Merge branch 'jc/verify-loose-object-header'
[git/git.git] / sha1_file.c
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
30 #ifndef O_NOATIME
31 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
32 #define O_NOATIME 01000000
33 #else
34 #define O_NOATIME 0
35 #endif
36 #endif
37
38 #define SZ_FMT PRIuMAX
39 static inline uintmax_t sz_fmt(size_t s) { return s; }
40
41 const unsigned char null_sha1[20];
42 const struct object_id null_oid;
43 const struct object_id empty_tree_oid = {
44 EMPTY_TREE_SHA1_BIN_LITERAL
45 };
46 const struct object_id empty_blob_oid = {
47 EMPTY_BLOB_SHA1_BIN_LITERAL
48 };
49
50 /*
51 * This is meant to hold a *small* number of objects that you would
52 * want read_sha1_file() to be able to return, but yet you do not want
53 * to write them into the object store (e.g. a browse-only
54 * application).
55 */
56 static struct cached_object {
57 unsigned char sha1[20];
58 enum object_type type;
59 void *buf;
60 unsigned long size;
61 } *cached_objects;
62 static int cached_object_nr, cached_object_alloc;
63
64 static struct cached_object empty_tree = {
65 EMPTY_TREE_SHA1_BIN_LITERAL,
66 OBJ_TREE,
67 "",
68 0
69 };
70
71 static struct cached_object *find_cached_object(const unsigned char *sha1)
72 {
73 int i;
74 struct cached_object *co = cached_objects;
75
76 for (i = 0; i < cached_object_nr; i++, co++) {
77 if (!hashcmp(co->sha1, sha1))
78 return co;
79 }
80 if (!hashcmp(sha1, empty_tree.sha1))
81 return &empty_tree;
82 return NULL;
83 }
84
85 int mkdir_in_gitdir(const char *path)
86 {
87 if (mkdir(path, 0777)) {
88 int saved_errno = errno;
89 struct stat st;
90 struct strbuf sb = STRBUF_INIT;
91
92 if (errno != EEXIST)
93 return -1;
94 /*
95 * Are we looking at a path in a symlinked worktree
96 * whose original repository does not yet have it?
97 * e.g. .git/rr-cache pointing at its original
98 * repository in which the user hasn't performed any
99 * conflict resolution yet?
100 */
101 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
102 strbuf_readlink(&sb, path, st.st_size) ||
103 !is_absolute_path(sb.buf) ||
104 mkdir(sb.buf, 0777)) {
105 strbuf_release(&sb);
106 errno = saved_errno;
107 return -1;
108 }
109 strbuf_release(&sb);
110 }
111 return adjust_shared_perm(path);
112 }
113
114 enum scld_error safe_create_leading_directories(char *path)
115 {
116 char *next_component = path + offset_1st_component(path);
117 enum scld_error ret = SCLD_OK;
118
119 while (ret == SCLD_OK && next_component) {
120 struct stat st;
121 char *slash = next_component, slash_character;
122
123 while (*slash && !is_dir_sep(*slash))
124 slash++;
125
126 if (!*slash)
127 break;
128
129 next_component = slash + 1;
130 while (is_dir_sep(*next_component))
131 next_component++;
132 if (!*next_component)
133 break;
134
135 slash_character = *slash;
136 *slash = '\0';
137 if (!stat(path, &st)) {
138 /* path exists */
139 if (!S_ISDIR(st.st_mode))
140 ret = SCLD_EXISTS;
141 } else if (mkdir(path, 0777)) {
142 if (errno == EEXIST &&
143 !stat(path, &st) && S_ISDIR(st.st_mode))
144 ; /* somebody created it since we checked */
145 else if (errno == ENOENT)
146 /*
147 * Either mkdir() failed because
148 * somebody just pruned the containing
149 * directory, or stat() failed because
150 * the file that was in our way was
151 * just removed. Either way, inform
152 * the caller that it might be worth
153 * trying again:
154 */
155 ret = SCLD_VANISHED;
156 else
157 ret = SCLD_FAILED;
158 } else if (adjust_shared_perm(path)) {
159 ret = SCLD_PERMS;
160 }
161 *slash = slash_character;
162 }
163 return ret;
164 }
165
166 enum scld_error safe_create_leading_directories_const(const char *path)
167 {
168 /* path points to cache entries, so xstrdup before messing with it */
169 char *buf = xstrdup(path);
170 enum scld_error result = safe_create_leading_directories(buf);
171 free(buf);
172 return result;
173 }
174
175 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
176 {
177 int i;
178 for (i = 0; i < 20; i++) {
179 static char hex[] = "0123456789abcdef";
180 unsigned int val = sha1[i];
181 char *pos = pathbuf + i*2 + (i > 0);
182 *pos++ = hex[val >> 4];
183 *pos = hex[val & 0xf];
184 }
185 }
186
187 const char *sha1_file_name(const unsigned char *sha1)
188 {
189 static char buf[PATH_MAX];
190 const char *objdir;
191 int len;
192
193 objdir = get_object_directory();
194 len = strlen(objdir);
195
196 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
197 if (len + 43 > PATH_MAX)
198 die("insanely long object directory %s", objdir);
199 memcpy(buf, objdir, len);
200 buf[len] = '/';
201 buf[len+3] = '/';
202 buf[len+42] = '\0';
203 fill_sha1_path(buf + len + 1, sha1);
204 return buf;
205 }
206
207 /*
208 * Return the name of the pack or index file with the specified sha1
209 * in its filename. *base and *name are scratch space that must be
210 * provided by the caller. which should be "pack" or "idx".
211 */
212 static char *sha1_get_pack_name(const unsigned char *sha1,
213 struct strbuf *buf,
214 const char *which)
215 {
216 strbuf_reset(buf);
217 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
218 sha1_to_hex(sha1), which);
219 return buf->buf;
220 }
221
222 char *sha1_pack_name(const unsigned char *sha1)
223 {
224 static struct strbuf buf = STRBUF_INIT;
225 return sha1_get_pack_name(sha1, &buf, "pack");
226 }
227
228 char *sha1_pack_index_name(const unsigned char *sha1)
229 {
230 static struct strbuf buf = STRBUF_INIT;
231 return sha1_get_pack_name(sha1, &buf, "idx");
232 }
233
234 struct alternate_object_database *alt_odb_list;
235 static struct alternate_object_database **alt_odb_tail;
236
237 /*
238 * Prepare alternate object database registry.
239 *
240 * The variable alt_odb_list points at the list of struct
241 * alternate_object_database. The elements on this list come from
242 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
243 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
244 * whose contents is similar to that environment variable but can be
245 * LF separated. Its base points at a statically allocated buffer that
246 * contains "/the/directory/corresponding/to/.git/objects/...", while
247 * its name points just after the slash at the end of ".git/objects/"
248 * in the example above, and has enough space to hold 40-byte hex
249 * SHA1, an extra slash for the first level indirection, and the
250 * terminating NUL.
251 */
252 static int link_alt_odb_entry(const char *entry, const char *relative_base,
253 int depth, const char *normalized_objdir)
254 {
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 size_t pfxlen, entlen;
258 struct strbuf pathbuf = STRBUF_INIT;
259
260 if (!is_absolute_path(entry) && relative_base) {
261 strbuf_addstr(&pathbuf, real_path(relative_base));
262 strbuf_addch(&pathbuf, '/');
263 }
264 strbuf_addstr(&pathbuf, entry);
265
266 normalize_path_copy(pathbuf.buf, pathbuf.buf);
267
268 pfxlen = strlen(pathbuf.buf);
269
270 /*
271 * The trailing slash after the directory name is given by
272 * this function at the end. Remove duplicates.
273 */
274 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
275 pfxlen -= 1;
276
277 entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
278 ent = xmalloc(st_add(sizeof(*ent), entlen));
279 memcpy(ent->base, pathbuf.buf, pfxlen);
280 strbuf_release(&pathbuf);
281
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent->base)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
290 ent->base);
291 free(ent);
292 return -1;
293 }
294
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
297 */
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (pfxlen == alt->name - alt->base - 1 &&
300 !memcmp(ent->base, alt->base, pfxlen)) {
301 free(ent);
302 return -1;
303 }
304 }
305 if (!fspathcmp(ent->base, normalized_objdir)) {
306 free(ent);
307 return -1;
308 }
309
310 /* add the alternate entry */
311 *alt_odb_tail = ent;
312 alt_odb_tail = &(ent->next);
313 ent->next = NULL;
314
315 /* recursively add alternates */
316 read_info_alternates(ent->base, depth + 1);
317
318 ent->base[pfxlen] = '/';
319
320 return 0;
321 }
322
323 static void link_alt_odb_entries(const char *alt, int len, int sep,
324 const char *relative_base, int depth)
325 {
326 struct string_list entries = STRING_LIST_INIT_NODUP;
327 char *alt_copy;
328 int i;
329 struct strbuf objdirbuf = STRBUF_INIT;
330
331 if (depth > 5) {
332 error("%s: ignoring alternate object stores, nesting too deep.",
333 relative_base);
334 return;
335 }
336
337 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
338 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
339
340 alt_copy = xmemdupz(alt, len);
341 string_list_split_in_place(&entries, alt_copy, sep, -1);
342 for (i = 0; i < entries.nr; i++) {
343 const char *entry = entries.items[i].string;
344 if (entry[0] == '\0' || entry[0] == '#')
345 continue;
346 if (!is_absolute_path(entry) && depth) {
347 error("%s: ignoring relative alternate object store %s",
348 relative_base, entry);
349 } else {
350 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
351 }
352 }
353 string_list_clear(&entries, 0);
354 free(alt_copy);
355 strbuf_release(&objdirbuf);
356 }
357
358 void read_info_alternates(const char * relative_base, int depth)
359 {
360 char *map;
361 size_t mapsz;
362 struct stat st;
363 char *path;
364 int fd;
365
366 path = xstrfmt("%s/info/alternates", relative_base);
367 fd = git_open_noatime(path);
368 free(path);
369 if (fd < 0)
370 return;
371 if (fstat(fd, &st) || (st.st_size == 0)) {
372 close(fd);
373 return;
374 }
375 mapsz = xsize_t(st.st_size);
376 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
377 close(fd);
378
379 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
380
381 munmap(map, mapsz);
382 }
383
384 void add_to_alternates_file(const char *reference)
385 {
386 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
387 char *alts = git_pathdup("objects/info/alternates");
388 FILE *in, *out;
389
390 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
391 out = fdopen_lock_file(lock, "w");
392 if (!out)
393 die_errno("unable to fdopen alternates lockfile");
394
395 in = fopen(alts, "r");
396 if (in) {
397 struct strbuf line = STRBUF_INIT;
398 int found = 0;
399
400 while (strbuf_getline(&line, in) != EOF) {
401 if (!strcmp(reference, line.buf)) {
402 found = 1;
403 break;
404 }
405 fprintf_or_die(out, "%s\n", line.buf);
406 }
407
408 strbuf_release(&line);
409 fclose(in);
410
411 if (found) {
412 rollback_lock_file(lock);
413 lock = NULL;
414 }
415 }
416 else if (errno != ENOENT)
417 die_errno("unable to read alternates file");
418
419 if (lock) {
420 fprintf_or_die(out, "%s\n", reference);
421 if (commit_lock_file(lock))
422 die_errno("unable to move new alternates file into place");
423 if (alt_odb_tail)
424 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
425 }
426 free(alts);
427 }
428
429 /*
430 * Compute the exact path an alternate is at and returns it. In case of
431 * error NULL is returned and the human readable error is added to `err`
432 * `path` may be relative and should point to $GITDIR.
433 * `err` must not be null.
434 */
435 char *compute_alternate_path(const char *path, struct strbuf *err)
436 {
437 char *ref_git = NULL;
438 const char *repo, *ref_git_s;
439 int seen_error = 0;
440
441 ref_git_s = real_path_if_valid(path);
442 if (!ref_git_s) {
443 seen_error = 1;
444 strbuf_addf(err, _("path '%s' does not exist"), path);
445 goto out;
446 } else
447 /*
448 * Beware: read_gitfile(), real_path() and mkpath()
449 * return static buffer
450 */
451 ref_git = xstrdup(ref_git_s);
452
453 repo = read_gitfile(ref_git);
454 if (!repo)
455 repo = read_gitfile(mkpath("%s/.git", ref_git));
456 if (repo) {
457 free(ref_git);
458 ref_git = xstrdup(repo);
459 }
460
461 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
462 char *ref_git_git = mkpathdup("%s/.git", ref_git);
463 free(ref_git);
464 ref_git = ref_git_git;
465 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
466 struct strbuf sb = STRBUF_INIT;
467 seen_error = 1;
468 if (get_common_dir(&sb, ref_git)) {
469 strbuf_addf(err,
470 _("reference repository '%s' as a linked "
471 "checkout is not supported yet."),
472 path);
473 goto out;
474 }
475
476 strbuf_addf(err, _("reference repository '%s' is not a "
477 "local repository."), path);
478 goto out;
479 }
480
481 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
482 strbuf_addf(err, _("reference repository '%s' is shallow"),
483 path);
484 seen_error = 1;
485 goto out;
486 }
487
488 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
489 strbuf_addf(err,
490 _("reference repository '%s' is grafted"),
491 path);
492 seen_error = 1;
493 goto out;
494 }
495
496 out:
497 if (seen_error) {
498 free(ref_git);
499 ref_git = NULL;
500 }
501
502 return ref_git;
503 }
504
505 int foreach_alt_odb(alt_odb_fn fn, void *cb)
506 {
507 struct alternate_object_database *ent;
508 int r = 0;
509
510 prepare_alt_odb();
511 for (ent = alt_odb_list; ent; ent = ent->next) {
512 r = fn(ent, cb);
513 if (r)
514 break;
515 }
516 return r;
517 }
518
519 void prepare_alt_odb(void)
520 {
521 const char *alt;
522
523 if (alt_odb_tail)
524 return;
525
526 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
527 if (!alt) alt = "";
528
529 alt_odb_tail = &alt_odb_list;
530 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
531
532 read_info_alternates(get_object_directory(), 0);
533 }
534
535 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
536 static int freshen_file(const char *fn)
537 {
538 struct utimbuf t;
539 t.actime = t.modtime = time(NULL);
540 return !utime(fn, &t);
541 }
542
543 /*
544 * All of the check_and_freshen functions return 1 if the file exists and was
545 * freshened (if freshening was requested), 0 otherwise. If they return
546 * 0, you should not assume that it is safe to skip a write of the object (it
547 * either does not exist on disk, or has a stale mtime and may be subject to
548 * pruning).
