Merge branch 'cb/open-noatime-clear-errno'
[git.git] / sha1_file.c
blob08302f5857b8ad298809adf04a6bffc136f843a4
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
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"
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
38 const unsigned char null_sha1[20];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git *last_found_pack;
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
71 int i;
72 struct cached_object *co = cached_objects;
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
83 int mkdir_in_gitdir(const char *path)
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
90 if (errno != EEXIST)
91 return -1;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
107 strbuf_release(&sb);
109 return adjust_shared_perm(path);
112 enum scld_error safe_create_leading_directories(char *path)
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
124 if (!*slash)
125 break;
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
159 *slash = slash_character;
161 return ret;
164 enum scld_error safe_create_leading_directories_const(const char *path)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
185 const char *sha1_file_name(const unsigned char *sha1)
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
191 objdir = get_object_directory();
192 len = strlen(objdir);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
226 buf = *name;
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
234 return *base;
237 char *sha1_pack_name(const unsigned char *sha1)
239 static char *name, *base;
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1)
246 static char *name, *base;
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
281 strbuf_addstr(&pathbuf, entry);
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
285 pfxlen = strlen(pathbuf.buf);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
335 ent->base[pfxlen] = '/';
337 return 0;
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
375 void read_info_alternates(const char * relative_base, int depth)
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 const char alt_file_name[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path[PATH_MAX + 1 + sizeof alt_file_name];
385 int fd;
387 sprintf(path, "%s/%s", relative_base, alt_file_name);
388 fd = git_open_noatime(path);
389 if (fd < 0)
390 return;
391 if (fstat(fd, &st) || (st.st_size == 0)) {
392 close(fd);
393 return;
395 mapsz = xsize_t(st.st_size);
396 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
397 close(fd);
399 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
401 munmap(map, mapsz);
404 void add_to_alternates_file(const char *reference)
406 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
407 char *alts = git_pathdup("objects/info/alternates");
408 FILE *in, *out;
410 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
411 out = fdopen_lock_file(lock, "w");
412 if (!out)
413 die_errno("unable to fdopen alternates lockfile");
415 in = fopen(alts, "r");
416 if (in) {
417 struct strbuf line = STRBUF_INIT;
418 int found = 0;
420 while (strbuf_getline(&line, in, '\n') != EOF) {
421 if (!strcmp(reference, line.buf)) {
422 found = 1;
423 break;
425 fprintf_or_die(out, "%s\n", line.buf);
428 strbuf_release(&line);
429 fclose(in);
431 if (found) {
432 rollback_lock_file(lock);
433 lock = NULL;
436 else if (errno != ENOENT)
437 die_errno("unable to read alternates file");
439 if (lock) {
440 fprintf_or_die(out, "%s\n", reference);
441 if (commit_lock_file(lock))
442 die_errno("unable to move new alternates file into place");
443 if (alt_odb_tail)
444 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
446 free(alts);
449 int foreach_alt_odb(alt_odb_fn fn, void *cb)
451 struct alternate_object_database *ent;
452 int r = 0;
454 prepare_alt_odb();
455 for (ent = alt_odb_list; ent; ent = ent->next) {
456 r = fn(ent, cb);
457 if (r)
458 break;
460 return r;
463 void prepare_alt_odb(void)
465 const char *alt;
467 if (alt_odb_tail)
468 return;
470 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
471 if (!alt) alt = "";
473 alt_odb_tail = &alt_odb_list;
474 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
476 read_info_alternates(get_object_directory(), 0);
479 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
480 static int freshen_file(const char *fn)
482 struct utimbuf t;
483 t.actime = t.modtime = time(NULL);
484 return !utime(fn, &t);
488 * All of the check_and_freshen functions return 1 if the file exists and was
489 * freshened (if freshening was requested), 0 otherwise. If they return
490 * 0, you should not assume that it is safe to skip a write of the object (it
491 * either does not exist on disk, or has a stale mtime and may be subject to
492 * pruning).
494 static int check_and_freshen_file(const char *fn, int freshen)
496 if (access(fn, F_OK))
497 return 0;
498 if (freshen && !freshen_file(fn))
499 return 0;
500 return 1;
503 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
505 return check_and_freshen_file(sha1_file_name(sha1), freshen);
508 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
510 struct alternate_object_database *alt;
511 prepare_alt_odb();
512 for (alt = alt_odb_list; alt; alt = alt->next) {
513 fill_sha1_path(alt->name, sha1);
514 if (check_and_freshen_file(alt->base, freshen))
515 return 1;
517 return 0;
520 static int check_and_freshen(const unsigned char *sha1, int freshen)
522 return check_and_freshen_local(sha1, freshen) ||
523 check_and_freshen_nonlocal(sha1, freshen);
526 int has_loose_object_nonlocal(const unsigned char *sha1)
528 return check_and_freshen_nonlocal(sha1, 0);
531 static int has_loose_object(const unsigned char *sha1)
533 return check_and_freshen(sha1, 0);
536 static unsigned int pack_used_ctr;
537 static unsigned int pack_mmap_calls;
538 static unsigned int peak_pack_open_windows;
539 static unsigned int pack_open_windows;
540 static unsigned int pack_open_fds;
541 static unsigned int pack_max_fds;
542 static size_t peak_pack_mapped;
543 static size_t pack_mapped;
544 struct packed_git *packed_git;
546 void pack_report(void)
548 fprintf(stderr,
549 "pack_report: getpagesize() = %10" SZ_FMT "\n"
550 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
551 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
552 sz_fmt(getpagesize()),
553 sz_fmt(packed_git_window_size),
554 sz_fmt(packed_git_limit));
555 fprintf(stderr,
556 "pack_report: pack_used_ctr = %10u\n"
557 "pack_report: pack_mmap_calls = %10u\n"
558 "pack_report: pack_open_windows = %10u / %10u\n"
559 "pack_report: pack_mapped = "
560 "%10" SZ_FMT " / %10" SZ_FMT "\n",
561 pack_used_ctr,
562 pack_mmap_calls,
563 pack_open_windows, peak_pack_open_windows,
564 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
568 * Open and mmap the index file at path, perform a couple of
569 * consistency checks, then record its information to p. Return 0 on
570 * success.
572 static int check_packed_git_idx(const char *path, struct packed_git *p)
574 void *idx_map;
575 struct pack_idx_header *hdr;
576 size_t idx_size;
577 uint32_t version, nr, i, *index;
578 int fd = git_open_noatime(path);
579 struct stat st;
581 if (fd < 0)
582 return -1;
583 if (fstat(fd, &st)) {
584 close(fd);
585 return -1;
587 idx_size = xsize_t(st.st_size);
588 if (idx_size < 4 * 256 + 20 + 20) {
589 close(fd);
590 return error("index file %s is too small", path);
592 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
593 close(fd);
595 hdr = idx_map;
596 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
597 version = ntohl(hdr->idx_version);
598 if (version < 2 || version > 2) {
599 munmap(idx_map, idx_size);
600 return error("index file %s is version %"PRIu32
601 " and is not supported by this binary"
602 " (try upgrading GIT to a newer version)",
603 path, version);
605 } else
606 version = 1;
608 nr = 0;
609 index = idx_map;
610 if (version > 1)
611 index += 2; /* skip index header */
612 for (i = 0; i < 256; i++) {
613 uint32_t n = ntohl(index[i]);
614 if (n < nr) {
615 munmap(idx_map, idx_size);
616 return error("non-monotonic index %s", path);
618 nr = n;
621 if (version == 1) {
623 * Total size:
624 * - 256 index entries 4 bytes each
625 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
626 * - 20-byte SHA1 of the packfile
627 * - 20-byte SHA1 file checksum
629 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
630 munmap(idx_map, idx_size);
631 return error("wrong index v1 file size in %s", path);
633 } else if (version == 2) {
635 * Minimum size:
636 * - 8 bytes of header
637 * - 256 index entries 4 bytes each
638 * - 20-byte sha1 entry * nr
639 * - 4-byte crc entry * nr
640 * - 4-byte offset entry * nr
641 * - 20-byte SHA1 of the packfile
642 * - 20-byte SHA1 file checksum
643 * And after the 4-byte offset table might be a
644 * variable sized table containing 8-byte entries
645 * for offsets larger than 2^31.
647 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
648 unsigned long max_size = min_size;
649 if (nr)
650 max_size += (nr - 1)*8;
651 if (idx_size < min_size || idx_size > max_size) {
652 munmap(idx_map, idx_size);
653 return error("wrong index v2 file size in %s", path);
655 if (idx_size != min_size &&
657 * make sure we can deal with large pack offsets.
658 * 31-bit signed offset won't be enough, neither
659 * 32-bit unsigned one will be.
661 (sizeof(off_t) <= 4)) {
662 munmap(idx_map, idx_size);
663 return error("pack too large for current definition of off_t in %s", path);
667 p->index_version = version;
668 p->index_data = idx_map;
669 p->index_size = idx_size;
670 p->num_objects = nr;
671 return 0;
674 int open_pack_index(struct packed_git *p)
676 char *idx_name;
677 int ret;
679 if (p->index_data)
680 return 0;
682 idx_name = xstrdup(p->pack_name);
683 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
684 ret = check_packed_git_idx(idx_name, p);
685 free(idx_name);
686 return ret;
689 static void scan_windows(struct packed_git *p,
690 struct packed_git **lru_p,
691 struct pack_window **lru_w,
692 struct pack_window **lru_l)
694 struct pack_window *w, *w_l;
696 for (w_l = NULL, w = p->windows; w; w = w->next) {
697 if (!w->inuse_cnt) {
698 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
699 *lru_p = p;
700 *lru_w = w;
701 *lru_l = w_l;
704 w_l = w;
708 static int unuse_one_window(struct packed_git *current)
710 struct packed_git *p, *lru_p = NULL;
711 struct pack_window *lru_w = NULL, *lru_l = NULL;
713 if (current)
714 scan_windows(current, &lru_p, &lru_w, &lru_l);
715 for (p = packed_git; p; p = p->next)
716 scan_windows(p, &lru_p, &lru_w, &lru_l);
717 if (lru_p) {
718 munmap(lru_w->base, lru_w->len);
719 pack_mapped -= lru_w->len;
720 if (lru_l)
721 lru_l->next = lru_w->next;
722 else
723 lru_p->windows = lru_w->next;
724 free(lru_w);
725 pack_open_windows--;
726 return 1;
728 return 0;
731 void release_pack_memory(size_t need)
733 size_t cur = pack_mapped;
734 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
735 ; /* nothing */
738 static void mmap_limit_check(size_t length)
740 static size_t limit = 0;
741 if (!limit) {
742 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
743 if (!limit)
744 limit = SIZE_MAX;
746 if (length > limit)
747 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
748 (uintmax_t)length, (uintmax_t)limit);
751 void *xmmap_gently(void *start, size_t length,
752 int prot, int flags, int fd, off_t offset)
754 void *ret;
756 mmap_limit_check(length);
757 ret = mmap(start, length, prot, flags, fd, offset);
758 if (ret == MAP_FAILED) {
759 if (!length)
760 return NULL;
761 release_pack_memory(length);
762 ret = mmap(start, length, prot, flags, fd, offset);
764 return ret;
767 void *xmmap(void *start, size_t length,
768 int prot, int flags, int fd, off_t offset)
770 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
771 if (ret == MAP_FAILED)
772 die_errno("mmap failed");
773 return ret;
776 void close_pack_windows(struct packed_git *p)
778 while (p->windows) {
779 struct pack_window *w = p->windows;
781 if (w->inuse_cnt)
782 die("pack '%s' still has open windows to it",
783 p->pack_name);
784 munmap(w->base, w->len);
785 pack_mapped -= w->len;
786 pack_open_windows--;
787 p->windows = w->next;
788 free(w);
793 * The LRU pack is the one with the oldest MRU window, preferring packs
794 * with no used windows, or the oldest mtime if it has no windows allocated.
