum object_type type, const char *path,
3651 unsigned flags)
3652{
3653 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3654}
3655
3656int index_fd(unsigned char *sha1, int fd, struct stat *st,
3657 enum object_type type, const char *path, unsigned flags)
3658{
3659 int ret;
3660
3661 /*
3662 * Call xsize_t() only when needed to avoid potentially unnecessary
3663 * die() for large files.
3664 */
3665 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3666 ret = index_stream_convert_blob(sha1, fd, path, flags);
3667 else if (!S_ISREG(st->st_mode))
3668 ret = index_pipe(sha1, fd, type, path, flags);
3669 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3670 (path && would_convert_to_git(path)))
3671 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3672 flags);
3673 else
3674 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3675 flags);
3676 close(fd);
3677 return ret;
3678}
3679
3680int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3681{
3682 int fd;
3683 struct strbuf sb = STRBUF_INIT;
3684
3685 switch (st->st_mode & S_IFMT) {
3686 case S_IFREG:
3687 fd = open(path, O_RDONLY);
3688 if (fd < 0)
3689 return error_errno("open(\"%s\")", path);
3690 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3691 return error("%s: failed to insert into database",
3692 path);
3693 break;
3694 case S_IFLNK:
3695 if (strbuf_readlink(&sb, path, st->st_size))
3696 return error_errno("readlink(\"%s\")", path);
3697 if (!(flags & HASH_WRITE_OBJECT))
3698 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3699 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3700 return error("%s: failed to insert into database",
3701 path);
3702 strbuf_release(&sb);
3703 break;
3704 case S_IFDIR:
3705 return resolve_gitlink_ref(path, "HEAD", sha1);
3706 default:
3707 return error("%s: unsupported file type", path);
3708 }
3709 return 0;
3710}
3711
3712int read_pack_header(int fd, struct pack_header *header)
3713{
3714 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3715 /* "eof before pack header was fully read" */
3716 return PH_ERROR_EOF;
3717
3718 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3719 /* "protocol error (pack signature mismatch detected)" */
3720 return PH_ERROR_PACK_SIGNATURE;
3721 if (!pack_version_ok(header->hdr_version))
3722 /* "protocol error (pack version unsupported)" */
3723 return PH_ERROR_PROTOCOL;
3724 return 0;
3725}
3726
3727void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3728{
3729 enum object_type type = sha1_object_info(sha1, NULL);
3730 if (type < 0)
3731 die("%s is not a valid object", sha1_to_hex(sha1));
3732 if (type != expect)
3733 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3734 typename(expect));
3735}
3736
3737static int for_each_file_in_obj_subdir(int subdir_nr,
3738 struct strbuf *path,
3739 each_loose_object_fn obj_cb,
3740 each_loose_cruft_fn cruft_cb,
3741 each_loose_subdir_fn subdir_cb,
3742 void *data)
3743{
3744 size_t baselen = path->len;
3745 DIR *dir = opendir(path->buf);
3746 struct dirent *de;
3747 int r = 0;
3748
3749 if (!dir) {
3750 if (errno == ENOENT)
3751 return 0;
3752 return error_errno("unable to open %s", path->buf);
3753 }
3754
3755 while ((de = readdir(dir))) {
3756 if (is_dot_or_dotdot(de->d_name))
3757 continue;
3758
3759 strbuf_setlen(path, baselen);
3760 strbuf_addf(path, "/%s", de->d_name);
3761
3762 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3763 char hex[GIT_MAX_HEXSZ+1];
3764 struct object_id oid;
3765
3766 xsnprintf(hex, sizeof(hex), "%02x%s",
3767 subdir_nr, de->d_name);
3768 if (!get_oid_hex(hex, &oid)) {
3769 if (obj_cb) {
3770 r = obj_cb(&oid, path->buf, data);
3771 if (r)
3772 break;
3773 }
3774 continue;
3775 }
3776 }
3777
3778 if (cruft_cb) {
3779 r = cruft_cb(de->d_name, path->buf, data);
3780 if (r)
3781 break;
3782 }
3783 }
3784 closedir(dir);
3785
3786 strbuf_setlen(path, baselen);
3787 if (!r && subdir_cb)
3788 r = subdir_cb(subdir_nr, path->buf, data);
3789
3790 return r;
3791}
3792
3793int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3794 each_loose_object_fn obj_cb,
3795 each_loose_cruft_fn cruft_cb,
3796 each_loose_subdir_fn subdir_cb,
3797 void *data)
3798{
3799 size_t baselen = path->len;
3800 int r = 0;
3801 int i;
3802
3803 for (i = 0; i < 256; i++) {
3804 strbuf_addf(path, "/%02x", i);
3805 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3806 subdir_cb, data);
3807 strbuf_setlen(path, baselen);
3808 if (r)
3809 break;
3810 }
3811
3812 return r;
3813}
3814
3815int for_each_loose_file_in_objdir(const char *path,
3816 each_loose_object_fn obj_cb,
3817 each_loose_cruft_fn cruft_cb,
3818 each_loose_subdir_fn subdir_cb,
3819 void *data)
3820{
3821 struct strbuf buf = STRBUF_INIT;
3822 int r;
3823
3824 strbuf_addstr(&buf, path);
3825 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3826 subdir_cb, data);
3827 strbuf_release(&buf);
3828
3829 return r;