549 */
550 static int check_and_freshen_file(const char *fn, int freshen)
551 {
552 if (access(fn, F_OK))
553 return 0;
554 if (freshen && !freshen_file(fn))
555 return 0;
556 return 1;
557 }
558
559 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
560 {
561 return check_and_freshen_file(sha1_file_name(sha1), freshen);
562 }
563
564 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
565 {
566 struct alternate_object_database *alt;
567 prepare_alt_odb();
568 for (alt = alt_odb_list; alt; alt = alt->next) {
569 fill_sha1_path(alt->name, sha1);
570 if (check_and_freshen_file(alt->base, freshen))
571 return 1;
572 }
573 return 0;
574 }
575
576 static int check_and_freshen(const unsigned char *sha1, int freshen)
577 {
578 return check_and_freshen_local(sha1, freshen) ||
579 check_and_freshen_nonlocal(sha1, freshen);
580 }
581
582 int has_loose_object_nonlocal(const unsigned char *sha1)
583 {
584 return check_and_freshen_nonlocal(sha1, 0);
585 }
586
587 static int has_loose_object(const unsigned char *sha1)
588 {
589 return check_and_freshen(sha1, 0);
590 }
591
592 static unsigned int pack_used_ctr;
593 static unsigned int pack_mmap_calls;
594 static unsigned int peak_pack_open_windows;
595 static unsigned int pack_open_windows;
596 static unsigned int pack_open_fds;
597 static unsigned int pack_max_fds;
598 static size_t peak_pack_mapped;
599 static size_t pack_mapped;
600 struct packed_git *packed_git;
601
602 static struct mru packed_git_mru_storage;
603 struct mru *packed_git_mru = &packed_git_mru_storage;
604
605 void pack_report(void)
606 {
607 fprintf(stderr,
608 "pack_report: getpagesize() = %10" SZ_FMT "\n"
609 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
610 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
611 sz_fmt(getpagesize()),
612 sz_fmt(packed_git_window_size),
613 sz_fmt(packed_git_limit));
614 fprintf(stderr,
615 "pack_report: pack_used_ctr = %10u\n"
616 "pack_report: pack_mmap_calls = %10u\n"
617 "pack_report: pack_open_windows = %10u / %10u\n"
618 "pack_report: pack_mapped = "
619 "%10" SZ_FMT " / %10" SZ_FMT "\n",
620 pack_used_ctr,
621 pack_mmap_calls,
622 pack_open_windows, peak_pack_open_windows,
623 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
624 }
625
626 /*
627 * Open and mmap the index file at path, perform a couple of
628 * consistency checks, then record its information to p. Return 0 on
629 * success.
630 */
631 static int check_packed_git_idx(const char *path, struct packed_git *p)
632 {
633 void *idx_map;
634 struct pack_idx_header *hdr;
635 size_t idx_size;
636 uint32_t version, nr, i, *index;
637 int fd = git_open_noatime(path);
638 struct stat st;
639
640 if (fd < 0)
641 return -1;
642 if (fstat(fd, &st)) {
643 close(fd);
644 return -1;
645 }
646 idx_size = xsize_t(st.st_size);
647 if (idx_size < 4 * 256 + 20 + 20) {
648 close(fd);
649 return error("index file %s is too small", path);
650 }
651 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
652 close(fd);
653
654 hdr = idx_map;
655 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
656 version = ntohl(hdr->idx_version);
657 if (version < 2 || version > 2) {
658 munmap(idx_map, idx_size);
659 return error("index file %s is version %"PRIu32
660 " and is not supported by this binary"
661 " (try upgrading GIT to a newer version)",
662 path, version);
663 }
664 } else
665 version = 1;
666
667 nr = 0;
668 index = idx_map;
669 if (version > 1)
670 index += 2; /* skip index header */
671 for (i = 0; i < 256; i++) {
672 uint32_t n = ntohl(index[i]);
673 if (n < nr) {
674 munmap(idx_map, idx_size);
675 return error("non-monotonic index %s", path);
676 }
677 nr = n;
678 }
679
680 if (version == 1) {
681 /*
682 * Total size:
683 * - 256 index entries 4 bytes each
684 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
685 * - 20-byte SHA1 of the packfile
686 * - 20-byte SHA1 file checksum
687 */
688 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
689 munmap(idx_map, idx_size);
690 return error("wrong index v1 file size in %s", path);
691 }
692 } else if (version == 2) {
693 /*
694 * Minimum size:
695 * - 8 bytes of header
696 * - 256 index entries 4 bytes each
697 * - 20-byte sha1 entry * nr
698 * - 4-byte crc entry * nr
699 * - 4-byte offset entry * nr
700 * - 20-byte SHA1 of the packfile
701 * - 20-byte SHA1 file checksum
702 * And after the 4-byte offset table might be a
703 * variable sized table containing 8-byte entries
704 * for offsets larger than 2^31.
705 */
706 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
707 unsigned long max_size = min_size;
708 if (nr)
709 max_size += (nr - 1)*8;
710 if (idx_size < min_size || idx_size > max_size) {
711 munmap(idx_map, idx_size);
712 return error("wrong index v2 file size in %s", path);
713 }
714 if (idx_size != min_size &&
715 /*
716 * make sure we can deal with large pack offsets.
717 * 31-bit signed offset won't be enough, neither
718 * 32-bit unsigned one will be.
719 */
720 (sizeof(off_t) <= 4)) {
721 munmap(idx_map, idx_size);
722 return error("pack too large for current definition of off_t in %s", path);
723 }
724 }
725
726 p->index_version = version;
727 p->index_data = idx_map;
728 p->index_size = idx_size;
729 p->num_objects = nr;
730 return 0;
731 }
732
733 int open_pack_index(struct packed_git *p)
734 {
735 char *idx_name;
736 size_t len;
737 int ret;
738
739 if (p->index_data)
740 return 0;
741
742 if (!strip_suffix(p->pack_name, ".pack", &len))
743 die("BUG: pack_name does not end in .pack");
744 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
745 ret = check_packed_git_idx(idx_name, p);
746 free(idx_name);
747 return ret;
748 }
749
750 static void scan_windows(struct packed_git *p,
751 struct packed_git **lru_p,
752 struct pack_window **lru_w,
753 struct pack_window **lru_l)
754 {
755 struct pack_window *w, *w_l;
756
757 for (w_l = NULL, w = p->windows; w; w = w->next) {
758 if (!w->inuse_cnt) {
759 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
760 *lru_p = p;
761 *lru_w = w;
762 *lru_l = w_l;
763 }
764 }
765 w_l = w;
766 }
767 }
768
769 static int unuse_one_window(struct packed_git *current)
770 {
771 struct packed_git *p, *lru_p = NULL;
772 struct pack_window *lru_w = NULL, *lru_l = NULL;
773
774 if (current)
775 scan_windows(current, &lru_p, &lru_w, &lru_l);
776 for (p = packed_git; p; p = p->next)
777 scan_windows(p, &lru_p, &lru_w, &lru_l);
778 if (lru_p) {
779 munmap(lru_w->base, lru_w->len);
780 pack_mapped -= lru_w->len;
781 if (lru_l)
782 lru_l->next = lru_w->next;
783 else
784 lru_p->windows = lru_w->next;
785 free(lru_w);
786 pack_open_windows--;
787 return 1;
788 }
789 return 0;
790 }
791
792 void release_pack_memory(size_t need)
793 {
794 size_t cur = pack_mapped;
795 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
796 ; /* nothing */
797 }
798
799 static void mmap_limit_check(size_t length)
800 {
801 static size_t limit = 0;
802 if (!limit) {
803 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
804 if (!limit)
805 limit = SIZE_MAX;
806 }
807 if (length > limit)
808 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
809 (uintmax_t)length, (uintmax_t)limit);
810 }
811
812 void *xmmap_gently(void *start, size_t length,
813 int prot, int flags, int fd, off_t offset)
814 {
815 void *ret;
816
817 mmap_limit_check(length);
818 ret = mmap(start, length, prot, flags, fd, offset);
819 if (ret == MAP_FAILED) {
820 if (!length)
821 return NULL;
822 release_pack_memory(length);
823 ret = mmap(start, length, prot, flags, fd, offset);
824 }
825 return ret;
826 }
827
828 void *xmmap(void *start, size_t length,
829 int prot, int flags, int fd, off_t offset)
830 {
831 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
832 if (ret == MAP_FAILED)
833 die_errno("mmap failed");
834 return ret;
835 }
836
837 void close_pack_windows(struct packed_git *p)
838 {
839 while (p->windows) {
840 struct pack_window *w = p->windows;
841
842 if (w->inuse_cnt)
843 die("pack '%s' still has open windows to it",
844 p->pack_name);
845 munmap(w->base, w->len);
846 pack_mapped -= w->len;
847 pack_open_windows--;
848 p->windows = w->next;
849 free(w);
850 }
851 }
852
853 static int close_pack_fd(struct packed_git *p)
854 {
855 if (p->pack_fd < 0)
856 return 0;
857
858 close(p->pack_fd);
859 pack_open_fds--;
860 p->pack_fd = -1;
861
862 return 1;
863 }
864
865 static void close_pack(struct packed_git *p)
866 {
867 close_pack_windows(p);
868 close_pack_fd(p);
869 close_pack_index(p);
870 }
871
872 void close_all_packs(void)
873 {
874 struct packed_git *p;
875
876 for (p = packed_git; p; p = p->next)
877 if (p->do_not_close)
878 die("BUG: want to close pack marked 'do-not-close'");
879 else
880 close_pack(p);
881 }
882
883
884 /*
885 * The LRU pack is the one with the oldest MRU window, preferring packs
886 * with no used windows, or the oldest mtime if it has no windows allocated.
887 */
888 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
889 {
890 struct pack_window *w, *this_mru_w;
891 int has_windows_inuse = 0;
892
893 /*
894 * Reject this pack if it has windows and the previously selected
895 * one does not. If this pack does not have windows, reject
896 * it if the pack file is newer than the previously selected one.
897 */
898 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
899 return;
900
901 for (w = this_mru_w = p->windows; w; w = w->next) {
902 /*
903 * Reject this pack if any of its windows are in use,
904 * but the previously selected pack did not have any
905 * inuse windows. Otherwise, record that this pack
906 * has windows in use.
907 */
908 if (w->inuse_cnt) {
909 if (*accept_windows_inuse)
910 has_windows_inuse = 1;
911 else
912 return;
913 }
914
915 if (w->last_used > this_mru_w->last_used)
916 this_mru_w = w;
917
918 /*
919 * Reject this pack if it has windows that have been
920 * used more recently than the previously selected pack.
921 * If the previously selected pack had windows inuse and
922 * we have not encountered a window in this pack that is
923 * inuse, skip this check since we prefer a pack with no
924 * inuse windows to one that has inuse windows.
925 */
926 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
927 this_mru_w->last_used > (*mru_w)->last_used)
928 return;
929 }
930
931 /*
932 * Select this pack.
933 */
934 *mru_w = this_mru_w;
935 *lru_p = p;
936 *accept_windows_inuse = has_windows_inuse;
937 }
938
939 static int close_one_pack(void)
940 {
941 struct packed_git *p, *lru_p = NULL;
942 struct pack_window *mru_w = NULL;
943 int accept_windows_inuse = 1;
944
945 for (p = packed_git; p; p = p->next) {
946 if (p->pack_fd == -1)
947 continue;
948 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
949 }
950
951 if (lru_p)
952 return close_pack_fd(lru_p);
953
954 return 0;
955 }
956
957 void unuse_pack(struct pack_window **w_cursor)
958 {
959 struct pack_window *w = *w_cursor;
960 if (w) {
961 w->inuse_cnt--;
962 *w_cursor = NULL;
963 }
964 }
965
966 void close_pack_index(struct packed_git *p)
967 {
968 if (p->index_data) {
969 munmap((void *)p->index_data, p->index_size);
970 p->index_data = NULL;
971 }
972 }
973
974 static unsigned int get_max_fd_limit(void)
975 {
976 #ifdef RLIMIT_NOFILE
977 {
978 struct rlimit lim;
979
980 if (!getrlimit(RLIMIT_NOFILE, &lim))
981 return lim.rlim_cur;
982 }
983 #endif
984
985 #ifdef _SC_OPEN_MAX
986 {
987 long open_max = sysconf(_SC_OPEN_MAX);
988 if (0 < open_max)
989 return open_max;
990 /*
991 * Otherwise, we got -1 for one of the two
992 * reasons:
993 *
994 * (1) sysconf() did not understand _SC_OPEN_MAX
995 * and signaled an error with -1; or
996 * (2) sysconf() said there is no limit.