796 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
798 struct pack_window *w, *this_mru_w;
799 int has_windows_inuse = 0;
802 * Reject this pack if it has windows and the previously selected
803 * one does not. If this pack does not have windows, reject
804 * it if the pack file is newer than the previously selected one.
806 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
807 return;
809 for (w = this_mru_w = p->windows; w; w = w->next) {
811 * Reject this pack if any of its windows are in use,
812 * but the previously selected pack did not have any
813 * inuse windows. Otherwise, record that this pack
814 * has windows in use.
816 if (w->inuse_cnt) {
817 if (*accept_windows_inuse)
818 has_windows_inuse = 1;
819 else
820 return;
823 if (w->last_used > this_mru_w->last_used)
824 this_mru_w = w;
827 * Reject this pack if it has windows that have been
828 * used more recently than the previously selected pack.
829 * If the previously selected pack had windows inuse and
830 * we have not encountered a window in this pack that is
831 * inuse, skip this check since we prefer a pack with no
832 * inuse windows to one that has inuse windows.
834 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
835 this_mru_w->last_used > (*mru_w)->last_used)
836 return;
840 * Select this pack.
842 *mru_w = this_mru_w;
843 *lru_p = p;
844 *accept_windows_inuse = has_windows_inuse;
847 static int close_one_pack(void)
849 struct packed_git *p, *lru_p = NULL;
850 struct pack_window *mru_w = NULL;
851 int accept_windows_inuse = 1;
853 for (p = packed_git; p; p = p->next) {
854 if (p->pack_fd == -1)
855 continue;
856 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
859 if (lru_p) {
860 close(lru_p->pack_fd);
861 pack_open_fds--;
862 lru_p->pack_fd = -1;
863 return 1;
866 return 0;
869 void unuse_pack(struct pack_window **w_cursor)
871 struct pack_window *w = *w_cursor;
872 if (w) {
873 w->inuse_cnt--;
874 *w_cursor = NULL;
878 void close_pack_index(struct packed_git *p)
880 if (p->index_data) {
881 munmap((void *)p->index_data, p->index_size);
882 p->index_data = NULL;
887 * This is used by git-repack in case a newly created pack happens to
888 * contain the same set of objects as an existing one. In that case
889 * the resulting file might be different even if its name would be the
890 * same. It is best to close any reference to the old pack before it is
891 * replaced on disk. Of course no index pointers or windows for given pack
892 * must subsist at this point. If ever objects from this pack are requested
893 * again, the new version of the pack will be reinitialized through
894 * reprepare_packed_git().
896 void free_pack_by_name(const char *pack_name)
898 struct packed_git *p, **pp = &packed_git;
900 while (*pp) {
901 p = *pp;
902 if (strcmp(pack_name, p->pack_name) == 0) {
903 clear_delta_base_cache();
904 close_pack_windows(p);
905 if (p->pack_fd != -1) {
906 close(p->pack_fd);
907 pack_open_fds--;
909 close_pack_index(p);
910 free(p->bad_object_sha1);
911 *pp = p->next;
912 if (last_found_pack == p)
913 last_found_pack = NULL;
914 free(p);
915 return;
917 pp = &p->next;
921 static unsigned int get_max_fd_limit(void)
923 #ifdef RLIMIT_NOFILE
925 struct rlimit lim;
927 if (!getrlimit(RLIMIT_NOFILE, &lim))
928 return lim.rlim_cur;
930 #endif
932 #ifdef _SC_OPEN_MAX
934 long open_max = sysconf(_SC_OPEN_MAX);
935 if (0 < open_max)
936 return open_max;
938 * Otherwise, we got -1 for one of the two
939 * reasons:
941 * (1) sysconf() did not understand _SC_OPEN_MAX
942 * and signaled an error with -1; or
943 * (2) sysconf() said there is no limit.
945 * We _could_ clear errno before calling sysconf() to
946 * tell these two cases apart and return a huge number
947 * in the latter case to let the caller cap it to a
948 * value that is not so selfish, but letting the
949 * fallback OPEN_MAX codepath take care of these cases
950 * is a lot simpler.
953 #endif
955 #ifdef OPEN_MAX
956 return OPEN_MAX;
957 #else
958 return 1; /* see the caller ;-) */
959 #endif
963 * Do not call this directly as this leaks p->pack_fd on error return;
964 * call open_packed_git() instead.
966 static int open_packed_git_1(struct packed_git *p)
968 struct stat st;
969 struct pack_header hdr;
970 unsigned char sha1[20];
971 unsigned char *idx_sha1;
972 long fd_flag;
974 if (!p->index_data && open_pack_index(p))
975 return error("packfile %s index unavailable", p->pack_name);
977 if (!pack_max_fds) {
978 unsigned int max_fds = get_max_fd_limit();
980 /* Save 3 for stdin/stdout/stderr, 22 for work */
981 if (25 < max_fds)
982 pack_max_fds = max_fds - 25;
983 else
984 pack_max_fds = 1;
987 while (pack_max_fds <= pack_open_fds && close_one_pack())
988 ; /* nothing */
990 p->pack_fd = git_open_noatime(p->pack_name);
991 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
992 return -1;
993 pack_open_fds++;
995 /* If we created the struct before we had the pack we lack size. */
996 if (!p->pack_size) {
997 if (!S_ISREG(st.st_mode))
998 return error("packfile %s not a regular file", p->pack_name);
999 p->pack_size = st.st_size;
1000 } else if (p->pack_size != st.st_size)
1001 return error("packfile %s size changed", p->pack_name);
1003 /* We leave these file descriptors open with sliding mmap;
1004 * there is no point keeping them open across exec(), though.
1006 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1007 if (fd_flag < 0)
1008 return error("cannot determine file descriptor flags");
1009 fd_flag |= FD_CLOEXEC;
1010 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1011 return error("cannot set FD_CLOEXEC");
1013 /* Verify we recognize this pack file format. */
1014 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1015 return error("file %s is far too short to be a packfile", p->pack_name);
1016 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1017 return error("file %s is not a GIT packfile", p->pack_name);
1018 if (!pack_version_ok(hdr.hdr_version))
1019 return error("packfile %s is version %"PRIu32" and not"
1020 " supported (try upgrading GIT to a newer version)",
1021 p->pack_name, ntohl(hdr.hdr_version));
1023 /* Verify the pack matches its index. */
1024 if (p->num_objects != ntohl(hdr.hdr_entries))
1025 return error("packfile %s claims to have %"PRIu32" objects"
1026 " while index indicates %"PRIu32" objects",
1027 p->pack_name, ntohl(hdr.hdr_entries),
1028 p->num_objects);
1029 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1030 return error("end of packfile %s is unavailable", p->pack_name);
1031 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1032 return error("packfile %s signature is unavailable", p->pack_name);
1033 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1034 if (hashcmp(sha1, idx_sha1))
1035 return error("packfile %s does not match index", p->pack_name);
1036 return 0;
1039 static int open_packed_git(struct packed_git *p)
1041 if (!open_packed_git_1(p))
1042 return 0;
1043 if (p->pack_fd != -1) {
1044 close(p->pack_fd);
1045 pack_open_fds--;
1046 p->pack_fd = -1;
1048 return -1;
1051 static int in_window(struct pack_window *win, off_t offset)
1053 /* We must promise at least 20 bytes (one hash) after the
1054 * offset is available from this window, otherwise the offset
1055 * is not actually in this window and a different window (which
1056 * has that one hash excess) must be used. This is to support
1057 * the object header and delta base parsing routines below.
1059 off_t win_off = win->offset;
1060 return win_off <= offset
1061 && (offset + 20) <= (win_off + win->len);
1064 unsigned char *use_pack(struct packed_git *p,
1065 struct pack_window **w_cursor,
1066 off_t offset,
1067 unsigned long *left)
1069 struct pack_window *win = *w_cursor;
1071 /* Since packfiles end in a hash of their content and it's
1072 * pointless to ask for an offset into the middle of that
1073 * hash, and the in_window function above wouldn't match
1074 * don't allow an offset too close to the end of the file.
1076 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1077 die("packfile %s cannot be accessed", p->pack_name);
1078 if (offset > (p->pack_size - 20))
1079 die("offset beyond end of packfile (truncated pack?)");
1081 if (!win || !in_window(win, offset)) {
1082 if (win)
1083 win->inuse_cnt--;
1084 for (win = p->windows; win; win = win->next) {
1085 if (in_window(win, offset))
1086 break;
1088 if (!win) {
1089 size_t window_align = packed_git_window_size / 2;
1090 off_t len;
1092 if (p->pack_fd == -1 && open_packed_git(p))
1093 die("packfile %s cannot be accessed", p->pack_name);
1095 win = xcalloc(1, sizeof(*win));
1096 win->offset = (offset / window_align) * window_align;
1097 len = p->pack_size - win->offset;
1098 if (len > packed_git_window_size)
1099 len = packed_git_window_size;
1100 win->len = (size_t)len;
1101 pack_mapped += win->len;
1102 while (packed_git_limit < pack_mapped
1103 && unuse_one_window(p))
1104 ; /* nothing */
1105 win->base = xmmap(NULL, win->len,
1106 PROT_READ, MAP_PRIVATE,
1107 p->pack_fd, win->offset);
1108 if (win->base == MAP_FAILED)
1109 die("packfile %s cannot be mapped: %s",
1110 p->pack_name,
1111 strerror(errno));
1112 if (!win->offset && win->len == p->pack_size
1113 && !p->do_not_close) {
1114 close(p->pack_fd);
1115 pack_open_fds--;
1116 p->pack_fd = -1;
1118 pack_mmap_calls++;
1119 pack_open_windows++;
1120 if (pack_mapped > peak_pack_mapped)
1121 peak_pack_mapped = pack_mapped;
1122 if (pack_open_windows > peak_pack_open_windows)
1123 peak_pack_open_windows = pack_open_windows;
1124 win->next = p->windows;
1125 p->windows = win;
1128 if (win != *w_cursor) {
1129 win->last_used = pack_used_ctr++;
1130 win->inuse_cnt++;
1131 *w_cursor = win;
1133 offset -= win->offset;
1134 if (left)
1135 *left = win->len - xsize_t(offset);
1136 return win->base + offset;
1139 static struct packed_git *alloc_packed_git(int extra)
1141 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1142 memset(p, 0, sizeof(*p));
1143 p->pack_fd = -1;
1144 return p;
1147 static void try_to_free_pack_memory(size_t size)
1149 release_pack_memory(size);
1152 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1154 static int have_set_try_to_free_routine;
1155 struct stat st;
1156 struct packed_git *p = alloc_packed_git(path_len + 2);
1158 if (!have_set_try_to_free_routine) {
1159 have_set_try_to_free_routine = 1;
1160 set_try_to_free_routine(try_to_free_pack_memory);
1164 * Make sure a corresponding .pack file exists and that
1165 * the index looks sane.
