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