parse-options: inline parse_options_usage() at its only remaining caller
[git/raj.git] / sha1_file.c
blob3d56746a9be12e78c2b5262948bd411acf3230a2
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 struct strbuf *buf,
212 const char *which)
214 strbuf_reset(buf);
215 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
216 sha1_to_hex(sha1), which);
217 return buf->buf;
220 char *sha1_pack_name(const unsigned char *sha1)
222 static struct strbuf buf = STRBUF_INIT;
223 return sha1_get_pack_name(sha1, &buf, "pack");
226 char *sha1_pack_index_name(const unsigned char *sha1)
228 static struct strbuf buf = STRBUF_INIT;
229 return sha1_get_pack_name(sha1, &buf, "idx");
232 struct alternate_object_database *alt_odb_list;
233 static struct alternate_object_database **alt_odb_tail;
236 * Prepare alternate object database registry.
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
250 static int link_alt_odb_entry(const char *entry, const char *relative_base,
251 int depth, const char *normalized_objdir)
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
258 if (!is_absolute_path(entry) && relative_base) {
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
262 strbuf_addstr(&pathbuf, entry);
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
266 pfxlen = strlen(pathbuf.buf);
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284 /* Detect cases where alternate disappeared */
285 if (!is_directory(ent->base)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (pfxlen == alt->name - alt->base - 1 &&
298 !memcmp(ent->base, alt->base, pfxlen)) {
299 free(ent);
300 return -1;
303 if (!strcmp_icase(ent->base, normalized_objdir)) {
304 free(ent);
305 return -1;
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
316 ent->base[pfxlen] = '/';
318 return 0;
321 static void link_alt_odb_entries(const char *alt, int len, int sep,
322 const char *relative_base, int depth)
324 struct string_list entries = STRING_LIST_INIT_NODUP;
325 char *alt_copy;
326 int i;
327 struct strbuf objdirbuf = STRBUF_INIT;
329 if (depth > 5) {
330 error("%s: ignoring alternate object stores, nesting too deep.",
331 relative_base);
332 return;
335 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
336 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
338 alt_copy = xmemdupz(alt, len);
339 string_list_split_in_place(&entries, alt_copy, sep, -1);
340 for (i = 0; i < entries.nr; i++) {
341 const char *entry = entries.items[i].string;
342 if (entry[0] == '\0' || entry[0] == '#')
343 continue;
344 if (!is_absolute_path(entry) && depth) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base, entry);
347 } else {
348 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
351 string_list_clear(&entries, 0);
352 free(alt_copy);
353 strbuf_release(&objdirbuf);
356 void read_info_alternates(const char * relative_base, int depth)
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 char *path;
362 int fd;
364 path = xstrfmt("%s/info/alternates", relative_base);
365 fd = git_open_noatime(path);
366 free(path);
367 if (fd < 0)
368 return;
369 if (fstat(fd, &st) || (st.st_size == 0)) {
370 close(fd);
371 return;
373 mapsz = xsize_t(st.st_size);
374 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
375 close(fd);
377 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
379 munmap(map, mapsz);
382 void add_to_alternates_file(const char *reference)
384 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
385 char *alts = git_pathdup("objects/info/alternates");
386 FILE *in, *out;
388 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
389 out = fdopen_lock_file(lock, "w");
390 if (!out)
391 die_errno("unable to fdopen alternates lockfile");
393 in = fopen(alts, "r");
394 if (in) {
395 struct strbuf line = STRBUF_INIT;
396 int found = 0;
398 while (strbuf_getline(&line, in, '\n') != EOF) {