1167 path_len -= strlen(".idx");
1168 if (path_len < 1) {
1169 free(p);
1170 return NULL;
1172 memcpy(p->pack_name, path, path_len);
1174 strcpy(p->pack_name + path_len, ".keep");
1175 if (!access(p->pack_name, F_OK))
1176 p->pack_keep = 1;
1178 strcpy(p->pack_name + path_len, ".pack");
1179 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1180 free(p);
1181 return NULL;
1184 /* ok, it looks sane as far as we can check without
1185 * actually mapping the pack file.
1187 p->pack_size = st.st_size;
1188 p->pack_local = local;
1189 p->mtime = st.st_mtime;
1190 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1191 hashclr(p->sha1);
1192 return p;
1195 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1197 const char *path = sha1_pack_name(sha1);
1198 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1200 strcpy(p->pack_name, path);
1201 hashcpy(p->sha1, sha1);
1202 if (check_packed_git_idx(idx_path, p)) {
1203 free(p);
1204 return NULL;
1207 return p;
1210 void install_packed_git(struct packed_git *pack)
1212 if (pack->pack_fd != -1)
1213 pack_open_fds++;
1215 pack->next = packed_git;
1216 packed_git = pack;
1219 void (*report_garbage)(const char *desc, const char *path);
1221 static void report_helper(const struct string_list *list,
1222 int seen_bits, int first, int last)
1224 const char *msg;
1225 switch (seen_bits) {
1226 case 0:
1227 msg = "no corresponding .idx or .pack";
1228 break;
1229 case 1:
1230 msg = "no corresponding .idx";
1231 break;
1232 case 2:
1233 msg = "no corresponding .pack";
1234 break;
1235 default:
1236 return;
1238 for (; first < last; first++)
1239 report_garbage(msg, list->items[first].string);
1242 static void report_pack_garbage(struct string_list *list)
1244 int i, baselen = -1, first = 0, seen_bits = 0;
1246 if (!report_garbage)
1247 return;
1249 string_list_sort(list);
1251 for (i = 0; i < list->nr; i++) {
1252 const char *path = list->items[i].string;
1253 if (baselen != -1 &&
1254 strncmp(path, list->items[first].string, baselen)) {
1255 report_helper(list, seen_bits, first, i);
1256 baselen = -1;
1257 seen_bits = 0;
1259 if (baselen == -1) {
1260 const char *dot = strrchr(path, '.');
1261 if (!dot) {
1262 report_garbage("garbage found", path);
1263 continue;
1265 baselen = dot - path + 1;
1266 first = i;
1268 if (!strcmp(path + baselen, "pack"))
1269 seen_bits |= 1;
1270 else if (!strcmp(path + baselen, "idx"))
1271 seen_bits |= 2;
1273 report_helper(list, seen_bits, first, list->nr);
1276 static void prepare_packed_git_one(char *objdir, int local)
1278 struct strbuf path = STRBUF_INIT;
1279 size_t dirnamelen;
1280 DIR *dir;
1281 struct dirent *de;
1282 struct string_list garbage = STRING_LIST_INIT_DUP;
1284 strbuf_addstr(&path, objdir);
1285 strbuf_addstr(&path, "/pack");
1286 dir = opendir(path.buf);
1287 if (!dir) {
1288 if (errno != ENOENT)
1289 error("unable to open object pack directory: %s: %s",
1290 path.buf, strerror(errno));
1291 strbuf_release(&path);
1292 return;
1294 strbuf_addch(&path, '/');
1295 dirnamelen = path.len;
1296 while ((de = readdir(dir)) != NULL) {
1297 struct packed_git *p;
1298 size_t base_len;
1300 if (is_dot_or_dotdot(de->d_name))
1301 continue;
1303 strbuf_setlen(&path, dirnamelen);
1304 strbuf_addstr(&path, de->d_name);
1306 base_len = path.len;
1307 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1308 /* Don't reopen a pack we already have. */
1309 for (p = packed_git; p; p = p->next) {
1310 size_t len;
1311 if (strip_suffix(p->pack_name, ".pack", &len) &&
1312 len == base_len &&
1313 !memcmp(p->pack_name, path.buf, len))
1314 break;
1316 if (p == NULL &&
1318 * See if it really is a valid .idx file with
1319 * corresponding .pack file that we can map.
1321 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1322 install_packed_git(p);
1325 if (!report_garbage)
1326 continue;
1328 if (ends_with(de->d_name, ".idx") ||
1329 ends_with(de->d_name, ".pack") ||
1330 ends_with(de->d_name, ".bitmap") ||
1331 ends_with(de->d_name, ".keep"))
1332 string_list_append(&garbage, path.buf);
1333 else
1334 report_garbage("garbage found", path.buf);
1336 closedir(dir);
1337 report_pack_garbage(&garbage);
1338 string_list_clear(&garbage, 0);
1339 strbuf_release(&path);
1342 static int sort_pack(const void *a_, const void *b_)
1344 struct packed_git *a = *((struct packed_git **)a_);
1345 struct packed_git *b = *((struct packed_git **)b_);
1346 int st;
1349 * Local packs tend to contain objects specific to our
1350 * variant of the project than remote ones. In addition,
1351 * remote ones could be on a network mounted filesystem.
1352 * Favor local ones for these reasons.
1354 st = a->pack_local - b->pack_local;
1355 if (st)
1356 return -st;
1359 * Younger packs tend to contain more recent objects,
1360 * and more recent objects tend to get accessed more
1361 * often.
1363 if (a->mtime < b->mtime)
1364 return 1;
1365 else if (a->mtime == b->mtime)
1366 return 0;
1367 return -1;
1370 static void rearrange_packed_git(void)
1372 struct packed_git **ary, *p;
1373 int i, n;
1375 for (n = 0, p = packed_git; p; p = p->next)
1376 n++;
1377 if (n < 2)
1378 return;
1380 /* prepare an array of packed_git for easier sorting */
1381 ary = xcalloc(n, sizeof(struct packed_git *));
1382 for (n = 0, p = packed_git; p; p = p->next)
1383 ary[n++] = p;
1385 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1387 /* link them back again */
1388 for (i = 0; i < n - 1; i++)
1389 ary[i]->next = ary[i + 1];
1390 ary[n - 1]->next = NULL;
1391 packed_git = ary[0];
1393 free(ary);
1396 static int prepare_packed_git_run_once = 0;
1397 void prepare_packed_git(void)
1399 struct alternate_object_database *alt;
1401 if (prepare_packed_git_run_once)
1402 return;
1403 prepare_packed_git_one(get_object_directory(), 1);
1404 prepare_alt_odb();
1405 for (alt = alt_odb_list; alt; alt = alt->next) {
1406 alt->name[-1] = 0;
1407 prepare_packed_git_one(alt->base, 0);
1408 alt->name[-1] = '/';
1410 rearrange_packed_git();
1411 prepare_packed_git_run_once = 1;
1414 void reprepare_packed_git(void)
1416 prepare_packed_git_run_once = 0;
1417 prepare_packed_git();
1420 static void mark_bad_packed_object(struct packed_git *p,
1421 const unsigned char *sha1)
1423 unsigned i;
1424 for (i = 0; i < p->num_bad_objects; i++)
1425 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1426 return;
1427 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1428 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1429 p->num_bad_objects++;
1432 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1434 struct packed_git *p;
1435 unsigned i;
1437 for (p = packed_git; p; p = p->next)
1438 for (i = 0; i < p->num_bad_objects; i++)
1439 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1440 return p;
1441 return NULL;
1445 * With an in-core object data in "map", rehash it to make sure the
1446 * object name actually matches "sha1" to detect object corruption.
1447 * With "map" == NULL, try reading the object named with "sha1" using
1448 * the streaming interface and rehash it to do the same.
1450 int check_sha1_signature(const unsigned char *sha1, void *map,
1451 unsigned long size, const char *type)
1453 unsigned char real_sha1[20];
1454 enum object_type obj_type;
1455 struct git_istream *st;
1456 git_SHA_CTX c;
1457 char hdr[32];
1458 int hdrlen;
1460 if (map) {
1461 hash_sha1_file(map, size, type, real_sha1);
1462 return hashcmp(sha1, real_sha1) ? -1 : 0;
1465 st = open_istream(sha1, &obj_type, &size, NULL);
1466 if (!st)
1467 return -1;
1469 /* Generate the header */
1470 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1472 /* Sha1.. */
1473 git_SHA1_Init(&c);
1474 git_SHA1_Update(&c, hdr, hdrlen);
1475 for (;;) {
1476 char buf[1024 * 16];
1477 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1479 if (readlen < 0) {
1480 close_istream(st);
1481 return -1;
1483 if (!readlen)
1484 break;
1485 git_SHA1_Update(&c, buf, readlen);
1487 git_SHA1_Final(real_sha1, &c);
1488 close_istream(st);
1489 return hashcmp(sha1, real_sha1) ? -1 : 0;
1492 int git_open_noatime(const char *name)
1494 static int sha1_file_open_flag = O_NOATIME;
1496 for (;;) {
1497 int fd;
1499 errno = 0;
1500 fd = open(name, O_RDONLY | sha1_file_open_flag);
1501 if (fd >= 0)
1502 return fd;
1504 /* Might the failure be due to O_NOATIME? */
1505 if (errno != ENOENT && sha1_file_open_flag) {
1506 sha1_file_open_flag = 0;
1507 continue;
1510 return -1;
1514 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1516 struct alternate_object_database *alt;
1518 if (!lstat(sha1_file_name(sha1), st))
1519 return 0;
1521 prepare_alt_odb();
1522 errno = ENOENT;
1523 for (alt = alt_odb_list; alt; alt = alt->next) {
1524 fill_sha1_path(alt->name, sha1);
1525 if (!lstat(alt->base, st))
1526 return 0;
1529 return -1;
1532 static int open_sha1_file(const unsigned char *sha1)
1534 int fd;
1535 struct alternate_object_database *alt;
1536 int most_interesting_errno;
1538 fd = git_open_noatime(sha1_file_name(sha1));
1539 if (fd >= 0)
1540 return fd;
1541 most_interesting_errno = errno;
1543 prepare_alt_odb();
1544 for (alt = alt_odb_list; alt; alt = alt->next) {
1545 fill_sha1_path(alt->name, sha1);
1546 fd = git_open_noatime(alt->base);
1547 if (fd >= 0)
1548 return fd;
1549 if (most_interesting_errno == ENOENT)
1550 most_interesting_errno = errno;
1552 errno = most_interesting_errno;
1553 return -1;
1556 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1558 void *map;
1559 int fd;
1561 fd = open_sha1_file(sha1);
1562 map = NULL;
1563 if (fd >= 0) {
1564 struct stat st;
1566 if (!fstat(fd, &st)) {
1567 *size = xsize_t(st.st_size);
1568 if (!*size) {
1569 /* mmap() is forbidden on empty files */
1570 error("object file %s is empty", sha1_file_name(sha1));
1571 return NULL;
1573 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1575 close(fd);
1577 return map;
1580 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1581 unsigned long len, enum object_type *type, unsigned long *sizep)
1583 unsigned shift;
1584 unsigned long size, c;
1585 unsigned long used = 0;
1587 c = buf[used++];
1588 *type = (c >> 4) & 7;
1589 size = c & 15;
1590 shift = 4;
1591 while (c & 0x80) {
1592 if (len <= used || bitsizeof(long) <= shift) {
1593 error("bad object header");
1594 size = used = 0;
1595 break;
1597 c = buf[used++];
1598 size += (c & 0x7f) << shift;
1599 shift += 7;
1601 *sizep = size;
1602 return used;
1605 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1607 /* Get the data stream */
1608 memset(stream, 0, sizeof(*stream));
1609 stream->next_in = map;
1610 stream->avail_in = mapsize;
1611 stream->next_out = buffer;
1612 stream->avail_out = bufsiz;
1614 git_inflate_init(stream);
1615 return git_inflate(stream, 0);
1618 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1619 unsigned long mapsize, void *buffer,
1620 unsigned long bufsiz, struct strbuf *header)
1622 int status;
1624 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1627 * Check if entire header is unpacked in the first iteration.