997 *
998 * We _could_ clear errno before calling sysconf() to
999 * tell these two cases apart and return a huge number
1000 * in the latter case to let the caller cap it to a
1001 * value that is not so selfish, but letting the
1002 * fallback OPEN_MAX codepath take care of these cases
1003 * is a lot simpler.
1004 */
1005 }
1006 #endif
1007
1008 #ifdef OPEN_MAX
1009 return OPEN_MAX;
1010 #else
1011 return 1; /* see the caller ;-) */
1012 #endif
1013 }
1014
1015 /*
1016 * Do not call this directly as this leaks p->pack_fd on error return;
1017 * call open_packed_git() instead.
1018 */
1019 static int open_packed_git_1(struct packed_git *p)
1020 {
1021 struct stat st;
1022 struct pack_header hdr;
1023 unsigned char sha1[20];
1024 unsigned char *idx_sha1;
1025 long fd_flag;
1026
1027 if (!p->index_data && open_pack_index(p))
1028 return error("packfile %s index unavailable", p->pack_name);
1029
1030 if (!pack_max_fds) {
1031 unsigned int max_fds = get_max_fd_limit();
1032
1033 /* Save 3 for stdin/stdout/stderr, 22 for work */
1034 if (25 < max_fds)
1035 pack_max_fds = max_fds - 25;
1036 else
1037 pack_max_fds = 1;
1038 }
1039
1040 while (pack_max_fds <= pack_open_fds && close_one_pack())
1041 ; /* nothing */
1042
1043 p->pack_fd = git_open_noatime(p->pack_name);
1044 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1045 return -1;
1046 pack_open_fds++;
1047
1048 /* If we created the struct before we had the pack we lack size. */
1049 if (!p->pack_size) {
1050 if (!S_ISREG(st.st_mode))
1051 return error("packfile %s not a regular file", p->pack_name);
1052 p->pack_size = st.st_size;
1053 } else if (p->pack_size != st.st_size)
1054 return error("packfile %s size changed", p->pack_name);
1055
1056 /* We leave these file descriptors open with sliding mmap;
1057 * there is no point keeping them open across exec(), though.
1058 */
1059 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1060 if (fd_flag < 0)
1061 return error("cannot determine file descriptor flags");
1062 fd_flag |= FD_CLOEXEC;
1063 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1064 return error("cannot set FD_CLOEXEC");
1065
1066 /* Verify we recognize this pack file format. */
1067 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1068 return error("file %s is far too short to be a packfile", p->pack_name);
1069 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1070 return error("file %s is not a GIT packfile", p->pack_name);
1071 if (!pack_version_ok(hdr.hdr_version))
1072 return error("packfile %s is version %"PRIu32" and not"
1073 " supported (try upgrading GIT to a newer version)",
1074 p->pack_name, ntohl(hdr.hdr_version));
1075
1076 /* Verify the pack matches its index. */
1077 if (p->num_objects != ntohl(hdr.hdr_entries))
1078 return error("packfile %s claims to have %"PRIu32" objects"
1079 " while index indicates %"PRIu32" objects",
1080 p->pack_name, ntohl(hdr.hdr_entries),
1081 p->num_objects);
1082 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1083 return error("end of packfile %s is unavailable", p->pack_name);
1084 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1085 return error("packfile %s signature is unavailable", p->pack_name);
1086 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1087 if (hashcmp(sha1, idx_sha1))
1088 return error("packfile %s does not match index", p->pack_name);
1089 return 0;
1090 }
1091
1092 static int open_packed_git(struct packed_git *p)
1093 {
1094 if (!open_packed_git_1(p))
1095 return 0;
1096 close_pack_fd(p);
1097 return -1;
1098 }
1099
1100 static int in_window(struct pack_window *win, off_t offset)
1101 {
1102 /* We must promise at least 20 bytes (one hash) after the
1103 * offset is available from this window, otherwise the offset
1104 * is not actually in this window and a different window (which
1105 * has that one hash excess) must be used. This is to support
1106 * the object header and delta base parsing routines below.
1107 */
1108 off_t win_off = win->offset;
1109 return win_off <= offset
1110 && (offset + 20) <= (win_off + win->len);
1111 }
1112
1113 unsigned char *use_pack(struct packed_git *p,
1114 struct pack_window **w_cursor,
1115 off_t offset,
1116 unsigned long *left)
1117 {
1118 struct pack_window *win = *w_cursor;
1119
1120 /* Since packfiles end in a hash of their content and it's
1121 * pointless to ask for an offset into the middle of that
1122 * hash, and the in_window function above wouldn't match
1123 * don't allow an offset too close to the end of the file.
1124 */
1125 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1126 die("packfile %s cannot be accessed", p->pack_name);
1127 if (offset > (p->pack_size - 20))
1128 die("offset beyond end of packfile (truncated pack?)");
1129 if (offset < 0)
1130 die(_("offset before end of packfile (broken .idx?)"));
1131
1132 if (!win || !in_window(win, offset)) {
1133 if (win)
1134 win->inuse_cnt--;
1135 for (win = p->windows; win; win = win->next) {
1136 if (in_window(win, offset))
1137 break;
1138 }
1139 if (!win) {
1140 size_t window_align = packed_git_window_size / 2;
1141 off_t len;
1142
1143 if (p->pack_fd == -1 && open_packed_git(p))
1144 die("packfile %s cannot be accessed", p->pack_name);
1145
1146 win = xcalloc(1, sizeof(*win));
1147 win->offset = (offset / window_align) * window_align;
1148 len = p->pack_size - win->offset;
1149 if (len > packed_git_window_size)
1150 len = packed_git_window_size;
1151 win->len = (size_t)len;
1152 pack_mapped += win->len;
1153 while (packed_git_limit < pack_mapped
1154 && unuse_one_window(p))
1155 ; /* nothing */
1156 win->base = xmmap(NULL, win->len,
1157 PROT_READ, MAP_PRIVATE,
1158 p->pack_fd, win->offset);
1159 if (win->base == MAP_FAILED)
1160 die_errno("packfile %s cannot be mapped",
1161 p->pack_name);
1162 if (!win->offset && win->len == p->pack_size
1163 && !p->do_not_close)
1164 close_pack_fd(p);
1165 pack_mmap_calls++;
1166 pack_open_windows++;
1167 if (pack_mapped > peak_pack_mapped)
1168 peak_pack_mapped = pack_mapped;
1169 if (pack_open_windows > peak_pack_open_windows)
1170 peak_pack_open_windows = pack_open_windows;
1171 win->next = p->windows;
1172 p->windows = win;
1173 }
1174 }
1175 if (win != *w_cursor) {
1176 win->last_used = pack_used_ctr++;
1177 win->inuse_cnt++;
1178 *w_cursor = win;
1179 }
1180 offset -= win->offset;
1181 if (left)
1182 *left = win->len - xsize_t(offset);
1183 return win->base + offset;
1184 }
1185
1186 static struct packed_git *alloc_packed_git(int extra)
1187 {
1188 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1189 memset(p, 0, sizeof(*p));
1190 p->pack_fd = -1;
1191 return p;
1192 }
1193
1194 static void try_to_free_pack_memory(size_t size)
1195 {
1196 release_pack_memory(size);
1197 }
1198
1199 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1200 {
1201 static int have_set_try_to_free_routine;
1202 struct stat st;
1203 size_t alloc;
1204 struct packed_git *p;
1205
1206 if (!have_set_try_to_free_routine) {
1207 have_set_try_to_free_routine = 1;
1208 set_try_to_free_routine(try_to_free_pack_memory);
1209 }
1210
1211 /*
1212 * Make sure a corresponding .pack file exists and that
1213 * the index looks sane.
1214 */
1215 if (!strip_suffix_mem(path, &path_len, ".idx"))
1216 return NULL;
1217
1218 /*
1219 * ".pack" is long enough to hold any suffix we're adding (and
1220 * the use xsnprintf double-checks that)
1221 */
1222 alloc = st_add3(path_len, strlen(".pack"), 1);
1223 p = alloc_packed_git(alloc);
1224 memcpy(p->pack_name, path, path_len);
1225
1226 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1227 if (!access(p->pack_name, F_OK))
1228 p->pack_keep = 1;
1229
1230 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1231 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1232 free(p);
1233 return NULL;
1234 }
1235
1236 /* ok, it looks sane as far as we can check without
1237 * actually mapping the pack file.
1238 */
1239 p->pack_size = st.st_size;
1240 p->pack_local = local;
1241 p->mtime = st.st_mtime;
1242 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1243 hashclr(p->sha1);
1244 return p;
1245 }
1246
1247 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1248 {
1249 const char *path = sha1_pack_name(sha1);
1250 size_t alloc = st_add(strlen(path), 1);
1251 struct packed_git *p = alloc_packed_git(alloc);
1252
1253 memcpy(p->pack_name, path, alloc); /* includes NUL */
1254 hashcpy(p->sha1, sha1);
1255 if (check_packed_git_idx(idx_path, p)) {
1256 free(p);
1257 return NULL;
1258 }
1259
1260 return p;
1261 }
1262
1263 void install_packed_git(struct packed_git *pack)
1264 {
1265 if (pack->pack_fd != -1)
1266 pack_open_fds++;
1267
1268 pack->next = packed_git;
1269 packed_git = pack;
1270 }
1271
1272 void (*report_garbage)(unsigned seen_bits, const char *path);
1273
1274 static void report_helper(const struct string_list *list,
1275 int seen_bits, int first, int last)
1276 {
1277 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1278 return;
1279
1280 for (; first < last; first++)
1281 report_garbage(seen_bits, list->items[first].string);
1282 }
1283
1284 static void report_pack_garbage(struct string_list *list)
1285 {
1286 int i, baselen = -1, first = 0, seen_bits = 0;
1287
1288 if (!report_garbage)
1289 return;
1290
1291 string_list_sort(list);
1292
1293 for (i = 0; i < list->nr; i++) {
1294 const char *path = list->items[i].string;
1295 if (baselen != -1 &&
1296 strncmp(path, list->items[first].string, baselen)) {
1297 report_helper(list, seen_bits, first, i);
1298 baselen = -1;
1299 seen_bits = 0;
1300 }
1301 if (baselen == -1) {
1302 const char *dot = strrchr(path, '.');
1303 if (!dot) {
1304 report_garbage(PACKDIR_FILE_GARBAGE, path);
1305 continue;
1306 }
1307 baselen = dot - path + 1;
1308 first = i;
1309 }
1310 if (!strcmp(path + baselen, "pack"))
1311 seen_bits |= 1;
1312 else if (!strcmp(path + baselen, "idx"))
1313 seen_bits |= 2;
1314 }
1315 report_helper(list, seen_bits, first, list->nr);
1316 }
1317
1318 static void prepare_packed_git_one(char *objdir, int local)
1319 {
1320 struct strbuf path = STRBUF_INIT;
1321 size_t dirnamelen;
1322 DIR *dir;
1323 struct dirent *de;
1324 struct string_list garbage = STRING_LIST_INIT_DUP;
1325
1326 strbuf_addstr(&path, objdir);
1327 strbuf_addstr(&path, "/pack");
1328 dir = opendir(path.buf);
1329 if (!dir) {
1330 if (errno != ENOENT)
1331 error_errno("unable to open object pack directory: %s",
1332 path.buf);
1333 strbuf_release(&path);
1334 return;
1335 }
1336 strbuf_addch(&path, '/');
1337 dirnamelen = path.len;
1338 while ((de = readdir(dir)) != NULL) {
1339 struct packed_git *p;
1340 size_t base_len;
1341
1342 if (is_dot_or_dotdot(de->d_name))
1343 continue;
1344
1345 strbuf_setlen(&path, dirnamelen);
1346 strbuf_addstr(&path, de->d_name);
1347
1348 base_len = path.len;
1349 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1350 /* Don't reopen a pack we already have. */
1351 for (p = packed_git; p; p = p->next) {
1352 size_t len;
1353 if (strip_suffix(p->pack_name, ".pack", &len) &&
1354 len == base_len &&
1355 !memcmp(p->pack_name, path.buf, len))
1356 break;
1357 }
1358 if (p == NULL &&
1359 /*
1360 * See if it really is a valid .idx file with
1361 * corresponding .pack file that we can map.
1362 */
1363 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1364 install_packed_git(p);
1365 }
1366
1367 if (!report_garbage)
1368 continue;
1369
1370 if (ends_with(de->d_name, ".idx") ||
1371 ends_with(de->d_name, ".pack") ||
1372 ends_with(de->d_name, ".bitmap") ||
1373 ends_with(de->d_name, ".keep"))
1374 string_list_append(&garbage, path.buf);
1375 else
1376 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1377 }
1378 closedir(dir);
1379 report_pack_garbage(&garbage);
1380 string_list_clear(&garbage, 0);
1381 strbuf_release(&path);
1382 }
1383
1384 static void *get_next_packed_git(const void *p)
1385 {
1386 return ((const struct packed_git *)p)->next;
1387 }
1388
1389 static void set_next_packed_git(void *p, void *next)
1390 {
1391 ((struct packed_git *)p)->next = next;
1392 }
1393
1394 static int sort_pack(const void *a_, const void *b_)
1395 {
1396 const struct packed_git *a = a_;
1397 const struct packed_git *b = b_;
1398 int st;
1399
1400 /*
1401 * Local packs tend to contain objects specific to our
1402 * variant of the project than remote ones. In addition,
1403 * remote ones could be on a network mounted filesystem.
1404 * Favor local ones for these reasons.