399 if (!strcmp(reference, line.buf)) {
400 found = 1;
401 break;
403 fprintf_or_die(out, "%s\n", line.buf);
406 strbuf_release(&line);
407 fclose(in);
409 if (found) {
410 rollback_lock_file(lock);
411 lock = NULL;
414 else if (errno != ENOENT)
415 die_errno("unable to read alternates file");
417 if (lock) {
418 fprintf_or_die(out, "%s\n", reference);
419 if (commit_lock_file(lock))
420 die_errno("unable to move new alternates file into place");
421 if (alt_odb_tail)
422 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
424 free(alts);
427 int foreach_alt_odb(alt_odb_fn fn, void *cb)
429 struct alternate_object_database *ent;
430 int r = 0;
432 prepare_alt_odb();
433 for (ent = alt_odb_list; ent; ent = ent->next) {
434 r = fn(ent, cb);
435 if (r)
436 break;
438 return r;
441 void prepare_alt_odb(void)
443 const char *alt;
445 if (alt_odb_tail)
446 return;
448 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
449 if (!alt) alt = "";
451 alt_odb_tail = &alt_odb_list;
452 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
454 read_info_alternates(get_object_directory(), 0);
457 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
458 static int freshen_file(const char *fn)
460 struct utimbuf t;
461 t.actime = t.modtime = time(NULL);
462 return !utime(fn, &t);
466 * All of the check_and_freshen functions return 1 if the file exists and was
467 * freshened (if freshening was requested), 0 otherwise. If they return
468 * 0, you should not assume that it is safe to skip a write of the object (it
469 * either does not exist on disk, or has a stale mtime and may be subject to
470 * pruning).
472 static int check_and_freshen_file(const char *fn, int freshen)
474 if (access(fn, F_OK))
475 return 0;
476 if (freshen && !freshen_file(fn))
477 return 0;
478 return 1;
481 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
483 return check_and_freshen_file(sha1_file_name(sha1), freshen);
486 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
488 struct alternate_object_database *alt;
489 prepare_alt_odb();
490 for (alt = alt_odb_list; alt; alt = alt->next) {
491 fill_sha1_path(alt->name, sha1);
492 if (check_and_freshen_file(alt->base, freshen))
493 return 1;
495 return 0;
498 static int check_and_freshen(const unsigned char *sha1, int freshen)
500 return check_and_freshen_local(sha1, freshen) ||
501 check_and_freshen_nonlocal(sha1, freshen);
504 int has_loose_object_nonlocal(const unsigned char *sha1)
506 return check_and_freshen_nonlocal(sha1, 0);
509 static int has_loose_object(const unsigned char *sha1)
511 return check_and_freshen(sha1, 0);
514 static unsigned int pack_used_ctr;
515 static unsigned int pack_mmap_calls;
516 static unsigned int peak_pack_open_windows;
517 static unsigned int pack_open_windows;
518 static unsigned int pack_open_fds;
519 static unsigned int pack_max_fds;
520 static size_t peak_pack_mapped;
521 static size_t pack_mapped;
522 struct packed_git *packed_git;
524 void pack_report(void)
526 fprintf(stderr,
527 "pack_report: getpagesize() = %10" SZ_FMT "\n"
528 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
529 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
530 sz_fmt(getpagesize()),
531 sz_fmt(packed_git_window_size),
532 sz_fmt(packed_git_limit));
533 fprintf(stderr,
534 "pack_report: pack_used_ctr = %10u\n"
535 "pack_report: pack_mmap_calls = %10u\n"
536 "pack_report: pack_open_windows = %10u / %10u\n"
537 "pack_report: pack_mapped = "
538 "%10" SZ_FMT " / %10" SZ_FMT "\n",
539 pack_used_ctr,
540 pack_mmap_calls,
541 pack_open_windows, peak_pack_open_windows,
542 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
546 * Open and mmap the index file at path, perform a couple of
547 * consistency checks, then record its information to p. Return 0 on
548 * success.