1629 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1630 return 0;
1633 * buffer[0..bufsiz] was not large enough. Copy the partial
1634 * result out to header, and then append the result of further
1635 * reading the stream.
1637 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1638 stream->next_out = buffer;
1639 stream->avail_out = bufsiz;
1641 do {
1642 status = git_inflate(stream, 0);
1643 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1644 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1645 return 0;
1646 stream->next_out = buffer;
1647 stream->avail_out = bufsiz;
1648 } while (status != Z_STREAM_END);
1649 return -1;
1652 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1654 int bytes = strlen(buffer) + 1;
1655 unsigned char *buf = xmallocz(size);
1656 unsigned long n;
1657 int status = Z_OK;
1659 n = stream->total_out - bytes;
1660 if (n > size)
1661 n = size;
1662 memcpy(buf, (char *) buffer + bytes, n);
1663 bytes = n;
1664 if (bytes <= size) {
1666 * The above condition must be (bytes <= size), not
1667 * (bytes < size). In other words, even though we
1668 * expect no more output and set avail_out to zero,
1669 * the input zlib stream may have bytes that express
1670 * "this concludes the stream", and we *do* want to
1671 * eat that input.
1673 * Otherwise we would not be able to test that we
1674 * consumed all the input to reach the expected size;
1675 * we also want to check that zlib tells us that all
1676 * went well with status == Z_STREAM_END at the end.
1678 stream->next_out = buf + bytes;
1679 stream->avail_out = size - bytes;
1680 while (status == Z_OK)
1681 status = git_inflate(stream, Z_FINISH);
1683 if (status == Z_STREAM_END && !stream->avail_in) {
1684 git_inflate_end(stream);
1685 return buf;
1688 if (status < 0)
1689 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1690 else if (stream->avail_in)
1691 error("garbage at end of loose object '%s'",
1692 sha1_to_hex(sha1));
1693 free(buf);
1694 return NULL;
1698 * We used to just use "sscanf()", but that's actually way
1699 * too permissive for what we want to check. So do an anal
1700 * object header parse by hand.
1702 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1703 unsigned int flags)
1705 const char *type_buf = hdr;
1706 unsigned long size;
1707 int type, type_len = 0;
1710 * The type can be of any size but is followed by
1711 * a space.
1713 for (;;) {
1714 char c = *hdr++;
1715 if (c == ' ')
1716 break;
1717 type_len++;
1720 type = type_from_string_gently(type_buf, type_len, 1);
1721 if (oi->typename)
1722 strbuf_add(oi->typename, type_buf, type_len);
1724 * Set type to 0 if its an unknown object and
1725 * we're obtaining the type using '--allow-unkown-type'
1726 * option.
1728 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1729 type = 0;
1730 else if (type < 0)
1731 die("invalid object type");
1732 if (oi->typep)
1733 *oi->typep = type;
1736 * The length must follow immediately, and be in canonical
1737 * decimal format (ie "010" is not valid).
1739 size = *hdr++ - '0';
1740 if (size > 9)
1741 return -1;
1742 if (size) {
1743 for (;;) {
1744 unsigned long c = *hdr - '0';
1745 if (c > 9)
1746 break;
1747 hdr++;
1748 size = size * 10 + c;
1752 if (oi->sizep)
1753 *oi->sizep = size;
1756 * The length must be followed by a zero byte
1758 return *hdr ? -1 : type;
1761 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1763 struct object_info oi;
1765 oi.sizep = sizep;
1766 oi.typename = NULL;
1767 oi.typep = NULL;
1768 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1771 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1773 int ret;
1774 git_zstream stream;
1775 char hdr[8192];
1777 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1778 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1779 return NULL;
1781 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1784 unsigned long get_size_from_delta(struct packed_git *p,
1785 struct pack_window **w_curs,
1786 off_t curpos)
1788 const unsigned char *data;
1789 unsigned char delta_head[20], *in;
1790 git_zstream stream;
1791 int st;
1793 memset(&stream, 0, sizeof(stream));
1794 stream.next_out = delta_head;
1795 stream.avail_out = sizeof(delta_head);
1797 git_inflate_init(&stream);
1798 do {
1799 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1800 stream.next_in = in;
1801 st = git_inflate(&stream, Z_FINISH);
1802 curpos += stream.next_in - in;
1803 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1804 stream.total_out < sizeof(delta_head));
1805 git_inflate_end(&stream);
1806 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1807 error("delta data unpack-initial failed");
1808 return 0;
1811 /* Examine the initial part of the delta to figure out
1812 * the result size.
1814 data = delta_head;
1816 /* ignore base size */
1817 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1819 /* Read the result size */
1820 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1823 static off_t get_delta_base(struct packed_git *p,
1824 struct pack_window **w_curs,
1825 off_t *curpos,
1826 enum object_type type,
1827 off_t delta_obj_offset)
1829 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1830 off_t base_offset;
1832 /* use_pack() assured us we have [base_info, base_info + 20)
1833 * as a range that we can look at without walking off the
1834 * end of the mapped window. Its actually the hash size
1835 * that is assured. An OFS_DELTA longer than the hash size
1836 * is stupid, as then a REF_DELTA would be smaller to store.
1838 if (type == OBJ_OFS_DELTA) {
1839 unsigned used = 0;
1840 unsigned char c = base_info[used++];
1841 base_offset = c & 127;
1842 while (c & 128) {
1843 base_offset += 1;
1844 if (!base_offset || MSB(base_offset, 7))
1845 return 0; /* overflow */
1846 c = base_info[used++];
1847 base_offset = (base_offset << 7) + (c & 127);
1849 base_offset = delta_obj_offset - base_offset;
1850 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1851 return 0; /* out of bound */
1852 *curpos += used;
1853 } else if (type == OBJ_REF_DELTA) {
1854 /* The base entry _must_ be in the same pack */
1855 base_offset = find_pack_entry_one(base_info, p);
1856 *curpos += 20;
1857 } else
1858 die("I am totally screwed");
1859 return base_offset;
1863 * Like get_delta_base above, but we return the sha1 instead of the pack
1864 * offset. This means it is cheaper for REF deltas (we do not have to do
1865 * the final object lookup), but more expensive for OFS deltas (we
1866 * have to load the revidx to convert the offset back into a sha1).
1868 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1869 struct pack_window **w_curs,
1870 off_t curpos,
1871 enum object_type type,
1872 off_t delta_obj_offset)
1874 if (type == OBJ_REF_DELTA) {
1875 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1876 return base;
1877 } else if (type == OBJ_OFS_DELTA) {
1878 struct revindex_entry *revidx;
1879 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1880 type, delta_obj_offset);
1882 if (!base_offset)
1883 return NULL;
1885 revidx = find_pack_revindex(p, base_offset);
1886 if (!revidx)
1887 return NULL;
1889 return nth_packed_object_sha1(p, revidx->nr);
1890 } else
1891 return NULL;
1894 int unpack_object_header(struct packed_git *p,
1895 struct pack_window **w_curs,
1896 off_t *curpos,
1897 unsigned long *sizep)
1899 unsigned char *base;
1900 unsigned long left;
1901 unsigned long used;
1902 enum object_type type;
1904 /* use_pack() assures us we have [base, base + 20) available
1905 * as a range that we can look at. (Its actually the hash
1906 * size that is assured.) With our object header encoding
1907 * the maximum deflated object size is 2^137, which is just
1908 * insane, so we know won't exceed what we have been given.
1910 base = use_pack(p, w_curs, *curpos, &left);
1911 used = unpack_object_header_buffer(base, left, &type, sizep);
1912 if (!used) {
1913 type = OBJ_BAD;
1914 } else
1915 *curpos += used;
1917 return type;
1920 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1922 int type;
1923 struct revindex_entry *revidx;
1924 const unsigned char *sha1;
1925 revidx = find_pack_revindex(p, obj_offset);
1926 if (!revidx)
1927 return OBJ_BAD;
1928 sha1 = nth_packed_object_sha1(p, revidx->nr);
1929 mark_bad_packed_object(p, sha1);
1930 type = sha1_object_info(sha1, NULL);
1931 if (type <= OBJ_NONE)
1932 return OBJ_BAD;
1933 return type;
1936 #define POI_STACK_PREALLOC 64
1938 static enum object_type packed_to_object_type(struct packed_git *p,
1939 off_t obj_offset,
1940 enum object_type type,
1941 struct pack_window **w_curs,
1942 off_t curpos)
1944 off_t small_poi_stack[POI_STACK_PREALLOC];
1945 off_t *poi_stack = small_poi_stack;
1946 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1948 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1949 off_t base_offset;
1950 unsigned long size;
1951 /* Push the object we're going to leave behind */
1952 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1953 poi_stack_alloc = alloc_nr(poi_stack_nr);
1954 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1955 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1956 } else {
1957 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1959 poi_stack[poi_stack_nr++] = obj_offset;
1960 /* If parsing the base offset fails, just unwind */
1961 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1962 if (!base_offset)
1963 goto unwind;
1964 curpos = obj_offset = base_offset;
1965 type = unpack_object_header(p, w_curs, &curpos, &size);
1966 if (type <= OBJ_NONE) {
1967 /* If getting the base itself fails, we first
1968 * retry the base, otherwise unwind */
1969 type = retry_bad_packed_offset(p, base_offset);
1970 if (type > OBJ_NONE)
1971 goto out;
1972 goto unwind;
1976 switch (type) {
1977 case OBJ_BAD:
1978 case OBJ_COMMIT:
1979 case OBJ_TREE:
1980 case OBJ_BLOB:
1981 case OBJ_TAG:
1982 break;
1983 default:
1984 error("unknown object type %i at offset %"PRIuMAX" in %s",
1985 type, (uintmax_t)obj_offset, p->pack_name);
1986 type = OBJ_BAD;
1989 out:
1990 if (poi_stack != small_poi_stack)
1991 free(poi_stack);
1992 return type;
1994 unwind:
1995 while (poi_stack_nr) {
1996 obj_offset = poi_stack[--poi_stack_nr];
1997 type = retry_bad_packed_offset(p, obj_offset);
1998 if (type > OBJ_NONE)
1999 goto out;
2001 type = OBJ_BAD;
2002 goto out;
2005 static int packed_object_info(struct packed_git *p, off_t obj_offset,
2006 struct object_info *oi)
2008 struct pack_window *w_curs = NULL;
2009 unsigned long size;
2010 off_t curpos = obj_offset;
2011 enum object_type type;
2014 * We always get the representation type, but only convert it to
2015 * a "real" type later if the caller is interested.