1405 */
1406 st = a->pack_local - b->pack_local;
1407 if (st)
1408 return -st;
1409
1410 /*
1411 * Younger packs tend to contain more recent objects,
1412 * and more recent objects tend to get accessed more
1413 * often.
1414 */
1415 if (a->mtime < b->mtime)
1416 return 1;
1417 else if (a->mtime == b->mtime)
1418 return 0;
1419 return -1;
1420 }
1421
1422 static void rearrange_packed_git(void)
1423 {
1424 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1425 set_next_packed_git, sort_pack);
1426 }
1427
1428 static void prepare_packed_git_mru(void)
1429 {
1430 struct packed_git *p;
1431
1432 mru_clear(packed_git_mru);
1433 for (p = packed_git; p; p = p->next)
1434 mru_append(packed_git_mru, p);
1435 }
1436
1437 static int prepare_packed_git_run_once = 0;
1438 void prepare_packed_git(void)
1439 {
1440 struct alternate_object_database *alt;
1441
1442 if (prepare_packed_git_run_once)
1443 return;
1444 prepare_packed_git_one(get_object_directory(), 1);
1445 prepare_alt_odb();
1446 for (alt = alt_odb_list; alt; alt = alt->next) {
1447 alt->name[-1] = 0;
1448 prepare_packed_git_one(alt->base, 0);
1449 alt->name[-1] = '/';
1450 }
1451 rearrange_packed_git();
1452 prepare_packed_git_mru();
1453 prepare_packed_git_run_once = 1;
1454 }
1455
1456 void reprepare_packed_git(void)
1457 {
1458 prepare_packed_git_run_once = 0;
1459 prepare_packed_git();
1460 }
1461
1462 static void mark_bad_packed_object(struct packed_git *p,
1463 const unsigned char *sha1)
1464 {
1465 unsigned i;
1466 for (i = 0; i < p->num_bad_objects; i++)
1467 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1468 return;
1469 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1470 st_mult(GIT_SHA1_RAWSZ,
1471 st_add(p->num_bad_objects, 1)));
1472 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1473 p->num_bad_objects++;
1474 }
1475
1476 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1477 {
1478 struct packed_git *p;
1479 unsigned i;
1480
1481 for (p = packed_git; p; p = p->next)
1482 for (i = 0; i < p->num_bad_objects; i++)
1483 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1484 return p;
1485 return NULL;
1486 }
1487
1488 /*
1489 * With an in-core object data in "map", rehash it to make sure the
1490 * object name actually matches "sha1" to detect object corruption.
1491 * With "map" == NULL, try reading the object named with "sha1" using
1492 * the streaming interface and rehash it to do the same.
1493 */
1494 int check_sha1_signature(const unsigned char *sha1, void *map,
1495 unsigned long size, const char *type)
1496 {
1497 unsigned char real_sha1[20];
1498 enum object_type obj_type;
1499 struct git_istream *st;
1500 git_SHA_CTX c;
1501 char hdr[32];
1502 int hdrlen;
1503
1504 if (map) {
1505 hash_sha1_file(map, size, type, real_sha1);
1506 return hashcmp(sha1, real_sha1) ? -1 : 0;
1507 }
1508
1509 st = open_istream(sha1, &obj_type, &size, NULL);
1510 if (!st)
1511 return -1;
1512
1513 /* Generate the header */
1514 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1515
1516 /* Sha1.. */
1517 git_SHA1_Init(&c);
1518 git_SHA1_Update(&c, hdr, hdrlen);
1519 for (;;) {
1520 char buf[1024 * 16];
1521 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1522
1523 if (readlen < 0) {
1524 close_istream(st);
1525 return -1;
1526 }
1527 if (!readlen)
1528 break;
1529 git_SHA1_Update(&c, buf, readlen);
1530 }
1531 git_SHA1_Final(real_sha1, &c);
1532 close_istream(st);
1533 return hashcmp(sha1, real_sha1) ? -1 : 0;
1534 }
1535
1536 int git_open_noatime(const char *name)
1537 {
1538 static int sha1_file_open_flag = O_NOATIME;
1539
1540 for (;;) {
1541 int fd;
1542
1543 errno = 0;
1544 fd = open(name, O_RDONLY | sha1_file_open_flag);
1545 if (fd >= 0)
1546 return fd;
1547
1548 /* Might the failure be due to O_NOATIME? */
1549 if (errno != ENOENT && sha1_file_open_flag) {
1550 sha1_file_open_flag = 0;
1551 continue;
1552 }
1553
1554 return -1;
1555 }
1556 }
1557
1558 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1559 {
1560 struct alternate_object_database *alt;
1561
1562 if (!lstat(sha1_file_name(sha1), st))
1563 return 0;
1564
1565 prepare_alt_odb();
1566 errno = ENOENT;
1567 for (alt = alt_odb_list; alt; alt = alt->next) {
1568 fill_sha1_path(alt->name, sha1);
1569 if (!lstat(alt->base, st))
1570 return 0;
1571 }
1572
1573 return -1;
1574 }
1575
1576 static int open_sha1_file(const unsigned char *sha1)
1577 {
1578 int fd;
1579 struct alternate_object_database *alt;
1580 int most_interesting_errno;
1581
1582 fd = git_open_noatime(sha1_file_name(sha1));
1583 if (fd >= 0)
1584 return fd;
1585 most_interesting_errno = errno;
1586
1587 prepare_alt_odb();
1588 for (alt = alt_odb_list; alt; alt = alt->next) {
1589 fill_sha1_path(alt->name, sha1);
1590 fd = git_open_noatime(alt->base);
1591 if (fd >= 0)
1592 return fd;
1593 if (most_interesting_errno == ENOENT)
1594 most_interesting_errno = errno;
1595 }
1596 errno = most_interesting_errno;
1597 return -1;
1598 }
1599
1600 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1601 {
1602 void *map;
1603 int fd;
1604
1605 fd = open_sha1_file(sha1);
1606 map = NULL;
1607 if (fd >= 0) {
1608 struct stat st;
1609
1610 if (!fstat(fd, &st)) {
1611 *size = xsize_t(st.st_size);
1612 if (!*size) {
1613 /* mmap() is forbidden on empty files */
1614 error("object file %s is empty", sha1_file_name(sha1));
1615 return NULL;
1616 }
1617 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1618 }
1619 close(fd);
1620 }
1621 return map;
1622 }
1623
1624 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1625 unsigned long len, enum object_type *type, unsigned long *sizep)
1626 {
1627 unsigned shift;
1628 unsigned long size, c;
1629 unsigned long used = 0;
1630
1631 c = buf[used++];
1632 *type = (c >> 4) & 7;
1633 size = c & 15;
1634 shift = 4;
1635 while (c & 0x80) {
1636 if (len <= used || bitsizeof(long) <= shift) {
1637 error("bad object header");
1638 size = used = 0;
1639 break;
1640 }
1641 c = buf[used++];
1642 size += (c & 0x7f) << shift;
1643 shift += 7;
1644 }
1645 *sizep = size;
1646 return used;
1647 }
1648
1649 static int unpack_sha1_short_header(git_zstream *stream,
1650 unsigned char *map, unsigned long mapsize,
1651 void *buffer, unsigned long bufsiz)
1652 {
1653 /* Get the data stream */
1654 memset(stream, 0, sizeof(*stream));
1655 stream->next_in = map;
1656 stream->avail_in = mapsize;
1657 stream->next_out = buffer;
1658 stream->avail_out = bufsiz;
1659
1660 git_inflate_init(stream);
1661 return git_inflate(stream, 0);
1662 }
1663
1664 int unpack_sha1_header(git_zstream *stream,
1665 unsigned char *map, unsigned long mapsize,
1666 void *buffer, unsigned long bufsiz)
1667 {
1668 int status = unpack_sha1_short_header(stream, map, mapsize,
1669 buffer, bufsiz);
1670
1671 if (status < Z_OK)
1672 return status;
1673
1674 /* Make sure we have the terminating NUL */
1675 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1676 return -1;
1677 return 0;
1678 }
1679
1680 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1681 unsigned long mapsize, void *buffer,
1682 unsigned long bufsiz, struct strbuf *header)
1683 {
1684 int status;
1685
1686 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1687 if (status < Z_OK)
1688 return -1;
1689
1690 /*
1691 * Check if entire header is unpacked in the first iteration.
1692 */
1693 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1694 return 0;
1695
1696 /*
1697 * buffer[0..bufsiz] was not large enough. Copy the partial
1698 * result out to header, and then append the result of further
1699 * reading the stream.
1700 */
1701 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1702 stream->next_out = buffer;
1703 stream->avail_out = bufsiz;
1704
1705 do {
1706 status = git_inflate(stream, 0);
1707 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1708 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1709 return 0;
1710 stream->next_out = buffer;
1711 stream->avail_out = bufsiz;
1712 } while (status != Z_STREAM_END);
1713 return -1;
1714 }
1715
1716 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1717 {
1718 int bytes = strlen(buffer) + 1;
1719 unsigned char *buf = xmallocz(size);
1720 unsigned long n;
1721 int status = Z_OK;
1722
1723 n = stream->total_out - bytes;
1724 if (n > size)
1725 n = size;
1726 memcpy(buf, (char *) buffer + bytes, n);
1727 bytes = n;
1728 if (bytes <= size) {
1729 /*
1730 * The above condition must be (bytes <= size), not
1731 * (bytes < size). In other words, even though we
1732 * expect no more output and set avail_out to zero,
1733 * the input zlib stream may have bytes that express
1734 * "this concludes the stream", and we *do* want to
1735 * eat that input.
1736 *
1737 * Otherwise we would not be able to test that we
1738 * consumed all the input to reach the expected size;
1739 * we also want to check that zlib tells us that all
1740 * went well with status == Z_STREAM_END at the end.
1741 */
1742 stream->next_out = buf + bytes;
1743 stream->avail_out = size - bytes;
1744 while (status == Z_OK)
1745 status = git_inflate(stream, Z_FINISH);
1746 }
1747 if (status == Z_STREAM_END && !stream->avail_in) {
1748 git_inflate_end(stream);
1749 return buf;
1750 }
1751
1752 if (status < 0)
1753 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1754 else if (stream->avail_in)
1755 error("garbage at end of loose object '%s'",
1756 sha1_to_hex(sha1));
1757 free(buf);
1758 return NULL;
1759 }
1760
1761 /*
1762 * We used to just use "sscanf()", but that's actually way
1763 * too permissive for what we want to check. So do an anal
1764 * object header parse by hand.
1765 */
1766 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1767 unsigned int flags)
1768 {
1769 const char *type_buf = hdr;
1770 unsigned long size;
1771 int type, type_len = 0;
1772
1773 /*
1774 * The type can be of any size but is followed by
1775 * a space.
1776 */
1777 for (;;) {
1778 char c = *hdr++;
1779 if (!c)
1780 return -1;
1781 if (c == ' ')
1782 break;
1783 type_len++;
1784 }
1785
1786 type = type_from_string_gently(type_buf, type_len, 1);
1787 if (oi->typename)
1788 strbuf_add(oi->typename, type_buf, type_len);
1789 /*
1790 * Set type to 0 if its an unknown object and
1791 * we're obtaining the type using '--allow-unknown-type'
1792 * option.
1793 */
1794 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1795 type = 0;
1796 else if (type < 0)
1797 die("invalid object type");
1798 if (oi->typep)
1799 *oi->typep = type;
1800
1801 /*
1802 * The length must follow immediately, and be in canonical
1803 * decimal format (ie "010" is not valid).
1804 */
1805 size = *hdr++ - '0';
1806 if (size > 9)
1807 return -1;
1808 if (size) {
1809 for (;;) {
1810 unsigned long c = *hdr - '0';
1811 if (c > 9)
1812 break;
1813 hdr++;
1814 size = size * 10 + c;
1815 }
1816 }
1817
1818 if (oi->sizep)
1819 *oi->sizep = size;
1820
1821 /*
1822 * The length must be followed by a zero byte
1823 */
1824 return *hdr ? -1 : type;
1825 }
1826
1827 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1828 {
1829 struct object_info oi;
1830
1831 oi.sizep = sizep;
1832 oi.typename = NULL;
1833 oi.typep = NULL;
1834 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1835 }
1836
1837 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1838 {
1839 int ret;
1840 git_zstream stream;
1841 char hdr[8192];
1842
1843 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1844 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1845 return NULL;
1846
1847 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1848 }
1849
1850 unsigned long get_size_from_delta(struct packed_git *p,
1851 struct pack_window **w_curs,
1852 off_t curpos)
1853 {
1854 const unsigned char *data;
1855 unsigned char delta_head[20], *in;
1856 git_zstream stream;
1857 int st;
1858
1859 memset(&stream, 0, sizeof(stream));
1860 stream.next_out = delta_head;
1861 stream.avail_out = sizeof(delta_head);
1862
1863 git_inflate_init(&stream);
1864 do {
1865 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1866 stream.next_in = in;
1867 st = git_inflate(&stream, Z_FINISH);
1868 curpos += stream.next_in - in;
1869 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1870 stream.total_out < sizeof(delta_head));
1871 git_inflate_end(&stream);
1872 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1873 error("delta data unpack-initial failed");
1874 return 0;
1875 }
1876
1877 /* Examine the initial part of the delta to figure out
1878 * the result size.
1879 */
1880 data = delta_head;
1881
1882 /* ignore base size */
1883 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1884
1885 /* Read the result size */
1886 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1887 }
1888
1889 static off_t get_delta_base(struct packed_git *p,
1890 struct pack_window **w_curs,
1891 off_t *curpos,
1892 enum object_type type,
1893 off_t delta_obj_offset)
1894 {
1895 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1896 off_t base_offset;
1897
1898 /* use_pack() assured us we have [base_info, base_info + 20)
1899 * as a range that we can look at without walking off the
1900 * end of the mapped window. Its actually the hash size
1901 * that is assured. An OFS_DELTA longer than the hash size
1902 * is stupid, as then a REF_DELTA would be smaller to store.