3830}
3831
3832struct loose_alt_odb_data {
3833 each_loose_object_fn *cb;
3834 void *data;
3835};
3836
3837static int loose_from_alt_odb(struct alternate_object_database *alt,
3838 void *vdata)
3839{
3840 struct loose_alt_odb_data *data = vdata;
3841 struct strbuf buf = STRBUF_INIT;
3842 int r;
3843
3844 strbuf_addstr(&buf, alt->path);
3845 r = for_each_loose_file_in_objdir_buf(&buf,
3846 data->cb, NULL, NULL,
3847 data->data);
3848 strbuf_release(&buf);
3849 return r;
3850}
3851
3852int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3853{
3854 struct loose_alt_odb_data alt;
3855 int r;
3856
3857 r = for_each_loose_file_in_objdir(get_object_directory(),
3858 cb, NULL, NULL, data);
3859 if (r)
3860 return r;
3861
3862 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3863 return 0;
3864
3865 alt.cb = cb;
3866 alt.data = data;
3867 return foreach_alt_odb(loose_from_alt_odb, &alt);
3868}
3869
3870static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3871{
3872 uint32_t i;
3873 int r = 0;
3874
3875 for (i = 0; i < p->num_objects; i++) {
3876 struct object_id oid;
3877
3878 if (!nth_packed_object_oid(&oid, p, i))
3879 return error("unable to get sha1 of object %u in %s",
3880 i, p->pack_name);
3881
3882 r = cb(&oid, p, i, data);
3883 if (r)
3884 break;
3885 }
3886 return r;
3887}
3888
3889int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3890{
3891 struct packed_git *p;
3892 int r = 0;
3893 int pack_errors = 0;
3894
3895 prepare_packed_git();
3896 for (p = packed_git; p; p = p->next) {
3897 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3898 continue;
3899 if (open_pack_index(p)) {
3900 pack_errors = 1;
3901 continue;
3902 }
3903 r = for_each_object_in_pack(p, cb, data);
3904 if (r)
3905 break;
3906 }
3907 return r ? r : pack_errors;
3908}
3909
3910static int check_stream_sha1(git_zstream *stream,
3911 const char *hdr,
3912 unsigned long size,
3913 const char *path,
3914 const unsigned char *expected_sha1)
3915{
3916 git_SHA_CTX c;
3917 unsigned char real_sha1[GIT_MAX_RAWSZ];
3918 unsigned char buf[4096];
3919 unsigned long total_read;
3920 int status = Z_OK;
3921
3922 git_SHA1_Init(&c);
3923 git_SHA1_Update(&c, hdr, stream->total_out);
3924
3925 /*
3926 * We already read some bytes into hdr, but the ones up to the NUL
3927 * do not count against the object's content size.
3928 */
3929 total_read = stream->total_out - strlen(hdr) - 1;
3930
3931 /*
3932 * This size comparison must be "<=" to read the final zlib packets;
3933 * see the comment in unpack_sha1_rest for details.
3934 */
3935 while (total_read <= size &&
3936 (status == Z_OK || status == Z_BUF_ERROR)) {
3937 stream->next_out = buf;
3938 stream->avail_out = sizeof(buf);
3939 if (size - total_read < stream->avail_out)
3940 stream->avail_out = size - total_read;
3941 status = git_inflate(stream, Z_FINISH);
3942 git_SHA1_Update(&c, buf, stream->next_out - buf);
3943 total_read += stream->next_out - buf;
3944 }
3945 git_inflate_end(stream);
3946
3947 if (status != Z_STREAM_END) {
3948 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3949 return -1;
3950 }
3951 if (stream->avail_in) {
3952 error("garbage at end of loose object '%s'",
3953 sha1_to_hex(expected_sha1));
3954 return -1;
3955 }
3956
3957 git_SHA1_Final(real_sha1, &c);
3958 if (hashcmp(expected_sha1, real_sha1)) {
3959 error("sha1 mismatch for %s (expected %s)", path,
3960 sha1_to_hex(expected_sha1));
3961 return -1;
3962 }
3963
3964 return 0;
3965}
3966
3967int read_loose_object(const char *path,
3968 const unsigned char *expected_sha1,
3969 enum object_type *type,
3970 unsigned long *size,
3971 void **contents)
3972{
3973 int ret = -1;
3974 void *map = NULL;
3975 unsigned long mapsize;
3976 git_zstream stream;
3977 char hdr[32];
3978
3979 *contents = NULL;
3980
3981 map = map_sha1_file_1(path, NULL, &mapsize);
3982 if (!map) {
3983 error_errno("unable to mmap %s", path);
3984 goto out;
3985 }
3986
3987 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3988 error("unable to unpack header of %s", path);
3989 goto out;
3990 }
3991
3992 *type = parse_sha1_header(hdr, size);
3993 if (*type < 0) {
3994 error("unable to parse header of %s", path);
3995 git_inflate_end(&stream);
3996 goto out;
3997 }
3998
3999 if (*type == OBJ_BLOB) {
4000 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
4001 goto out;
4002 } else {
4003 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
4004 if (!*contents) {
4005 error("unable to unpack contents of %s", path);
4006 git_inflate_end(&stream);
4007 goto out;
4008 }
4009 if (check_sha1_signature(expected_sha1, *contents,
4010 *size, typename(*type))) {
4011 error("sha1 mismatch for %s (expected %s)", path,
4012 sha1_to_hex(expected_sha1));
4013 free(*contents);
4014 goto out;
4015 }
4016 }
4017
4018 ret = 0; /* everything checks out */
4019
4020out:
4021 if (map)
4022 munmap(map, mapsize);
4023 return ret;
4024}