2017 type = unpack_object_header(p, &w_curs, &curpos, &size);
2019 if (oi->sizep) {
2020 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2021 off_t tmp_pos = curpos;
2022 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2023 type, obj_offset);
2024 if (!base_offset) {
2025 type = OBJ_BAD;
2026 goto out;
2028 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2029 if (*oi->sizep == 0) {
2030 type = OBJ_BAD;
2031 goto out;
2033 } else {
2034 *oi->sizep = size;
2038 if (oi->disk_sizep) {
2039 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2040 *oi->disk_sizep = revidx[1].offset - obj_offset;
2043 if (oi->typep) {
2044 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2045 if (*oi->typep < 0) {
2046 type = OBJ_BAD;
2047 goto out;
2051 if (oi->delta_base_sha1) {
2052 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2053 const unsigned char *base;
2055 base = get_delta_base_sha1(p, &w_curs, curpos,
2056 type, obj_offset);
2057 if (!base) {
2058 type = OBJ_BAD;
2059 goto out;
2062 hashcpy(oi->delta_base_sha1, base);
2063 } else
2064 hashclr(oi->delta_base_sha1);
2067 out:
2068 unuse_pack(&w_curs);
2069 return type;
2072 static void *unpack_compressed_entry(struct packed_git *p,
2073 struct pack_window **w_curs,
2074 off_t curpos,
2075 unsigned long size)
2077 int st;
2078 git_zstream stream;
2079 unsigned char *buffer, *in;
2081 buffer = xmallocz_gently(size);
2082 if (!buffer)
2083 return NULL;
2084 memset(&stream, 0, sizeof(stream));
2085 stream.next_out = buffer;
2086 stream.avail_out = size + 1;
2088 git_inflate_init(&stream);
2089 do {
2090 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2091 stream.next_in = in;
2092 st = git_inflate(&stream, Z_FINISH);
2093 if (!stream.avail_out)
2094 break; /* the payload is larger than it should be */
2095 curpos += stream.next_in - in;
2096 } while (st == Z_OK || st == Z_BUF_ERROR);
2097 git_inflate_end(&stream);
2098 if ((st != Z_STREAM_END) || stream.total_out != size) {
2099 free(buffer);
2100 return NULL;
2103 return buffer;
2106 #define MAX_DELTA_CACHE (256)
2108 static size_t delta_base_cached;
2110 static struct delta_base_cache_lru_list {
2111 struct delta_base_cache_lru_list *prev;
2112 struct delta_base_cache_lru_list *next;
2113 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2115 static struct delta_base_cache_entry {
2116 struct delta_base_cache_lru_list lru;
2117 void *data;
2118 struct packed_git *p;
2119 off_t base_offset;
2120 unsigned long size;
2121 enum object_type type;
2122 } delta_base_cache[MAX_DELTA_CACHE];
2124 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2126 unsigned long hash;
2128 hash = (unsigned long)p + (unsigned long)base_offset;
2129 hash += (hash >> 8) + (hash >> 16);
2130 return hash % MAX_DELTA_CACHE;
2133 static struct delta_base_cache_entry *
2134 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2136 unsigned long hash = pack_entry_hash(p, base_offset);
2137 return delta_base_cache + hash;
2140 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2141 struct packed_git *p, off_t base_offset)
2143 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2146 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2148 struct delta_base_cache_entry *ent;
2149 ent = get_delta_base_cache_entry(p, base_offset);
2150 return eq_delta_base_cache_entry(ent, p, base_offset);
2153 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2155 ent->data = NULL;
2156 ent->lru.next->prev = ent->lru.prev;
2157 ent->lru.prev->next = ent->lru.next;
2158 delta_base_cached -= ent->size;
2161 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2162 unsigned long *base_size, enum object_type *type, int keep_cache)
2164 struct delta_base_cache_entry *ent;
2165 void *ret;
2167 ent = get_delta_base_cache_entry(p, base_offset);
2169 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2170 return unpack_entry(p, base_offset, type, base_size);
2172 ret = ent->data;
2174 if (!keep_cache)
2175 clear_delta_base_cache_entry(ent);
2176 else
2177 ret = xmemdupz(ent->data, ent->size);
2178 *type = ent->type;
2179 *base_size = ent->size;
2180 return ret;
2183 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2185 if (ent->data) {
2186 free(ent->data);
2187 ent->data = NULL;
2188 ent->lru.next->prev = ent->lru.prev;
2189 ent->lru.prev->next = ent->lru.next;
2190 delta_base_cached -= ent->size;
2194 void clear_delta_base_cache(void)
2196 unsigned long p;
2197 for (p = 0; p < MAX_DELTA_CACHE; p++)
2198 release_delta_base_cache(&delta_base_cache[p]);
2201 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2202 void *base, unsigned long base_size, enum object_type type)
2204 unsigned long hash = pack_entry_hash(p, base_offset);
2205 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2206 struct delta_base_cache_lru_list *lru;
2208 release_delta_base_cache(ent);
2209 delta_base_cached += base_size;
2211 for (lru = delta_base_cache_lru.next;
2212 delta_base_cached > delta_base_cache_limit
2213 && lru != &delta_base_cache_lru;
2214 lru = lru->next) {
2215 struct delta_base_cache_entry *f = (void *)lru;
2216 if (f->type == OBJ_BLOB)
2217 release_delta_base_cache(f);
2219 for (lru = delta_base_cache_lru.next;
2220 delta_base_cached > delta_base_cache_limit
2221 && lru != &delta_base_cache_lru;
2222 lru = lru->next) {
2223 struct delta_base_cache_entry *f = (void *)lru;
2224 release_delta_base_cache(f);
2227 ent->p = p;
2228 ent->base_offset = base_offset;
2229 ent->type = type;
2230 ent->data = base;
2231 ent->size = base_size;
2232 ent->lru.next = &delta_base_cache_lru;
2233 ent->lru.prev = delta_base_cache_lru.prev;
2234 delta_base_cache_lru.prev->next = &ent->lru;
2235 delta_base_cache_lru.prev = &ent->lru;
2238 static void *read_object(const unsigned char *sha1, enum object_type *type,
2239 unsigned long *size);
2241 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2243 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2244 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2245 p->pack_name, (uintmax_t)obj_offset);
2248 int do_check_packed_object_crc;
2250 #define UNPACK_ENTRY_STACK_PREALLOC 64
2251 struct unpack_entry_stack_ent {
2252 off_t obj_offset;
2253 off_t curpos;
2254 unsigned long size;
2257 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2258 enum object_type *final_type, unsigned long *final_size)
2260 struct pack_window *w_curs = NULL;
2261 off_t curpos = obj_offset;
2262 void *data = NULL;
2263 unsigned long size;
2264 enum object_type type;
2265 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2266 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2267 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2268 int base_from_cache = 0;
2270 write_pack_access_log(p, obj_offset);
2272 /* PHASE 1: drill down to the innermost base object */
2273 for (;;) {
2274 off_t base_offset;
2275 int i;
2276 struct delta_base_cache_entry *ent;
2278 ent = get_delta_base_cache_entry(p, curpos);
2279 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2280 type = ent->type;
2281 data = ent->data;
2282 size = ent->size;
2283 clear_delta_base_cache_entry(ent);
2284 base_from_cache = 1;
2285 break;
2288 if (do_check_packed_object_crc && p->index_version > 1) {
2289 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2290 unsigned long len = revidx[1].offset - obj_offset;
2291 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2292 const unsigned char *sha1 =
2293 nth_packed_object_sha1(p, revidx->nr);
2294 error("bad packed object CRC for %s",
2295 sha1_to_hex(sha1));
2296 mark_bad_packed_object(p, sha1);
2297 unuse_pack(&w_curs);
2298 return NULL;
2302 type = unpack_object_header(p, &w_curs, &curpos, &size);
2303 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2304 break;
2306 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2307 if (!base_offset) {
2308 error("failed to validate delta base reference "
2309 "at offset %"PRIuMAX" from %s",
2310 (uintmax_t)curpos, p->pack_name);
2311 /* bail to phase 2, in hopes of recovery */
2312 data = NULL;
2313 break;
2316 /* push object, proceed to base */
2317 if (delta_stack_nr >= delta_stack_alloc
2318 && delta_stack == small_delta_stack) {
2319 delta_stack_alloc = alloc_nr(delta_stack_nr);
2320 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2321 memcpy(delta_stack, small_delta_stack,
2322 sizeof(*delta_stack)*delta_stack_nr);
2323 } else {
2324 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2326 i = delta_stack_nr++;
2327 delta_stack[i].obj_offset = obj_offset;
2328 delta_stack[i].curpos = curpos;
2329 delta_stack[i].size = size;
2331 curpos = obj_offset = base_offset;
2334 /* PHASE 2: handle the base */
2335 switch (type) {
2336 case OBJ_OFS_DELTA:
2337 case OBJ_REF_DELTA:
2338 if (data)
2339 die("BUG in unpack_entry: left loop at a valid delta");
2340 break;
2341 case OBJ_COMMIT:
2342 case OBJ_TREE:
2343 case OBJ_BLOB:
2344 case OBJ_TAG:
2345 if (!base_from_cache)
2346 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2347 break;
2348 default:
2349 data = NULL;
2350 error("unknown object type %i at offset %"PRIuMAX" in %s",
2351 type, (uintmax_t)obj_offset, p->pack_name);
2354 /* PHASE 3: apply deltas in order */
2356 /* invariants:
2357 * 'data' holds the base data, or NULL if there was corruption
2359 while (delta_stack_nr) {
2360 void *delta_data;
2361 void *base = data;
2362 unsigned long delta_size, base_size = size;
2363 int i;
2365 data = NULL;
2367 if (base)
2368 add_delta_base_cache(p, obj_offset, base, base_size, type);
2370 if (!base) {
2372 * We're probably in deep shit, but let's try to fetch
2373 * the required base anyway from another pack or loose.
2374 * This is costly but should happen only in the presence
2375 * of a corrupted pack, and is better than failing outright.