1903 */
1904 if (type == OBJ_OFS_DELTA) {
1905 unsigned used = 0;
1906 unsigned char c = base_info[used++];
1907 base_offset = c & 127;
1908 while (c & 128) {
1909 base_offset += 1;
1910 if (!base_offset || MSB(base_offset, 7))
1911 return 0; /* overflow */
1912 c = base_info[used++];
1913 base_offset = (base_offset << 7) + (c & 127);
1914 }
1915 base_offset = delta_obj_offset - base_offset;
1916 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1917 return 0; /* out of bound */
1918 *curpos += used;
1919 } else if (type == OBJ_REF_DELTA) {
1920 /* The base entry _must_ be in the same pack */
1921 base_offset = find_pack_entry_one(base_info, p);
1922 *curpos += 20;
1923 } else
1924 die("I am totally screwed");
1925 return base_offset;
1926 }
1927
1928 /*
1929 * Like get_delta_base above, but we return the sha1 instead of the pack
1930 * offset. This means it is cheaper for REF deltas (we do not have to do
1931 * the final object lookup), but more expensive for OFS deltas (we
1932 * have to load the revidx to convert the offset back into a sha1).
1933 */
1934 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1935 struct pack_window **w_curs,
1936 off_t curpos,
1937 enum object_type type,
1938 off_t delta_obj_offset)
1939 {
1940 if (type == OBJ_REF_DELTA) {
1941 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1942 return base;
1943 } else if (type == OBJ_OFS_DELTA) {
1944 struct revindex_entry *revidx;
1945 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1946 type, delta_obj_offset);
1947
1948 if (!base_offset)
1949 return NULL;
1950
1951 revidx = find_pack_revindex(p, base_offset);
1952 if (!revidx)
1953 return NULL;
1954
1955 return nth_packed_object_sha1(p, revidx->nr);
1956 } else
1957 return NULL;
1958 }
1959
1960 int unpack_object_header(struct packed_git *p,
1961 struct pack_window **w_curs,
1962 off_t *curpos,
1963 unsigned long *sizep)
1964 {
1965 unsigned char *base;
1966 unsigned long left;
1967 unsigned long used;
1968 enum object_type type;
1969
1970 /* use_pack() assures us we have [base, base + 20) available
1971 * as a range that we can look at. (Its actually the hash
1972 * size that is assured.) With our object header encoding
1973 * the maximum deflated object size is 2^137, which is just
1974 * insane, so we know won't exceed what we have been given.
1975 */
1976 base = use_pack(p, w_curs, *curpos, &left);
1977 used = unpack_object_header_buffer(base, left, &type, sizep);
1978 if (!used) {
1979 type = OBJ_BAD;
1980 } else
1981 *curpos += used;
1982
1983 return type;
1984 }
1985
1986 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1987 {
1988 int type;
1989 struct revindex_entry *revidx;
1990 const unsigned char *sha1;
1991 revidx = find_pack_revindex(p, obj_offset);
1992 if (!revidx)
1993 return OBJ_BAD;
1994 sha1 = nth_packed_object_sha1(p, revidx->nr);
1995 mark_bad_packed_object(p, sha1);
1996 type = sha1_object_info(sha1, NULL);
1997 if (type <= OBJ_NONE)
1998 return OBJ_BAD;
1999 return type;
2000 }
2001
2002 #define POI_STACK_PREALLOC 64
2003
2004 static enum object_type packed_to_object_type(struct packed_git *p,
2005 off_t obj_offset,
2006 enum object_type type,
2007 struct pack_window **w_curs,
2008 off_t curpos)
2009 {
2010 off_t small_poi_stack[POI_STACK_PREALLOC];
2011 off_t *poi_stack = small_poi_stack;
2012 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2013
2014 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2015 off_t base_offset;
2016 unsigned long size;
2017 /* Push the object we're going to leave behind */
2018 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2019 poi_stack_alloc = alloc_nr(poi_stack_nr);
2020 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2021 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2022 } else {
2023 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2024 }
2025 poi_stack[poi_stack_nr++] = obj_offset;
2026 /* If parsing the base offset fails, just unwind */
2027 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2028 if (!base_offset)
2029 goto unwind;
2030 curpos = obj_offset = base_offset;
2031 type = unpack_object_header(p, w_curs, &curpos, &size);
2032 if (type <= OBJ_NONE) {
2033 /* If getting the base itself fails, we first
2034 * retry the base, otherwise unwind */
2035 type = retry_bad_packed_offset(p, base_offset);
2036 if (type > OBJ_NONE)
2037 goto out;
2038 goto unwind;
2039 }
2040 }
2041
2042 switch (type) {
2043 case OBJ_BAD:
2044 case OBJ_COMMIT:
2045 case OBJ_TREE:
2046 case OBJ_BLOB:
2047 case OBJ_TAG:
2048 break;
2049 default:
2050 error("unknown object type %i at offset %"PRIuMAX" in %s",
2051 type, (uintmax_t)obj_offset, p->pack_name);
2052 type = OBJ_BAD;
2053 }
2054
2055 out:
2056 if (poi_stack != small_poi_stack)
2057 free(poi_stack);
2058 return type;
2059
2060 unwind:
2061 while (poi_stack_nr) {
2062 obj_offset = poi_stack[--poi_stack_nr];
2063 type = retry_bad_packed_offset(p, obj_offset);
2064 if (type > OBJ_NONE)
2065 goto out;
2066 }
2067 type = OBJ_BAD;
2068 goto out;
2069 }
2070
2071 static int packed_object_info(struct packed_git *p, off_t obj_offset,
2072 struct object_info *oi)
2073 {
2074 struct pack_window *w_curs = NULL;
2075 unsigned long size;
2076 off_t curpos = obj_offset;
2077 enum object_type type;
2078
2079 /*
2080 * We always get the representation type, but only convert it to
2081 * a "real" type later if the caller is interested.
2082 */
2083 type = unpack_object_header(p, &w_curs, &curpos, &size);
2084
2085 if (oi->sizep) {
2086 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2087 off_t tmp_pos = curpos;
2088 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2089 type, obj_offset);
2090 if (!base_offset) {
2091 type = OBJ_BAD;
2092 goto out;
2093 }
2094 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2095 if (*oi->sizep == 0) {
2096 type = OBJ_BAD;
2097 goto out;
2098 }
2099 } else {
2100 *oi->sizep = size;
2101 }
2102 }
2103
2104 if (oi->disk_sizep) {
2105 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2106 *oi->disk_sizep = revidx[1].offset - obj_offset;
2107 }
2108
2109 if (oi->typep) {
2110 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2111 if (*oi->typep < 0) {
2112 type = OBJ_BAD;
2113 goto out;
2114 }
2115 }
2116
2117 if (oi->delta_base_sha1) {
2118 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2119 const unsigned char *base;
2120
2121 base = get_delta_base_sha1(p, &w_curs, curpos,
2122 type, obj_offset);
2123 if (!base) {
2124 type = OBJ_BAD;
2125 goto out;
2126 }
2127
2128 hashcpy(oi->delta_base_sha1, base);
2129 } else
2130 hashclr(oi->delta_base_sha1);
2131 }
2132
2133 out:
2134 unuse_pack(&w_curs);
2135 return type;
2136 }
2137
2138 static void *unpack_compressed_entry(struct packed_git *p,
2139 struct pack_window **w_curs,
2140 off_t curpos,
2141 unsigned long size)
2142 {
2143 int st;
2144 git_zstream stream;
2145 unsigned char *buffer, *in;
2146
2147 buffer = xmallocz_gently(size);
2148 if (!buffer)
2149 return NULL;
2150 memset(&stream, 0, sizeof(stream));
2151 stream.next_out = buffer;
2152 stream.avail_out = size + 1;
2153
2154 git_inflate_init(&stream);
2155 do {
2156 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2157 stream.next_in = in;
2158 st = git_inflate(&stream, Z_FINISH);
2159 if (!stream.avail_out)
2160 break; /* the payload is larger than it should be */
2161 curpos += stream.next_in - in;
2162 } while (st == Z_OK || st == Z_BUF_ERROR);
2163 git_inflate_end(&stream);
2164 if ((st != Z_STREAM_END) || stream.total_out != size) {
2165 free(buffer);
2166 return NULL;
2167 }
2168
2169 return buffer;
2170 }
2171
2172 static struct hashmap delta_base_cache;
2173 static size_t delta_base_cached;
2174
2175 static LIST_HEAD(delta_base_cache_lru);
2176
2177 struct delta_base_cache_key {
2178 struct packed_git *p;
2179 off_t base_offset;
2180 };
2181
2182 struct delta_base_cache_entry {
2183 struct hashmap hash;
2184 struct delta_base_cache_key key;
2185 struct list_head lru;
2186 void *data;
2187 unsigned long size;
2188 enum object_type type;
2189 };
2190
2191 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2192 {
2193 unsigned int hash;
2194
2195 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2196 hash += (hash >> 8) + (hash >> 16);
2197 return hash;
2198 }
2199
2200 static struct delta_base_cache_entry *
2201 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2202 {
2203 struct hashmap_entry entry;
2204 struct delta_base_cache_key key;
2205
2206 if (!delta_base_cache.cmpfn)
2207 return NULL;
2208
2209 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2210 key.p = p;
2211 key.base_offset = base_offset;
2212 return hashmap_get(&delta_base_cache, &entry, &key);
2213 }
2214
2215 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2216 const struct delta_base_cache_key *b)
2217 {
2218 return a->p == b->p && a->base_offset == b->base_offset;
2219 }
2220
2221 static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2222 const void *vkey)
2223 {
2224 const struct delta_base_cache_entry *a = va, *b = vb;
2225 const struct delta_base_cache_key *key = vkey;
2226 if (key)
2227 return !delta_base_cache_key_eq(&a->key, key);
2228 else
2229 return !delta_base_cache_key_eq(&a->key, &b->key);
2230 }
2231
2232 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2233 {
2234 return !!get_delta_base_cache_entry(p, base_offset);
2235 }
2236
2237 /*
2238 * Remove the entry from the cache, but do _not_ free the associated
2239 * entry data. The caller takes ownership of the "data" buffer, and
2240 * should copy out any fields it wants before detaching.
2241 */
2242 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2243 {
2244 hashmap_remove(&delta_base_cache, ent, &ent->key);
2245 list_del(&ent->lru);
2246 delta_base_cached -= ent->size;
2247 free(ent);
2248 }
2249
2250 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2251 unsigned long *base_size, enum object_type *type)
2252 {
2253 struct delta_base_cache_entry *ent;
2254
2255 ent = get_delta_base_cache_entry(p, base_offset);
2256 if (!ent)
2257 return unpack_entry(p, base_offset, type, base_size);
2258
2259 *type = ent->type;
2260 *base_size = ent->size;
2261 return xmemdupz(ent->data, ent->size);
2262 }
2263
2264 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2265 {
2266 free(ent->data);
2267 detach_delta_base_cache_entry(ent);
2268 }
2269
2270 void clear_delta_base_cache(void)
2271 {
2272 struct hashmap_iter iter;
2273 struct delta_base_cache_entry *entry;
2274 for (entry = hashmap_iter_first(&delta_base_cache, &iter);
2275 entry;
2276 entry = hashmap_iter_next(&iter)) {
2277 release_delta_base_cache(entry);
2278 }
2279 }
2280
2281 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2282 void *base, unsigned long base_size, enum object_type type)
2283 {
2284 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2285 struct list_head *lru, *tmp;
2286
2287 delta_base_cached += base_size;
2288
2289 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2290 struct delta_base_cache_entry *f =
2291 list_entry(lru, struct delta_base_cache_entry, lru);
2292 if (delta_base_cached <= delta_base_cache_limit)
2293 break;
2294 release_delta_base_cache(f);
2295 }
2296
2297 ent->key.p = p;
2298 ent->key.base_offset = base_offset;
2299 ent->type = type;
2300 ent->data = base;
2301 ent->size = base_size;
2302 list_add_tail(&ent->lru, &delta_base_cache_lru);
2303
2304 if (!delta_base_cache.cmpfn)
2305 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2306 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2307 hashmap_add(&delta_base_cache, ent);
2308 }
2309
2310 static void *read_object(const unsigned char *sha1, enum object_type *type,
2311 unsigned long *size);
2312
2313 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2314 {
2315 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2316 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2317 p->pack_name, (uintmax_t)obj_offset);
2318 }
2319
2320 int do_check_packed_object_crc;
2321
2322 #define UNPACK_ENTRY_STACK_PREALLOC 64
2323 struct unpack_entry_stack_ent {
2324 off_t obj_offset;
2325 off_t curpos;
2326 unsigned long size;
2327 };
2328
2329 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2330 enum object_type *final_type, unsigned long *final_size)
2331 {
2332 struct pack_window *w_curs = NULL;
2333 off_t curpos = obj_offset;
2334 void *data = NULL;
2335 unsigned long size;
2336 enum object_type type;
2337 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2338 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2339 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2340 int base_from_cache = 0;
2341
2342 write_pack_access_log(p, obj_offset);
2343
2344 /* PHASE 1: drill down to the innermost base object */
2345 for (;;) {
2346 off_t base_offset;
2347 int i;
2348 struct delta_base_cache_entry *ent;
2349
2350 ent = get_delta_base_cache_entry(p, curpos);
2351 if (ent) {
2352 type = ent->type;
2353 data = ent->data;
2354 size = ent->size;
2355 detach_delta_base_cache_entry(ent);
2356 base_from_cache = 1;
2357 break;
2358 }
2359
2360 if (do_check_packed_object_crc && p->index_version > 1) {
2361 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2362 off_t len = revidx[1].offset - obj_offset;
2363 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2364 const unsigned char *sha1 =
2365 nth_packed_object_sha1(p, revidx->nr);
2366 error("bad packed object CRC for %s",
2367 sha1_to_hex(sha1));
2368 mark_bad_packed_object(p, sha1);
2369 unuse_pack(&w_curs);
2370 return NULL;
2371 }
2372 }
2373
2374 type = unpack_object_header(p, &w_curs, &curpos, &size);
2375 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2376 break;
2377
2378 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2379 if (!base_offset) {
2380 error("failed to validate delta base reference "
2381 "at offset %"PRIuMAX" from %s",
2382 (uintmax_t)curpos, p->pack_name);
2383 /* bail to phase 2, in hopes of recovery */
2384 data = NULL;
2385 break;
2386 }
2387
2388 /* push object, proceed to base */
2389 if (delta_stack_nr >= delta_stack_alloc
2390 && delta_stack == small_delta_stack) {
2391 delta_stack_alloc = alloc_nr(delta_stack_nr);
2392 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2393 memcpy(delta_stack, small_delta_stack,
2394 sizeof(*delta_stack)*delta_stack_nr);
2395 } else {
2396 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2397 }
2398 i = delta_stack_nr++;
2399 delta_stack[i].obj_offset = obj_offset;
2400 delta_stack[i].curpos = curpos;
2401 delta_stack[i].size = size;
2402
2403 curpos = obj_offset = base_offset;
2404 }
2405
2406 /* PHASE 2: handle the base */
2407 switch (type) {
2408 case OBJ_OFS_DELTA:
2409 case OBJ_REF_DELTA:
2410 if (data)
2411 die("BUG: unpack_entry: left loop at a valid delta");
2412 break;
2413 case OBJ_COMMIT:
2414 case OBJ_TREE:
2415 case OBJ_BLOB:
2416 case OBJ_TAG:
2417 if (!base_from_cache)
2418 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2419 break;
2420 default:
2421 data = NULL;
2422 error("unknown object type %i at offset %"PRIuMAX" in %s",
2423 type, (uintmax_t)obj_offset, p->pack_name);
2424 }
2425
2426 /* PHASE 3: apply deltas in order */
2427
2428 /* invariants:
2429 * 'data' holds the base data, or NULL if there was corruption
2430 */
2431 while (delta_stack_nr) {
2432 void *delta_data;
2433 void *base = data;
2434 unsigned long delta_size, base_size = size;
2435 int i;
2436
2437 data = NULL;
2438
2439 if (base)
2440 add_delta_base_cache(p, obj_offset, base, base_size, type);
2441
2442 if (!base) {
2443 /*
2444 * We're probably in deep shit, but let's try to fetch
2445 * the required base anyway from another pack or loose.