2377 struct revindex_entry *revidx;
2378 const unsigned char *base_sha1;
2379 revidx = find_pack_revindex(p, obj_offset);
2380 if (revidx) {
2381 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2382 error("failed to read delta base object %s"
2383 " at offset %"PRIuMAX" from %s",
2384 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2385 p->pack_name);
2386 mark_bad_packed_object(p, base_sha1);
2387 base = read_object(base_sha1, &type, &base_size);
2391 i = --delta_stack_nr;
2392 obj_offset = delta_stack[i].obj_offset;
2393 curpos = delta_stack[i].curpos;
2394 delta_size = delta_stack[i].size;
2396 if (!base)
2397 continue;
2399 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2401 if (!delta_data) {
2402 error("failed to unpack compressed delta "
2403 "at offset %"PRIuMAX" from %s",
2404 (uintmax_t)curpos, p->pack_name);
2405 data = NULL;
2406 continue;
2409 data = patch_delta(base, base_size,
2410 delta_data, delta_size,
2411 &size);
2414 * We could not apply the delta; warn the user, but keep going.
2415 * Our failure will be noticed either in the next iteration of
2416 * the loop, or if this is the final delta, in the caller when
2417 * we return NULL. Those code paths will take care of making
2418 * a more explicit warning and retrying with another copy of
2419 * the object.
2421 if (!data)
2422 error("failed to apply delta");
2424 free(delta_data);
2427 *final_type = type;
2428 *final_size = size;
2430 unuse_pack(&w_curs);
2432 if (delta_stack != small_delta_stack)
2433 free(delta_stack);
2435 return data;
2438 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2439 uint32_t n)
2441 const unsigned char *index = p->index_data;
2442 if (!index) {
2443 if (open_pack_index(p))
2444 return NULL;
2445 index = p->index_data;
2447 if (n >= p->num_objects)
2448 return NULL;
2449 index += 4 * 256;
2450 if (p->index_version == 1) {
2451 return index + 24 * n + 4;
2452 } else {
2453 index += 8;
2454 return index + 20 * n;
2458 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2460 const unsigned char *index = p->index_data;
2461 index += 4 * 256;
2462 if (p->index_version == 1) {
2463 return ntohl(*((uint32_t *)(index + 24 * n)));
2464 } else {
2465 uint32_t off;
2466 index += 8 + p->num_objects * (20 + 4);
2467 off = ntohl(*((uint32_t *)(index + 4 * n)));
2468 if (!(off & 0x80000000))
2469 return off;
2470 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2471 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2472 ntohl(*((uint32_t *)(index + 4)));
2476 off_t find_pack_entry_one(const unsigned char *sha1,
2477 struct packed_git *p)
2479 const uint32_t *level1_ofs = p->index_data;
2480 const unsigned char *index = p->index_data;
2481 unsigned hi, lo, stride;
2482 static int use_lookup = -1;
2483 static int debug_lookup = -1;
2485 if (debug_lookup < 0)
2486 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2488 if (!index) {
2489 if (open_pack_index(p))
2490 return 0;
2491 level1_ofs = p->index_data;
2492 index = p->index_data;
2494 if (p->index_version > 1) {
2495 level1_ofs += 2;
2496 index += 8;
2498 index += 4 * 256;
2499 hi = ntohl(level1_ofs[*sha1]);
2500 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2501 if (p->index_version > 1) {
2502 stride = 20;
2503 } else {
2504 stride = 24;
2505 index += 4;
2508 if (debug_lookup)
2509 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2510 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2512 if (use_lookup < 0)
2513 use_lookup = !!getenv("GIT_USE_LOOKUP");
2514 if (use_lookup) {
2515 int pos = sha1_entry_pos(index, stride, 0,
2516 lo, hi, p->num_objects, sha1);
2517 if (pos < 0)
2518 return 0;
2519 return nth_packed_object_offset(p, pos);
2522 do {
2523 unsigned mi = (lo + hi) / 2;
2524 int cmp = hashcmp(index + mi * stride, sha1);
2526 if (debug_lookup)
2527 printf("lo %u hi %u rg %u mi %u\n",
2528 lo, hi, hi - lo, mi);
2529 if (!cmp)
2530 return nth_packed_object_offset(p, mi);
2531 if (cmp > 0)
2532 hi = mi;
2533 else
2534 lo = mi+1;
2535 } while (lo < hi);
2536 return 0;
2539 int is_pack_valid(struct packed_git *p)
2541 /* An already open pack is known to be valid. */
2542 if (p->pack_fd != -1)
2543 return 1;
2545 /* If the pack has one window completely covering the
2546 * file size, the pack is known to be valid even if
2547 * the descriptor is not currently open.
2549 if (p->windows) {
2550 struct pack_window *w = p->windows;
2552 if (!w->offset && w->len == p->pack_size)
2553 return 1;
2556 /* Force the pack to open to prove its valid. */
2557 return !open_packed_git(p);
2560 static int fill_pack_entry(const unsigned char *sha1,
2561 struct pack_entry *e,
2562 struct packed_git *p)
2564 off_t offset;
2566 if (p->num_bad_objects) {
2567 unsigned i;
2568 for (i = 0; i < p->num_bad_objects; i++)
2569 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2570 return 0;
2573 offset = find_pack_entry_one(sha1, p);
2574 if (!offset)
2575 return 0;
2578 * We are about to tell the caller where they can locate the
2579 * requested object. We better make sure the packfile is
2580 * still here and can be accessed before supplying that
2581 * answer, as it may have been deleted since the index was
2582 * loaded!
2584 if (!is_pack_valid(p))
2585 return 0;
2586 e->offset = offset;
2587 e->p = p;
2588 hashcpy(e->sha1, sha1);
2589 return 1;
2593 * Iff a pack file contains the object named by sha1, return true and
2594 * store its location to e.
2596 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2598 struct packed_git *p;
2600 prepare_packed_git();
2601 if (!packed_git)
2602 return 0;
2604 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2605 return 1;
2607 for (p = packed_git; p; p = p->next) {
2608 if (p == last_found_pack)
2609 continue; /* we already checked this one */
2611 if (fill_pack_entry(sha1, e, p)) {
2612 last_found_pack = p;
2613 return 1;
2616 return 0;
2619 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2620 struct packed_git *packs)
2622 struct packed_git *p;
2624 for (p = packs; p; p = p->next) {
2625 if (find_pack_entry_one(sha1, p))
2626 return p;
2628 return NULL;
2632 static int sha1_loose_object_info(const unsigned char *sha1,
2633 struct object_info *oi,
2634 int flags)
2636 int status = 0;
2637 unsigned long mapsize;
2638 void *map;
2639 git_zstream stream;
2640 char hdr[32];
2641 struct strbuf hdrbuf = STRBUF_INIT;
2643 if (oi->delta_base_sha1)
2644 hashclr(oi->delta_base_sha1);
2647 * If we don't care about type or size, then we don't
2648 * need to look inside the object at all. Note that we
2649 * do not optimize out the stat call, even if the
2650 * caller doesn't care about the disk-size, since our
2651 * return value implicitly indicates whether the
2652 * object even exists.
2654 if (!oi->typep && !oi->typename && !oi->sizep) {
2655 struct stat st;
2656 if (stat_sha1_file(sha1, &st) < 0)
2657 return -1;
2658 if (oi->disk_sizep)
2659 *oi->disk_sizep = st.st_size;
2660 return 0;
2663 map = map_sha1_file(sha1, &mapsize);
2664 if (!map)
2665 return -1;
2666 if (oi->disk_sizep)
2667 *oi->disk_sizep = mapsize;
2668 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2669 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2670 status = error("unable to unpack %s header with --allow-unknown-type",
2671 sha1_to_hex(sha1));
2672 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2673 status = error("unable to unpack %s header",
2674 sha1_to_hex(sha1));
2675 if (status < 0)
2676 ; /* Do nothing */
2677 else if (hdrbuf.len) {
2678 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2679 status = error("unable to parse %s header with --allow-unknown-type",
2680 sha1_to_hex(sha1));
2681 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2682 status = error("unable to parse %s header", sha1_to_hex(sha1));
2683 git_inflate_end(&stream);
2684 munmap(map, mapsize);
2685 if (status && oi->typep)
2686 *oi->typep = status;
2687 strbuf_release(&hdrbuf);
2688 return 0;
2691 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2693 struct cached_object *co;
2694 struct pack_entry e;
2695 int rtype;
2696 enum object_type real_type;
2697 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2699 co = find_cached_object(real);
2700 if (co) {
2701 if (oi->typep)
2702 *(oi->typep) = co->type;
2703 if (oi->sizep)
2704 *(oi->sizep) = co->size;
2705 if (oi->disk_sizep)
2706 *(oi->disk_sizep) = 0;
2707 if (oi->delta_base_sha1)
2708 hashclr(oi->delta_base_sha1);
2709 if (oi->typename)
2710 strbuf_addstr(oi->typename, typename(co->type));
2711 oi->whence = OI_CACHED;
2712 return 0;
2715 if (!find_pack_entry(real, &e)) {
2716 /* Most likely it's a loose object. */
2717 if (!sha1_loose_object_info(real, oi, flags)) {
2718 oi->whence = OI_LOOSE;
2719 return 0;
2722 /* Not a loose object; someone else may have just packed it. */
2723 reprepare_packed_git();
2724 if (!find_pack_entry(real, &e))
2725 return -1;
2729 * packed_object_info() does not follow the delta chain to
2730 * find out the real type, unless it is given oi->typep.
2732 if (oi->typename && !oi->typep)
2733 oi->typep = &real_type;
2735 rtype = packed_object_info(e.p, e.offset, oi);
2736 if (rtype < 0) {
2737 mark_bad_packed_object(e.p, real);
2738 if (oi->typep == &real_type)
2739 oi->typep = NULL;
2740 return sha1_object_info_extended(real, oi, 0);
2741 } else if (in_delta_base_cache(e.p, e.offset)) {
2742 oi->whence = OI_DBCACHED;
2743 } else {
2744 oi->whence = OI_PACKED;
2745 oi->u.packed.offset = e.offset;
2746 oi->u.packed.pack = e.p;
2747 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2748 rtype == OBJ_OFS_DELTA);
2750 if (oi->typename)
2751 strbuf_addstr(oi->typename, typename(*oi->typep));
2752 if (oi->typep == &real_type)
2753 oi->typep = NULL;
2755 return 0;
2758 /* returns enum object_type or negative */
2759 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2761 enum object_type type;
2762 struct object_info oi = {NULL};
2764 oi.typep = &type;
2765 oi.sizep = sizep;
2766 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2767 return -1;
2768 return type;
2771 static void *read_packed_sha1(const unsigned char *sha1,
2772 enum object_type *type, unsigned long *size)
2774 struct pack_entry e;
2775 void *data;
2777 if (!find_pack_entry(sha1, &e))
2778 return NULL;
2779 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2780 if (!data) {
2782 * We're probably in deep shit, but let's try to fetch
2783 * the required object anyway from another pack or loose.
2784 * This should happen only in the presence of a corrupted
2785 * pack, and is better than failing outright.