2446 * This is costly but should happen only in the presence
2447 * of a corrupted pack, and is better than failing outright.
2448 */
2449 struct revindex_entry *revidx;
2450 const unsigned char *base_sha1;
2451 revidx = find_pack_revindex(p, obj_offset);
2452 if (revidx) {
2453 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2454 error("failed to read delta base object %s"
2455 " at offset %"PRIuMAX" from %s",
2456 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2457 p->pack_name);
2458 mark_bad_packed_object(p, base_sha1);
2459 base = read_object(base_sha1, &type, &base_size);
2460 }
2461 }
2462
2463 i = --delta_stack_nr;
2464 obj_offset = delta_stack[i].obj_offset;
2465 curpos = delta_stack[i].curpos;
2466 delta_size = delta_stack[i].size;
2467
2468 if (!base)
2469 continue;
2470
2471 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2472
2473 if (!delta_data) {
2474 error("failed to unpack compressed delta "
2475 "at offset %"PRIuMAX" from %s",
2476 (uintmax_t)curpos, p->pack_name);
2477 data = NULL;
2478 continue;
2479 }
2480
2481 data = patch_delta(base, base_size,
2482 delta_data, delta_size,
2483 &size);
2484
2485 /*
2486 * We could not apply the delta; warn the user, but keep going.
2487 * Our failure will be noticed either in the next iteration of
2488 * the loop, or if this is the final delta, in the caller when
2489 * we return NULL. Those code paths will take care of making
2490 * a more explicit warning and retrying with another copy of
2491 * the object.
2492 */
2493 if (!data)
2494 error("failed to apply delta");
2495
2496 free(delta_data);
2497 }
2498
2499 *final_type = type;
2500 *final_size = size;
2501
2502 unuse_pack(&w_curs);
2503
2504 if (delta_stack != small_delta_stack)
2505 free(delta_stack);
2506
2507 return data;
2508 }
2509
2510 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2511 uint32_t n)
2512 {
2513 const unsigned char *index = p->index_data;
2514 if (!index) {
2515 if (open_pack_index(p))
2516 return NULL;
2517 index = p->index_data;
2518 }
2519 if (n >= p->num_objects)
2520 return NULL;
2521 index += 4 * 256;
2522 if (p->index_version == 1) {
2523 return index + 24 * n + 4;
2524 } else {
2525 index += 8;
2526 return index + 20 * n;
2527 }
2528 }
2529
2530 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2531 {
2532 const unsigned char *ptr = vptr;
2533 const unsigned char *start = p->index_data;
2534 const unsigned char *end = start + p->index_size;
2535 if (ptr < start)
2536 die(_("offset before start of pack index for %s (corrupt index?)"),
2537 p->pack_name);
2538 /* No need to check for underflow; .idx files must be at least 8 bytes */
2539 if (ptr >= end - 8)
2540 die(_("offset beyond end of pack index for %s (truncated index?)"),
2541 p->pack_name);
2542 }
2543
2544 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2545 {
2546 const unsigned char *index = p->index_data;
2547 index += 4 * 256;
2548 if (p->index_version == 1) {
2549 return ntohl(*((uint32_t *)(index + 24 * n)));
2550 } else {
2551 uint32_t off;
2552 index += 8 + p->num_objects * (20 + 4);
2553 off = ntohl(*((uint32_t *)(index + 4 * n)));
2554 if (!(off & 0x80000000))
2555 return off;
2556 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2557 check_pack_index_ptr(p, index);
2558 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2559 ntohl(*((uint32_t *)(index + 4)));
2560 }
2561 }
2562
2563 off_t find_pack_entry_one(const unsigned char *sha1,
2564 struct packed_git *p)
2565 {
2566 const uint32_t *level1_ofs = p->index_data;
2567 const unsigned char *index = p->index_data;
2568 unsigned hi, lo, stride;
2569 static int use_lookup = -1;
2570 static int debug_lookup = -1;
2571
2572 if (debug_lookup < 0)
2573 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2574
2575 if (!index) {
2576 if (open_pack_index(p))
2577 return 0;
2578 level1_ofs = p->index_data;
2579 index = p->index_data;
2580 }
2581 if (p->index_version > 1) {
2582 level1_ofs += 2;
2583 index += 8;
2584 }
2585 index += 4 * 256;
2586 hi = ntohl(level1_ofs[*sha1]);
2587 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2588 if (p->index_version > 1) {
2589 stride = 20;
2590 } else {
2591 stride = 24;
2592 index += 4;
2593 }
2594
2595 if (debug_lookup)
2596 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2597 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2598
2599 if (use_lookup < 0)
2600 use_lookup = !!getenv("GIT_USE_LOOKUP");
2601 if (use_lookup) {
2602 int pos = sha1_entry_pos(index, stride, 0,
2603 lo, hi, p->num_objects, sha1);
2604 if (pos < 0)
2605 return 0;
2606 return nth_packed_object_offset(p, pos);
2607 }
2608
2609 do {
2610 unsigned mi = (lo + hi) / 2;
2611 int cmp = hashcmp(index + mi * stride, sha1);
2612
2613 if (debug_lookup)
2614 printf("lo %u hi %u rg %u mi %u\n",
2615 lo, hi, hi - lo, mi);
2616 if (!cmp)
2617 return nth_packed_object_offset(p, mi);
2618 if (cmp > 0)
2619 hi = mi;
2620 else
2621 lo = mi+1;
2622 } while (lo < hi);
2623 return 0;
2624 }
2625
2626 int is_pack_valid(struct packed_git *p)
2627 {
2628 /* An already open pack is known to be valid. */
2629 if (p->pack_fd != -1)
2630 return 1;
2631
2632 /* If the pack has one window completely covering the
2633 * file size, the pack is known to be valid even if
2634 * the descriptor is not currently open.
2635 */
2636 if (p->windows) {
2637 struct pack_window *w = p->windows;
2638
2639 if (!w->offset && w->len == p->pack_size)
2640 return 1;
2641 }
2642
2643 /* Force the pack to open to prove its valid. */
2644 return !open_packed_git(p);
2645 }
2646
2647 static int fill_pack_entry(const unsigned char *sha1,
2648 struct pack_entry *e,
2649 struct packed_git *p)
2650 {
2651 off_t offset;
2652
2653 if (p->num_bad_objects) {
2654 unsigned i;
2655 for (i = 0; i < p->num_bad_objects; i++)
2656 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2657 return 0;
2658 }
2659
2660 offset = find_pack_entry_one(sha1, p);
2661 if (!offset)
2662 return 0;
2663
2664 /*
2665 * We are about to tell the caller where they can locate the
2666 * requested object. We better make sure the packfile is
2667 * still here and can be accessed before supplying that
2668 * answer, as it may have been deleted since the index was
2669 * loaded!
2670 */
2671 if (!is_pack_valid(p))
2672 return 0;
2673 e->offset = offset;
2674 e->p = p;
2675 hashcpy(e->sha1, sha1);
2676 return 1;
2677 }
2678
2679 /*
2680 * Iff a pack file contains the object named by sha1, return true and
2681 * store its location to e.
2682 */
2683 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2684 {
2685 struct mru_entry *p;
2686
2687 prepare_packed_git();
2688 if (!packed_git)
2689 return 0;
2690
2691 for (p = packed_git_mru->head; p; p = p->next) {
2692 if (fill_pack_entry(sha1, e, p->item)) {
2693 mru_mark(packed_git_mru, p);
2694 return 1;
2695 }
2696 }
2697 return 0;
2698 }
2699
2700 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2701 struct packed_git *packs)
2702 {
2703 struct packed_git *p;
2704
2705 for (p = packs; p; p = p->next) {
2706 if (find_pack_entry_one(sha1, p))
2707 return p;
2708 }
2709 return NULL;
2710
2711 }
2712
2713 static int sha1_loose_object_info(const unsigned char *sha1,
2714 struct object_info *oi,
2715 int flags)
2716 {
2717 int status = 0;
2718 unsigned long mapsize;
2719 void *map;
2720 git_zstream stream;
2721 char hdr[32];
2722 struct strbuf hdrbuf = STRBUF_INIT;
2723
2724 if (oi->delta_base_sha1)
2725 hashclr(oi->delta_base_sha1);
2726
2727 /*
2728 * If we don't care about type or size, then we don't
2729 * need to look inside the object at all. Note that we
2730 * do not optimize out the stat call, even if the
2731 * caller doesn't care about the disk-size, since our
2732 * return value implicitly indicates whether the
2733 * object even exists.
2734 */
2735 if (!oi->typep && !oi->typename && !oi->sizep) {
2736 struct stat st;
2737 if (stat_sha1_file(sha1, &st) < 0)
2738 return -1;
2739 if (oi->disk_sizep)
2740 *oi->disk_sizep = st.st_size;
2741 return 0;
2742 }
2743
2744 map = map_sha1_file(sha1, &mapsize);
2745 if (!map)
2746 return -1;
2747 if (oi->disk_sizep)
2748 *oi->disk_sizep = mapsize;
2749 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2750 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2751 status = error("unable to unpack %s header with --allow-unknown-type",
2752 sha1_to_hex(sha1));
2753 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2754 status = error("unable to unpack %s header",
2755 sha1_to_hex(sha1));
2756 if (status < 0)
2757 ; /* Do nothing */
2758 else if (hdrbuf.len) {
2759 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2760 status = error("unable to parse %s header with --allow-unknown-type",
2761 sha1_to_hex(sha1));
2762 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2763 status = error("unable to parse %s header", sha1_to_hex(sha1));
2764 git_inflate_end(&stream);
2765 munmap(map, mapsize);
2766 if (status && oi->typep)
2767 *oi->typep = status;
2768 strbuf_release(&hdrbuf);
2769 return 0;
2770 }
2771
2772 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2773 {
2774 struct cached_object *co;
2775 struct pack_entry e;
2776 int rtype;
2777 enum object_type real_type;
2778 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2779
2780 co = find_cached_object(real);
2781 if (co) {
2782 if (oi->typep)
2783 *(oi->typep) = co->type;
2784 if (oi->sizep)
2785 *(oi->sizep) = co->size;
2786 if (oi->disk_sizep)
2787 *(oi->disk_sizep) = 0;
2788 if (oi->delta_base_sha1)
2789 hashclr(oi->delta_base_sha1);
2790 if (oi->typename)
2791 strbuf_addstr(oi->typename, typename(co->type));
2792 oi->whence = OI_CACHED;
2793 return 0;
2794 }
2795
2796 if (!find_pack_entry(real, &e)) {
2797 /* Most likely it's a loose object. */
2798 if (!sha1_loose_object_info(real, oi, flags)) {
2799 oi->whence = OI_LOOSE;
2800 return 0;
2801 }
2802
2803 /* Not a loose object; someone else may have just packed it. */
2804 reprepare_packed_git();
2805 if (!find_pack_entry(real, &e))
2806 return -1;
2807 }
2808
2809 /*
2810 * packed_object_info() does not follow the delta chain to
2811 * find out the real type, unless it is given oi->typep.