2787 error("failed to read object %s at offset %"PRIuMAX" from %s",
2788 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2789 mark_bad_packed_object(e.p, sha1);
2790 data = read_object(sha1, type, size);
2792 return data;
2795 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2796 unsigned char *sha1)
2798 struct cached_object *co;
2800 hash_sha1_file(buf, len, typename(type), sha1);
2801 if (has_sha1_file(sha1) || find_cached_object(sha1))
2802 return 0;
2803 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2804 co = &cached_objects[cached_object_nr++];
2805 co->size = len;
2806 co->type = type;
2807 co->buf = xmalloc(len);
2808 memcpy(co->buf, buf, len);
2809 hashcpy(co->sha1, sha1);
2810 return 0;
2813 static void *read_object(const unsigned char *sha1, enum object_type *type,
2814 unsigned long *size)
2816 unsigned long mapsize;
2817 void *map, *buf;
2818 struct cached_object *co;
2820 co = find_cached_object(sha1);
2821 if (co) {
2822 *type = co->type;
2823 *size = co->size;
2824 return xmemdupz(co->buf, co->size);
2827 buf = read_packed_sha1(sha1, type, size);
2828 if (buf)
2829 return buf;
2830 map = map_sha1_file(sha1, &mapsize);
2831 if (map) {
2832 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2833 munmap(map, mapsize);
2834 return buf;
2836 reprepare_packed_git();
2837 return read_packed_sha1(sha1, type, size);
2841 * This function dies on corrupt objects; the callers who want to
2842 * deal with them should arrange to call read_object() and give error
2843 * messages themselves.
2845 void *read_sha1_file_extended(const unsigned char *sha1,
2846 enum object_type *type,
2847 unsigned long *size,
2848 unsigned flag)
2850 void *data;
2851 const struct packed_git *p;
2852 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2854 errno = 0;
2855 data = read_object(repl, type, size);
2856 if (data)
2857 return data;
2859 if (errno && errno != ENOENT)
2860 die_errno("failed to read object %s", sha1_to_hex(sha1));
2862 /* die if we replaced an object with one that does not exist */
2863 if (repl != sha1)
2864 die("replacement %s not found for %s",
2865 sha1_to_hex(repl), sha1_to_hex(sha1));
2867 if (has_loose_object(repl)) {
2868 const char *path = sha1_file_name(sha1);
2870 die("loose object %s (stored in %s) is corrupt",
2871 sha1_to_hex(repl), path);
2874 if ((p = has_packed_and_bad(repl)) != NULL)
2875 die("packed object %s (stored in %s) is corrupt",
2876 sha1_to_hex(repl), p->pack_name);
2878 return NULL;
2881 void *read_object_with_reference(const unsigned char *sha1,
2882 const char *required_type_name,
2883 unsigned long *size,
2884 unsigned char *actual_sha1_return)
2886 enum object_type type, required_type;
2887 void *buffer;
2888 unsigned long isize;
2889 unsigned char actual_sha1[20];
2891 required_type = type_from_string(required_type_name);
2892 hashcpy(actual_sha1, sha1);
2893 while (1) {
2894 int ref_length = -1;
2895 const char *ref_type = NULL;
2897 buffer = read_sha1_file(actual_sha1, &type, &isize);
2898 if (!buffer)
2899 return NULL;
2900 if (type == required_type) {
2901 *size = isize;
2902 if (actual_sha1_return)
2903 hashcpy(actual_sha1_return, actual_sha1);
2904 return buffer;
2906 /* Handle references */
2907 else if (type == OBJ_COMMIT)
2908 ref_type = "tree ";
2909 else if (type == OBJ_TAG)
2910 ref_type = "object ";
2911 else {
2912 free(buffer);
2913 return NULL;
2915 ref_length = strlen(ref_type);
2917 if (ref_length + 40 > isize ||
2918 memcmp(buffer, ref_type, ref_length) ||
2919 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2920 free(buffer);
2921 return NULL;
2923 free(buffer);
2924 /* Now we have the ID of the referred-to object in
2925 * actual_sha1. Check again. */
2929 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2930 const char *type, unsigned char *sha1,
2931 char *hdr, int *hdrlen)
2933 git_SHA_CTX c;
2935 /* Generate the header */
2936 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2938 /* Sha1.. */
2939 git_SHA1_Init(&c);
2940 git_SHA1_Update(&c, hdr, *hdrlen);
2941 git_SHA1_Update(&c, buf, len);
2942 git_SHA1_Final(sha1, &c);
2946 * Move the just written object into its final resting place.
2948 int finalize_object_file(const char *tmpfile, const char *filename)
2950 int ret = 0;
2952 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2953 goto try_rename;
2954 else if (link(tmpfile, filename))
2955 ret = errno;
2958 * Coda hack - coda doesn't like cross-directory links,
2959 * so we fall back to a rename, which will mean that it
2960 * won't be able to check collisions, but that's not a
2961 * big deal.
2963 * The same holds for FAT formatted media.
2965 * When this succeeds, we just return. We have nothing
2966 * left to unlink.
2968 if (ret && ret != EEXIST) {
2969 try_rename:
2970 if (!rename(tmpfile, filename))
2971 goto out;
2972 ret = errno;
2974 unlink_or_warn(tmpfile);
2975 if (ret) {
2976 if (ret != EEXIST) {
2977 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2979 /* FIXME!!! Collision check here ? */
2982 out:
2983 if (adjust_shared_perm(filename))
2984 return error("unable to set permission to '%s'", filename);
2985 return 0;
2988 static int write_buffer(int fd, const void *buf, size_t len)
2990 if (write_in_full(fd, buf, len) < 0)
2991 return error("file write error (%s)", strerror(errno));
2992 return 0;
2995 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2996 unsigned char *sha1)
2998 char hdr[32];
2999 int hdrlen;
3000 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3001 return 0;
3004 /* Finalize a file on disk, and close it. */
3005 static void close_sha1_file(int fd)
3007 if (fsync_object_files)
3008 fsync_or_die(fd, "sha1 file");
3009 if (close(fd) != 0)
3010 die_errno("error when closing sha1 file");
3013 /* Size of directory component, including the ending '/' */
3014 static inline int directory_size(const char *filename)
3016 const char *s = strrchr(filename, '/');
3017 if (!s)
3018 return 0;
3019 return s - filename + 1;
3023 * This creates a temporary file in the same directory as the final
3024 * 'filename'
3026 * We want to avoid cross-directory filename renames, because those
3027 * can have problems on various filesystems (FAT, NFS, Coda).
3029 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3031 int fd, dirlen = directory_size(filename);
3033 if (dirlen + 20 > bufsiz) {
3034 errno = ENAMETOOLONG;
3035 return -1;
3037 memcpy(buffer, filename, dirlen);
3038 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3039 fd = git_mkstemp_mode(buffer, 0444);
3040 if (fd < 0 && dirlen && errno == ENOENT) {
3041 /* Make sure the directory exists */
3042 memcpy(buffer, filename, dirlen);
3043 buffer[dirlen-1] = 0;
3044 if (mkdir(buffer, 0777) && errno != EEXIST)
3045 return -1;
3046 if (adjust_shared_perm(buffer))
3047 return -1;
3049 /* Try again */
3050 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3051 fd = git_mkstemp_mode(buffer, 0444);
3053 return fd;
3056 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3057 const void *buf, unsigned long len, time_t mtime)
3059 int fd, ret;
3060 unsigned char compressed[4096];
3061 git_zstream stream;
3062 git_SHA_CTX c;
3063 unsigned char parano_sha1[20];
3064 static char tmp_file[PATH_MAX];
3065 const char *filename = sha1_file_name(sha1);
3067 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3068 if (fd < 0) {
3069 if (errno == EACCES)
3070 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3071 else
3072 return error("unable to create temporary file: %s", strerror(errno));
3075 /* Set it up */
3076 git_deflate_init(&stream, zlib_compression_level);
3077 stream.next_out = compressed;
3078 stream.avail_out = sizeof(compressed);
3079 git_SHA1_Init(&c);
3081 /* First header.. */
3082 stream.next_in = (unsigned char *)hdr;
3083 stream.avail_in = hdrlen;
3084 while (git_deflate(&stream, 0) == Z_OK)
3085 ; /* nothing */
3086 git_SHA1_Update(&c, hdr, hdrlen);
3088 /* Then the data itself.. */
3089 stream.next_in = (void *)buf;
3090 stream.avail_in = len;
3091 do {
3092 unsigned char *in0 = stream.next_in;
3093 ret = git_deflate(&stream, Z_FINISH);
3094 git_SHA1_Update(&c, in0, stream.next_in - in0);
3095 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3096 die("unable to write sha1 file");
3097 stream.next_out = compressed;
3098 stream.avail_out = sizeof(compressed);
3099 } while (ret == Z_OK);
3101 if (ret != Z_STREAM_END)
3102 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3103 ret = git_deflate_end_gently(&stream);
3104 if (ret != Z_OK)
3105 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3106 git_SHA1_Final(parano_sha1, &c);
3107 if (hashcmp(sha1, parano_sha1) != 0)
3108 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3110 close_sha1_file(fd);
3112 if (mtime) {
3113 struct utimbuf utb;
3114 utb.actime = mtime;
3115 utb.modtime = mtime;
3116 if (utime(tmp_file, &utb) < 0)
3117 warning("failed utime() on %s: %s",
3118 tmp_file, strerror(errno));
3121 return finalize_object_file(tmp_file, filename);
3124 static int freshen_loose_object(const unsigned char *sha1)
3126 return check_and_freshen(sha1, 1);
3129 static int freshen_packed_object(const unsigned char *sha1)
3131 struct pack_entry e;
3132 if (!find_pack_entry(sha1, &e))
3133 return 0;
3134 if (e.p->freshened)
3135 return 1;
3136 if (!freshen_file(e.p->pack_name))
3137 return 0;
3138 e.p->freshened = 1;
3139 return 1;
3142 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3144 char hdr[32];
3145 int hdrlen;
3147 /* Normally if we have it in the pack then we do not bother writing
3148 * it out into .git/objects/??/?{38} file.