2812 */
2813 if (oi->typename && !oi->typep)
2814 oi->typep = &real_type;
2815
2816 rtype = packed_object_info(e.p, e.offset, oi);
2817 if (rtype < 0) {
2818 mark_bad_packed_object(e.p, real);
2819 if (oi->typep == &real_type)
2820 oi->typep = NULL;
2821 return sha1_object_info_extended(real, oi, 0);
2822 } else if (in_delta_base_cache(e.p, e.offset)) {
2823 oi->whence = OI_DBCACHED;
2824 } else {
2825 oi->whence = OI_PACKED;
2826 oi->u.packed.offset = e.offset;
2827 oi->u.packed.pack = e.p;
2828 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2829 rtype == OBJ_OFS_DELTA);
2830 }
2831 if (oi->typename)
2832 strbuf_addstr(oi->typename, typename(*oi->typep));
2833 if (oi->typep == &real_type)
2834 oi->typep = NULL;
2835
2836 return 0;
2837 }
2838
2839 /* returns enum object_type or negative */
2840 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2841 {
2842 enum object_type type;
2843 struct object_info oi = {NULL};
2844
2845 oi.typep = &type;
2846 oi.sizep = sizep;
2847 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2848 return -1;
2849 return type;
2850 }
2851
2852 static void *read_packed_sha1(const unsigned char *sha1,
2853 enum object_type *type, unsigned long *size)
2854 {
2855 struct pack_entry e;
2856 void *data;
2857
2858 if (!find_pack_entry(sha1, &e))
2859 return NULL;
2860 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2861 if (!data) {
2862 /*
2863 * We're probably in deep shit, but let's try to fetch
2864 * the required object anyway from another pack or loose.
2865 * This should happen only in the presence of a corrupted
2866 * pack, and is better than failing outright.
2867 */
2868 error("failed to read object %s at offset %"PRIuMAX" from %s",
2869 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2870 mark_bad_packed_object(e.p, sha1);
2871 data = read_object(sha1, type, size);
2872 }
2873 return data;
2874 }
2875
2876 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2877 unsigned char *sha1)
2878 {
2879 struct cached_object *co;
2880
2881 hash_sha1_file(buf, len, typename(type), sha1);
2882 if (has_sha1_file(sha1) || find_cached_object(sha1))
2883 return 0;
2884 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2885 co = &cached_objects[cached_object_nr++];
2886 co->size = len;
2887 co->type = type;
2888 co->buf = xmalloc(len);
2889 memcpy(co->buf, buf, len);
2890 hashcpy(co->sha1, sha1);
2891 return 0;
2892 }
2893
2894 static void *read_object(const unsigned char *sha1, enum object_type *type,
2895 unsigned long *size)
2896 {
2897 unsigned long mapsize;
2898 void *map, *buf;
2899 struct cached_object *co;
2900
2901 co = find_cached_object(sha1);
2902 if (co) {
2903 *type = co->type;
2904 *size = co->size;
2905 return xmemdupz(co->buf, co->size);
2906 }
2907
2908 buf = read_packed_sha1(sha1, type, size);
2909 if (buf)
2910 return buf;
2911 map = map_sha1_file(sha1, &mapsize);
2912 if (map) {
2913 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2914 munmap(map, mapsize);
2915 return buf;
2916 }
2917 reprepare_packed_git();
2918 return read_packed_sha1(sha1, type, size);
2919 }
2920
2921 /*
2922 * This function dies on corrupt objects; the callers who want to
2923 * deal with them should arrange to call read_object() and give error
2924 * messages themselves.
2925 */
2926 void *read_sha1_file_extended(const unsigned char *sha1,
2927 enum object_type *type,
2928 unsigned long *size,
2929 unsigned flag)
2930 {
2931 void *data;
2932 const struct packed_git *p;
2933 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2934
2935 errno = 0;
2936 data = read_object(repl, type, size);
2937 if (data)
2938 return data;
2939
2940 if (errno && errno != ENOENT)
2941 die_errno("failed to read object %s", sha1_to_hex(sha1));
2942
2943 /* die if we replaced an object with one that does not exist */
2944 if (repl != sha1)
2945 die("replacement %s not found for %s",
2946 sha1_to_hex(repl), sha1_to_hex(sha1));
2947
2948 if (has_loose_object(repl)) {
2949 const char *path = sha1_file_name(sha1);
2950
2951 die("loose object %s (stored in %s) is corrupt",
2952 sha1_to_hex(repl), path);
2953 }
2954
2955 if ((p = has_packed_and_bad(repl)) != NULL)
2956 die("packed object %s (stored in %s) is corrupt",
2957 sha1_to_hex(repl), p->pack_name);
2958
2959 return NULL;
2960 }
2961
2962 void *read_object_with_reference(const unsigned char *sha1,
2963 const char *required_type_name,
2964 unsigned long *size,
2965 unsigned char *actual_sha1_return)
2966 {
2967 enum object_type type, required_type;
2968 void *buffer;
2969 unsigned long isize;
2970 unsigned char actual_sha1[20];
2971
2972 required_type = type_from_string(required_type_name);
2973 hashcpy(actual_sha1, sha1);
2974 while (1) {
2975 int ref_length = -1;
2976 const char *ref_type = NULL;
2977
2978 buffer = read_sha1_file(actual_sha1, &type, &isize);
2979 if (!buffer)
2980 return NULL;
2981 if (type == required_type) {
2982 *size = isize;
2983 if (actual_sha1_return)
2984 hashcpy(actual_sha1_return, actual_sha1);
2985 return buffer;
2986 }
2987 /* Handle references */
2988 else if (type == OBJ_COMMIT)
2989 ref_type = "tree ";
2990 else if (type == OBJ_TAG)
2991 ref_type = "object ";
2992 else {
2993 free(buffer);
2994 return NULL;
2995 }
2996 ref_length = strlen(ref_type);
2997
2998 if (ref_length + 40 > isize ||
2999 memcmp(buffer, ref_type, ref_length) ||
3000 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3001 free(buffer);
3002 return NULL;
3003 }
3004 free(buffer);
3005 /* Now we have the ID of the referred-to object in
3006 * actual_sha1. Check again. */
3007 }
3008 }
3009
3010 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3011 const char *type, unsigned char *sha1,
3012 char *hdr, int *hdrlen)
3013 {
3014 git_SHA_CTX c;
3015
3016 /* Generate the header */
3017 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3018
3019 /* Sha1.. */
3020 git_SHA1_Init(&c);
3021 git_SHA1_Update(&c, hdr, *hdrlen);
3022 git_SHA1_Update(&c, buf, len);
3023 git_SHA1_Final(sha1, &c);
3024 }
3025
3026 /*
3027 * Move the just written object into its final resting place.
3028 */
3029 int finalize_object_file(const char *tmpfile, const char *filename)
3030 {
3031 int ret = 0;
3032
3033 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3034 goto try_rename;
3035 else if (link(tmpfile, filename))
3036 ret = errno;
3037
3038 /*
3039 * Coda hack - coda doesn't like cross-directory links,
3040 * so we fall back to a rename, which will mean that it
3041 * won't be able to check collisions, but that's not a
3042 * big deal.
3043 *
3044 * The same holds for FAT formatted media.
3045 *
3046 * When this succeeds, we just return. We have nothing
3047 * left to unlink.
3048 */
3049 if (ret && ret != EEXIST) {
3050 try_rename:
3051 if (!rename(tmpfile, filename))
3052 goto out;
3053 ret = errno;
3054 }
3055 unlink_or_warn(tmpfile);
3056 if (ret) {
3057 if (ret != EEXIST) {
3058 return error_errno("unable to write sha1 filename %s", filename);
3059 }
3060 /* FIXME!!! Collision check here ? */
3061 }
3062
3063 out:
3064 if (adjust_shared_perm(filename))
3065 return error("unable to set permission to '%s'", filename);
3066 return 0;
3067 }
3068
3069 static int write_buffer(int fd, const void *buf, size_t len)
3070 {
3071 if (write_in_full(fd, buf, len) < 0)
3072 return error_errno("file write error");
3073 return 0;
3074 }
3075
3076 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3077 unsigned char *sha1)
3078 {
3079 char hdr[32];
3080 int hdrlen = sizeof(hdr);
3081 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3082 return 0;
3083 }
3084
3085 /* Finalize a file on disk, and close it. */
3086 static void close_sha1_file(int fd)
3087 {
3088 if (fsync_object_files)
3089 fsync_or_die(fd, "sha1 file");
3090 if (close(fd) != 0)
3091 die_errno("error when closing sha1 file");
3092 }
3093
3094 /* Size of directory component, including the ending '/' */
3095 static inline int directory_size(const char *filename)
3096 {
3097 const char *s = strrchr(filename, '/');
3098 if (!s)
3099 return 0;
3100 return s - filename + 1;
3101 }
3102
3103 /*
3104 * This creates a temporary file in the same directory as the final
3105 * 'filename'
3106 *
3107 * We want to avoid cross-directory filename renames, because those
3108 * can have problems on various filesystems (FAT, NFS, Coda).
3109 */
3110 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3111 {
3112 int fd, dirlen = directory_size(filename);
3113
3114 strbuf_reset(tmp);
3115 strbuf_add(tmp, filename, dirlen);
3116 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3117 fd = git_mkstemp_mode(tmp->buf, 0444);
3118 if (fd < 0 && dirlen && errno == ENOENT) {
3119 /*
3120 * Make sure the directory exists; note that the contents
3121 * of the buffer are undefined after mkstemp returns an
3122 * error, so we have to rewrite the whole buffer from
3123 * scratch.
3124 */
3125 strbuf_reset(tmp);
3126 strbuf_add(tmp, filename, dirlen - 1);
3127 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3128 return -1;
3129 if (adjust_shared_perm(tmp->buf))
3130 return -1;
3131
3132 /* Try again */
3133 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3134 fd = git_mkstemp_mode(tmp->buf, 0444);
3135 }
3136 return fd;
3137 }
3138
3139 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3140 const void *buf, unsigned long len, time_t mtime)
3141 {
3142 int fd, ret;
3143 unsigned char compressed[4096];
3144 git_zstream stream;
3145 git_SHA_CTX c;
3146 unsigned char parano_sha1[20];
3147 static struct strbuf tmp_file = STRBUF_INIT;
3148 const char *filename = sha1_file_name(sha1);
3149
3150 fd = create_tmpfile(&tmp_file, filename);
3151 if (fd < 0) {
3152 if (errno == EACCES)
3153 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3154 else
3155 return error_errno("unable to create temporary file");
3156 }
3157
3158 /* Set it up */
3159 git_deflate_init(&stream, zlib_compression_level);
3160 stream.next_out = compressed;
3161 stream.avail_out = sizeof(compressed);
3162 git_SHA1_Init(&c);
3163
3164 /* First header.. */
3165 stream.next_in = (unsigned char *)hdr;
3166 stream.avail_in = hdrlen;
3167 while (git_deflate(&stream, 0) == Z_OK)
3168 ; /* nothing */
3169 git_SHA1_Update(&c, hdr, hdrlen);
3170
3171 /* Then the data itself.. */
3172 stream.next_in = (void *)buf;
3173 stream.avail_in = len;
3174 do {
3175 unsigned char *in0 = stream.next_in;
3176 ret = git_deflate(&stream, Z_FINISH);
3177 git_SHA1_Update(&c, in0, stream.next_in - in0);
3178 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3179 die("unable to write sha1 file");
3180 stream.next_out = compressed;
3181 stream.avail_out = sizeof(compressed);
3182 } while (ret == Z_OK);
3183
3184 if (ret != Z_STREAM_END)
3185 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3186 ret = git_deflate_end_gently(&stream);
3187 if (ret != Z_OK)
3188 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3189 git_SHA1_Final(parano_sha1, &c);
3190 if (hashcmp(sha1, parano_sha1) != 0)
3191 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3192
3193 close_sha1_file(fd);
3194
3195 if (mtime) {
3196 struct utimbuf utb;
3197 utb.actime = mtime;
3198 utb.modtime = mtime;
3199 if (utime(tmp_file.buf, &utb) < 0)
3200 warning_errno("failed utime() on %s", tmp_file.buf);
3201 }
3202
3203 return finalize_object_file(tmp_file.buf, filename);
3204 }
3205
3206 static int freshen_loose_object(const unsigned char *sha1)
3207 {
3208 return check_and_freshen(sha1, 1);
3209 }
3210
3211 static int freshen_packed_object(const unsigned char *sha1)
3212 {
3213 struct pack_entry e;
3214 if (!find_pack_entry(sha1, &e))
3215 return 0;
3216 if (e.p->freshened)
3217 return 1;
3218 if (!freshen_file(e.p->pack_name))
3219 return 0;
3220 e.p->freshened = 1;
3221 return 1;
3222 }
3223
3224 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3225 {
3226 char hdr[32];
3227 int hdrlen = sizeof(hdr);
3228
3229 /* Normally if we have it in the pack then we do not bother writing
3230 * it out into .git/objects/??/?{38} file.
3231 */
3232 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3233 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3234 return 0;
3235 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3236 }
3237
3238 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3239 unsigned char *sha1, unsigned flags)
3240 {
3241 char *header;
3242 int hdrlen, status = 0;
3243
3244 /* type string, SP, %lu of the length plus NUL must fit this */
3245 hdrlen = strlen(type) + 32;
3246 header = xmalloc(hdrlen);
3247 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3248
3249 if (!(flags & HASH_WRITE_OBJECT))
3250 goto cleanup;
3251 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3252 goto cleanup;
3253 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3254
3255 cleanup:
3256 free(header);
3257 return status;
3258 }
3259
3260 int force_object_loose(const unsigned char *sha1, time_t mtime)
3261 {
3262 void *buf;
3263 unsigned long len;
3264 enum object_type type;
3265 char hdr[32];
3266 int hdrlen;
3267 int ret;
3268
3269 if (has_loose_object(sha1))
3270 return 0;
3271 buf = read_packed_sha1(sha1, &type, &len);
3272 if (!buf)
3273 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3274 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3275 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3276 free(buf);
3277
3278 return ret;
3279 }
3280
3281 int has_pack_index(const unsigned char *sha1)
3282 {
3283 struct stat st;
3284 if (stat(sha1_pack_index_name(sha1), &st))
3285 return 0;
3286 return 1;
3287 }
3288
3289 int has_sha1_pack(const unsigned char *sha1)
3290 {
3291 struct pack_entry e;
3292 return find_pack_entry(sha1, &e);
3293 }
3294
3295 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3296 {
3297 struct pack_entry e;
3298
3299 if (find_pack_entry(sha1, &e))
3300 return 1;
3301 if (has_loose_object(sha1))
3302 return 1;
3303 if (flags & HAS_SHA1_QUICK)
3304 return 0;
3305 reprepare_packed_git();
3306 return find_pack_entry(sha1, &e);
3307 }
3308
3309 int has_object_file(const struct object_id *oid)
3310 {
3311 return has_sha1_file(oid->hash);
3312 }
3313
3314 static void check_tree(const void *buf, size_t size)
3315 {
3316 struct tree_desc desc;
3317 struct name_entry entry;
3318
3319 init_tree_desc(&desc, buf, size);
3320 while (tree_entry(&desc, &entry))
3321 /* do nothing
3322 * tree_entry() will die() on malformed entries */
3323 ;
3324 }
3325
3326 static void check_commit(const void *buf, size_t size)
3327 {
3328 struct commit c;
3329 memset(&c, 0, sizeof(c));
3330 if (parse_commit_buffer(&c, buf, size))
3331 die("corrupt commit");
3332 }
3333
3334 static void check_tag(const void *buf, size_t size)
3335 {
3336 struct tag t;
3337 memset(&t, 0, sizeof(t));
3338 if (parse_tag_buffer(&t, buf, size))
3339 die("corrupt tag");
3340 }
3341
3342 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3343 enum object_type type,
3344 const char *path, unsigned flags)
3345 {
3346 int ret, re_allocated = 0;
3347 int write_object = flags & HASH_WRITE_OBJECT;
3348
3349 if (!type)
3350 type = OBJ_BLOB;
3351
3352 /*
3353 * Convert blobs to git internal format
3354 */
3355 if ((type == OBJ_BLOB) && path) {
3356 struct strbuf nbuf = STRBUF_INIT;
3357 if (convert_to_git(path, buf, size, &nbuf,
3358 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3359 buf = strbuf_detach(&nbuf, &size);
3360 re_allocated = 1;
3361 }
3362 }
3363 if (flags & HASH_FORMAT_CHECK) {
3364 if (type == OBJ_TREE)
3365 check_tree(buf, size);
3366 if (type == OBJ_COMMIT)
3367 check_commit(buf, size);
3368 if (type == OBJ_TAG)
3369 check_tag(buf, size);
3370 }
3371
3372 if (write_object)
3373 ret = write_sha1_file(buf, size, typename(type), sha1);
3374 else
3375 ret = hash_sha1_file(buf, size, typename(type), sha1);
3376 if (re_allocated)
3377 free(buf);
3378 return ret;
3379 }
3380
3381 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3382 const char *path, unsigned flags)
3383 {
3384 int ret;
3385 const int write_object = flags & HASH_WRITE_OBJECT;
3386 struct strbuf sbuf = STRBUF_INIT;
3387
3388 assert(path);
3389 assert(would_convert_to_git_filter_fd(path));
3390
3391 convert_to_git_filter_fd(path, fd, &sbuf,
3392 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3393
3394 if (write_object)
3395 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3396 sha1);
3397 else
3398 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3399 sha1);
3400 strbuf_release(&sbuf);
3401 return ret;
3402 }
3403
3404 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3405 const char *path, unsigned flags)
3406 {
3407 struct strbuf sbuf = STRBUF_INIT;
3408 int ret;
3409
3410 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3411 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3412 else
3413 ret = -1;
3414 strbuf_release(&sbuf);
3415 return ret;
3416 }
3417
3418 #define SMALL_FILE_SIZE (32*1024)
3419
3420 static int index_core(unsigned char *sha1, int fd, size_t size,
3421 enum object_type type, const char *path,
3422 unsigned flags)
3423 {
3424 int ret;
3425
3426 if (!size) {
3427 ret = index_mem(sha1, "", size, type, path, flags);
3428 } else if (size <= SMALL_FILE_SIZE) {
3429 char *buf = xmalloc(size);
3430 if (size == read_in_full(fd, buf, size))
3431 ret = index_mem(sha1, buf, size, type, path, flags);
3432 else
3433 ret = error_errno("short read");
3434 free(buf);
3435 } else {
3436 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3437 ret = index_mem(sha1, buf, size, type, path, flags);
3438 munmap(buf, size);
3439 }
3440 return ret;
3441 }
3442
3443 /*
3444 * This creates one packfile per large blob unless bulk-checkin
3445 * machinery is "plugged".
3446 *
3447 * This also bypasses the usual "convert-to-git" dance, and that is on
3448 * purpose. We could write a streaming version of the converting
3449 * functions and insert that before feeding the data to fast-import
3450 * (or equivalent in-core API described above). However, that is
3451 * somewhat complicated, as we do not know the size of the filter
3452 * result, which we need to know beforehand when writing a git object.
3453 * Since the primary motivation for trying to stream from the working
3454 * tree file and to avoid mmaping it in core is to deal with large
3455 * binary blobs, they generally do not want to get any conversion, and
3456 * callers should avoid this code path when filters are requested.
3457 */
3458 static int index_stream(unsigned char *sha1, int fd, size_t size,
3459 enum object_type type, const char *path,
3460 unsigned flags)
3461 {
3462 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3463 }
3464
3465 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3466 enum object_type type, const char *path, unsigned flags)
3467 {
3468 int ret;
3469
3470 /*
3471 * Call xsize_t() only when needed to avoid potentially unnecessary
3472 * die() for large files.
3473 */
3474 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3475 ret = index_stream_convert_blob(sha1, fd, path, flags);
3476 else if (!S_ISREG(st->st_mode))
3477 ret = index_pipe(sha1, fd, type, path, flags);
3478 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3479 (path && would_convert_to_git(path)))
3480 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3481 flags);
3482 else
3483 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3484 flags);
3485 close(fd);
3486 return ret;
3487 }
3488
3489 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3490 {
3491 int fd;
3492 struct strbuf sb = STRBUF_INIT;
3493
3494 switch (st->st_mode & S_IFMT) {
3495 case S_IFREG:
3496 fd = open(path, O_RDONLY);
3497 if (fd < 0)
3498 return error_errno("open(\"%s\")", path);
3499 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3500 return error("%s: failed to insert into database",
3501 path);
3502 break;
3503 case S_IFLNK:
3504 if (strbuf_readlink(&sb, path, st->st_size))
3505 return error_errno("readlink(\"%s\")", path);
3506 if (!(flags & HASH_WRITE_OBJECT))
3507 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3508 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3509 return error("%s: failed to insert into database",
3510 path);
3511 strbuf_release(&sb);
3512 break;
3513 case S_IFDIR:
3514 return resolve_gitlink_ref(path, "HEAD", sha1);
3515 default:
3516 return error("%s: unsupported file type", path);
3517 }
3518 return 0;
3519 }
3520
3521 int read_pack_header(int fd, struct pack_header *header)
3522 {
3523 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3524 /* "eof before pack header was fully read" */
3525 return PH_ERROR_EOF;
3526
3527 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3528 /* "protocol error (pack signature mismatch detected)" */
3529 return PH_ERROR_PACK_SIGNATURE;
3530 if (!pack_version_ok(header->hdr_version))
3531 /* "protocol error (pack version unsupported)" */
3532 return PH_ERROR_PROTOCOL;
3533 return 0;
3534 }
3535
3536 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3537 {
3538 enum object_type type = sha1_object_info(sha1, NULL);
3539 if (type < 0)
3540 die("%s is not a valid object", sha1_to_hex(sha1));
3541 if (type != expect)
3542 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3543 typename(expect));
3544 }
3545
3546 static int for_each_file_in_obj_subdir(int subdir_nr,
3547 struct strbuf *path,
3548 each_loose_object_fn obj_cb,
3549 each_loose_cruft_fn cruft_cb,
3550 each_loose_subdir_fn subdir_cb,
3551 void *data)
3552 {
3553 size_t baselen = path->len;
3554 DIR *dir = opendir(path->buf);
3555 struct dirent *de;
3556 int r = 0;
3557
3558 if (!dir) {
3559 if (errno == ENOENT)
3560 return 0;
3561 return error_errno("unable to open %s", path->buf);
3562 }
3563
3564 while ((de = readdir(dir))) {
3565 if (is_dot_or_dotdot(de->d_name))
3566 continue;
3567
3568 strbuf_setlen(path, baselen);
3569 strbuf_addf(path, "/%s", de->d_name);
3570
3571 if (strlen(de->d_name) == 38) {
3572 char hex[41];
3573 unsigned char sha1[20];
3574
3575 snprintf(hex, sizeof(hex), "%02x%s",
3576 subdir_nr, de->d_name);
3577 if (!get_sha1_hex(hex, sha1)) {
3578 if (obj_cb) {
3579 r = obj_cb(sha1, path->buf, data);
3580 if (r)
3581 break;
3582 }
3583 continue;
3584 }
3585 }
3586
3587 if (cruft_cb) {
3588 r = cruft_cb(de->d_name, path->buf, data);
3589 if (r)
3590 break;
3591 }
3592 }
3593 closedir(dir);
3594
3595 strbuf_setlen(path, baselen);
3596 if (!r && subdir_cb)
3597 r = subdir_cb(subdir_nr, path->buf, data);
3598
3599 return r;
3600 }
3601
3602 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3603 each_loose_object_fn obj_cb,
3604 each_loose_cruft_fn cruft_cb,
3605 each_loose_subdir_fn subdir_cb,
3606 void *data)
3607 {
3608 size_t baselen = path->len;
3609 int r = 0;
3610 int i;
3611
3612 for (i = 0; i < 256; i++) {
3613 strbuf_addf(path, "/%02x", i);
3614 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3615 subdir_cb, data);
3616 strbuf_setlen(path, baselen);
3617 if (r)
3618 break;
3619 }
3620
3621 return r;
3622 }
3623
3624 int for_each_loose_file_in_objdir(const char *path,
3625 each_loose_object_fn obj_cb,
3626 each_loose_cruft_fn cruft_cb,
3627 each_loose_subdir_fn subdir_cb,
3628 void *data)
3629 {
3630 struct strbuf buf = STRBUF_INIT;
3631 int r;
3632
3633 strbuf_addstr(&buf, path);
3634 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3635 subdir_cb, data);
3636 strbuf_release(&buf);
3637
3638 return r;
3639 }
3640
3641 struct loose_alt_odb_data {
3642 each_loose_object_fn *cb;
3643 void *data;
3644 };
3645
3646 static int loose_from_alt_odb(struct alternate_object_database *alt,
3647 void *vdata)
3648 {
3649 struct loose_alt_odb_data *data = vdata;
3650 struct strbuf buf = STRBUF_INIT;
3651 int r;
3652
3653 /* copy base not including trailing '/' */
3654 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3655 r = for_each_loose_file_in_objdir_buf(&buf,
3656 data->cb, NULL, NULL,
3657 data->data);
3658 strbuf_release(&buf);
3659 return r;
3660 }
3661
3662 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3663 {
3664 struct loose_alt_odb_data alt;
3665 int r;
3666
3667 r = for_each_loose_file_in_objdir(get_object_directory(),
3668 cb, NULL, NULL, data);
3669 if (r)
3670 return r;
3671
3672 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3673 return 0;
3674
3675 alt.cb = cb;
3676 alt.data = data;
3677 return foreach_alt_odb(loose_from_alt_odb, &alt);
3678 }
3679
3680 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3681 {
3682 uint32_t i;
3683 int r = 0;
3684
3685 for (i = 0; i < p->num_objects; i++) {
3686 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3687
3688 if (!sha1)
3689 return error("unable to get sha1 of object %u in %s",
3690 i, p->pack_name);
3691
3692 r = cb(sha1, p, i, data);
3693 if (r)
3694 break;
3695 }
3696 return r;
3697 }
3698
3699 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3700 {
3701 struct packed_git *p;
3702 int r = 0;
3703 int pack_errors = 0;
3704
3705 prepare_packed_git();
3706 for (p = packed_git; p; p = p->next) {
3707 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3708 continue;
3709 if (open_pack_index(p)) {
3710 pack_errors = 1;
3711 continue;
3712 }
3713 r = for_each_object_in_pack(p, cb, data);
3714 if (r)
3715 break;
3716 }
3717 return r ? r : pack_errors;
3718 }