3150 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3151 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3152 return 0;
3153 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3156 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3157 unsigned char *sha1, unsigned flags)
3159 char *header;
3160 int hdrlen, status = 0;
3162 /* type string, SP, %lu of the length plus NUL must fit this */
3163 header = xmalloc(strlen(type) + 32);
3164 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3166 if (!(flags & HASH_WRITE_OBJECT))
3167 goto cleanup;
3168 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3169 goto cleanup;
3170 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3172 cleanup:
3173 free(header);
3174 return status;
3177 int force_object_loose(const unsigned char *sha1, time_t mtime)
3179 void *buf;
3180 unsigned long len;
3181 enum object_type type;
3182 char hdr[32];
3183 int hdrlen;
3184 int ret;
3186 if (has_loose_object(sha1))
3187 return 0;
3188 buf = read_packed_sha1(sha1, &type, &len);
3189 if (!buf)
3190 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3191 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3192 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3193 free(buf);
3195 return ret;
3198 int has_pack_index(const unsigned char *sha1)
3200 struct stat st;
3201 if (stat(sha1_pack_index_name(sha1), &st))
3202 return 0;
3203 return 1;
3206 int has_sha1_pack(const unsigned char *sha1)
3208 struct pack_entry e;
3209 return find_pack_entry(sha1, &e);
3212 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3214 struct pack_entry e;
3216 if (find_pack_entry(sha1, &e))
3217 return 1;
3218 if (has_loose_object(sha1))
3219 return 1;
3220 if (flags & HAS_SHA1_QUICK)
3221 return 0;
3222 reprepare_packed_git();
3223 return find_pack_entry(sha1, &e);
3226 static void check_tree(const void *buf, size_t size)
3228 struct tree_desc desc;
3229 struct name_entry entry;
3231 init_tree_desc(&desc, buf, size);
3232 while (tree_entry(&desc, &entry))
3233 /* do nothing
3234 * tree_entry() will die() on malformed entries */
3238 static void check_commit(const void *buf, size_t size)
3240 struct commit c;
3241 memset(&c, 0, sizeof(c));
3242 if (parse_commit_buffer(&c, buf, size))
3243 die("corrupt commit");
3246 static void check_tag(const void *buf, size_t size)
3248 struct tag t;
3249 memset(&t, 0, sizeof(t));
3250 if (parse_tag_buffer(&t, buf, size))
3251 die("corrupt tag");
3254 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3255 enum object_type type,
3256 const char *path, unsigned flags)
3258 int ret, re_allocated = 0;
3259 int write_object = flags & HASH_WRITE_OBJECT;
3261 if (!type)
3262 type = OBJ_BLOB;
3265 * Convert blobs to git internal format
3267 if ((type == OBJ_BLOB) && path) {
3268 struct strbuf nbuf = STRBUF_INIT;
3269 if (convert_to_git(path, buf, size, &nbuf,
3270 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3271 buf = strbuf_detach(&nbuf, &size);
3272 re_allocated = 1;
3275 if (flags & HASH_FORMAT_CHECK) {
3276 if (type == OBJ_TREE)
3277 check_tree(buf, size);
3278 if (type == OBJ_COMMIT)
3279 check_commit(buf, size);
3280 if (type == OBJ_TAG)
3281 check_tag(buf, size);
3284 if (write_object)
3285 ret = write_sha1_file(buf, size, typename(type), sha1);
3286 else
3287 ret = hash_sha1_file(buf, size, typename(type), sha1);
3288 if (re_allocated)
3289 free(buf);
3290 return ret;
3293 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3294 const char *path, unsigned flags)
3296 int ret;
3297 const int write_object = flags & HASH_WRITE_OBJECT;
3298 struct strbuf sbuf = STRBUF_INIT;
3300 assert(path);
3301 assert(would_convert_to_git_filter_fd(path));
3303 convert_to_git_filter_fd(path, fd, &sbuf,
3304 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3306 if (write_object)
3307 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3308 sha1);
3309 else
3310 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3311 sha1);
3312 strbuf_release(&sbuf);
3313 return ret;
3316 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3317 const char *path, unsigned flags)
3319 struct strbuf sbuf = STRBUF_INIT;
3320 int ret;
3322 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3323 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3324 else
3325 ret = -1;
3326 strbuf_release(&sbuf);
3327 return ret;
3330 #define SMALL_FILE_SIZE (32*1024)
3332 static int index_core(unsigned char *sha1, int fd, size_t size,
3333 enum object_type type, const char *path,
3334 unsigned flags)
3336 int ret;
3338 if (!size) {
3339 ret = index_mem(sha1, "", size, type, path, flags);
3340 } else if (size <= SMALL_FILE_SIZE) {
3341 char *buf = xmalloc(size);
3342 if (size == read_in_full(fd, buf, size))
3343 ret = index_mem(sha1, buf, size, type, path, flags);
3344 else
3345 ret = error("short read %s", strerror(errno));
3346 free(buf);
3347 } else {
3348 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3349 ret = index_mem(sha1, buf, size, type, path, flags);
3350 munmap(buf, size);
3352 return ret;
3356 * This creates one packfile per large blob unless bulk-checkin
3357 * machinery is "plugged".
3359 * This also bypasses the usual "convert-to-git" dance, and that is on
3360 * purpose. We could write a streaming version of the converting
3361 * functions and insert that before feeding the data to fast-import
3362 * (or equivalent in-core API described above). However, that is
3363 * somewhat complicated, as we do not know the size of the filter
3364 * result, which we need to know beforehand when writing a git object.
3365 * Since the primary motivation for trying to stream from the working
3366 * tree file and to avoid mmaping it in core is to deal with large
3367 * binary blobs, they generally do not want to get any conversion, and
3368 * callers should avoid this code path when filters are requested.
3370 static int index_stream(unsigned char *sha1, int fd, size_t size,
3371 enum object_type type, const char *path,
3372 unsigned flags)
3374 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3377 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3378 enum object_type type, const char *path, unsigned flags)
3380 int ret;
3383 * Call xsize_t() only when needed to avoid potentially unnecessary
3384 * die() for large files.
3386 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3387 ret = index_stream_convert_blob(sha1, fd, path, flags);
3388 else if (!S_ISREG(st->st_mode))
3389 ret = index_pipe(sha1, fd, type, path, flags);
3390 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3391 (path && would_convert_to_git(path)))
3392 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3393 flags);
3394 else
3395 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3396 flags);
3397 close(fd);
3398 return ret;
3401 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3403 int fd;
3404 struct strbuf sb = STRBUF_INIT;
3406 switch (st->st_mode & S_IFMT) {
3407 case S_IFREG:
3408 fd = open(path, O_RDONLY);
3409 if (fd < 0)
3410 return error("open(\"%s\"): %s", path,
3411 strerror(errno));
3412 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3413 return error("%s: failed to insert into database",
3414 path);
3415 break;
3416 case S_IFLNK:
3417 if (strbuf_readlink(&sb, path, st->st_size)) {
3418 char *errstr = strerror(errno);
3419 return error("readlink(\"%s\"): %s", path,
3420 errstr);
3422 if (!(flags & HASH_WRITE_OBJECT))
3423 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3424 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3425 return error("%s: failed to insert into database",
3426 path);
3427 strbuf_release(&sb);
3428 break;
3429 case S_IFDIR:
3430 return resolve_gitlink_ref(path, "HEAD", sha1);
3431 default:
3432 return error("%s: unsupported file type", path);
3434 return 0;
3437 int read_pack_header(int fd, struct pack_header *header)
3439 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3440 /* "eof before pack header was fully read" */
3441 return PH_ERROR_EOF;
3443 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3444 /* "protocol error (pack signature mismatch detected)" */
3445 return PH_ERROR_PACK_SIGNATURE;
3446 if (!pack_version_ok(header->hdr_version))
3447 /* "protocol error (pack version unsupported)" */
3448 return PH_ERROR_PROTOCOL;
3449 return 0;
3452 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3454 enum object_type type = sha1_object_info(sha1, NULL);
3455 if (type < 0)
3456 die("%s is not a valid object", sha1_to_hex(sha1));
3457 if (type != expect)
3458 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3459 typename(expect));
3462 static int for_each_file_in_obj_subdir(int subdir_nr,
3463 struct strbuf *path,
3464 each_loose_object_fn obj_cb,
3465 each_loose_cruft_fn cruft_cb,
3466 each_loose_subdir_fn subdir_cb,
3467 void *data)
3469 size_t baselen = path->len;
3470 DIR *dir = opendir(path->buf);
3471 struct dirent *de;
3472 int r = 0;
3474 if (!dir) {
3475 if (errno == ENOENT)
3476 return 0;
3477 return error("unable to open %s: %s", path->buf, strerror(errno));
3480 while ((de = readdir(dir))) {
3481 if (is_dot_or_dotdot(de->d_name))
3482 continue;
3484 strbuf_setlen(path, baselen);
3485 strbuf_addf(path, "/%s", de->d_name);
3487 if (strlen(de->d_name) == 38) {
3488 char hex[41];
3489 unsigned char sha1[20];
3491 snprintf(hex, sizeof(hex), "%02x%s",
3492 subdir_nr, de->d_name);
3493 if (!get_sha1_hex(hex, sha1)) {
3494 if (obj_cb) {
3495 r = obj_cb(sha1, path->buf, data);
3496 if (r)
3497 break;
3499 continue;
3503 if (cruft_cb) {
3504 r = cruft_cb(de->d_name, path->buf, data);
3505 if (r)
3506 break;
3509 strbuf_setlen(path, baselen);
3511 if (!r && subdir_cb)
3512 r = subdir_cb(subdir_nr, path->buf, data);
3514 closedir(dir);
3515 return r;
3518 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3519 each_loose_object_fn obj_cb,
3520 each_loose_cruft_fn cruft_cb,
3521 each_loose_subdir_fn subdir_cb,
3522 void *data)
3524 size_t baselen = path->len;
3525 int r = 0;
3526 int i;
3528 for (i = 0; i < 256; i++) {
3529 strbuf_addf(path, "/%02x", i);
3530 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3531 subdir_cb, data);
3532 strbuf_setlen(path, baselen);
3533 if (r)
3534 break;
3537 return r;
3540 int for_each_loose_file_in_objdir(const char *path,
3541 each_loose_object_fn obj_cb,
3542 each_loose_cruft_fn cruft_cb,
3543 each_loose_subdir_fn subdir_cb,
3544 void *data)
3546 struct strbuf buf = STRBUF_INIT;
3547 int r;
3549 strbuf_addstr(&buf, path);
3550 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3551 subdir_cb, data);
3552 strbuf_release(&buf);
3554 return r;
3557 struct loose_alt_odb_data {
3558 each_loose_object_fn *cb;
3559 void *data;
3562 static int loose_from_alt_odb(struct alternate_object_database *alt,
3563 void *vdata)
3565 struct loose_alt_odb_data *data = vdata;
3566 struct strbuf buf = STRBUF_INIT;
3567 int r;
3569 /* copy base not including trailing '/' */
3570 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3571 r = for_each_loose_file_in_objdir_buf(&buf,
3572 data->cb, NULL, NULL,
3573 data->data);
3574 strbuf_release(&buf);
3575 return r;
3578 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3580 struct loose_alt_odb_data alt;
3581 int r;
3583 r = for_each_loose_file_in_objdir(get_object_directory(),
3584 cb, NULL, NULL, data);
3585 if (r)
3586 return r;
3588 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3589 return 0;
3591 alt.cb = cb;
3592 alt.data = data;
3593 return foreach_alt_odb(loose_from_alt_odb, &alt);
3596 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3598 uint32_t i;
3599 int r = 0;
3601 for (i = 0; i < p->num_objects; i++) {
3602 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3604 if (!sha1)
3605 return error("unable to get sha1 of object %u in %s",
3606 i, p->pack_name);
3608 r = cb(sha1, p, i, data);
3609 if (r)
3610 break;
3612 return r;
3615 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3617 struct packed_git *p;
3618 int r = 0;
3619 int pack_errors = 0;
3621 prepare_packed_git();
3622 for (p = packed_git; p; p = p->next) {
3623 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3624 continue;
3625 if (open_pack_index(p)) {
3626 pack_errors = 1;
3627 continue;
3629 r = for_each_object_in_pack(p, cb, data);
3630 if (r)
3631 break;
3633 return r ? r : pack_errors;