550 static int check_packed_git_idx(const char *path, struct packed_git *p)
552 void *idx_map;
553 struct pack_idx_header *hdr;
554 size_t idx_size;
555 uint32_t version, nr, i, *index;
556 int fd = git_open_noatime(path);
557 struct stat st;
559 if (fd < 0)
560 return -1;
561 if (fstat(fd, &st)) {
562 close(fd);
563 return -1;
565 idx_size = xsize_t(st.st_size);
566 if (idx_size < 4 * 256 + 20 + 20) {
567 close(fd);
568 return error("index file %s is too small", path);
570 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
571 close(fd);
573 hdr = idx_map;
574 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
575 version = ntohl(hdr->idx_version);
576 if (version < 2 || version > 2) {
577 munmap(idx_map, idx_size);
578 return error("index file %s is version %"PRIu32
579 " and is not supported by this binary"
580 " (try upgrading GIT to a newer version)",
581 path, version);
583 } else
584 version = 1;
586 nr = 0;
587 index = idx_map;
588 if (version > 1)
589 index += 2; /* skip index header */
590 for (i = 0; i < 256; i++) {
591 uint32_t n = ntohl(index[i]);
592 if (n < nr) {
593 munmap(idx_map, idx_size);
594 return error("non-monotonic index %s", path);
596 nr = n;
599 if (version == 1) {
601 * Total size:
602 * - 256 index entries 4 bytes each
603 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
604 * - 20-byte SHA1 of the packfile
605 * - 20-byte SHA1 file checksum
607 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
608 munmap(idx_map, idx_size);
609 return error("wrong index v1 file size in %s", path);
611 } else if (version == 2) {
613 * Minimum size:
614 * - 8 bytes of header
615 * - 256 index entries 4 bytes each
616 * - 20-byte sha1 entry * nr
617 * - 4-byte crc entry * nr
618 * - 4-byte offset entry * nr
619 * - 20-byte SHA1 of the packfile
620 * - 20-byte SHA1 file checksum
621 * And after the 4-byte offset table might be a
622 * variable sized table containing 8-byte entries
623 * for offsets larger than 2^31.
625 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
626 unsigned long max_size = min_size;
627 if (nr)
628 max_size += (nr - 1)*8;
629 if (idx_size < min_size || idx_size > max_size) {
630 munmap(idx_map, idx_size);
631 return error("wrong index v2 file size in %s", path);
633 if (idx_size != min_size &&
635 * make sure we can deal with large pack offsets.
636 * 31-bit signed offset won't be enough, neither
637 * 32-bit unsigned one will be.
639 (sizeof(off_t) <= 4)) {
640 munmap(idx_map, idx_size);
641 return error("pack too large for current definition of off_t in %s", path);
645 p->index_version = version;
646 p->index_data = idx_map;
647 p->index_size = idx_size;
648 p->num_objects = nr;
649 return 0;
652 int open_pack_index(struct packed_git *p)
654 char *idx_name;
655 size_t len;
656 int ret;
658 if (p->index_data)
659 return 0;
661 if (!strip_suffix(p->pack_name, ".pack", &len))
662 die("BUG: pack_name does not end in .pack");
663 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
664 ret = check_packed_git_idx(idx_name, p);
665 free(idx_name);
666 return ret;
669 static void scan_windows(struct packed_git *p,
670 struct packed_git **lru_p,
671 struct pack_window **lru_w,
672 struct pack_window **lru_l)
674 struct pack_window *w, *w_l;
676 for (w_l = NULL, w = p->windows; w; w = w->next) {
677 if (!w->inuse_cnt) {
678 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
679 *lru_p = p;
680 *lru_w = w;
681 *lru_l = w_l;
684 w_l = w;
688 static int unuse_one_window(struct packed_git *current)
690 struct packed_git *p, *lru_p = NULL;
691 struct pack_window *lru_w = NULL, *lru_l = NULL;
693 if (current)
694 scan_windows(current, &lru_p, &lru_w, &lru_l);
695 for (p = packed_git; p; p = p->next)
696 scan_windows(p, &lru_p, &lru_w, &lru_l);
697 if (lru_p) {
698 munmap(lru_w->base, lru_w->len);
699 pack_mapped -= lru_w->len;
700 if (lru_l)
701 lru_l->next = lru_w->next;
702 else
703 lru_p->windows = lru_w->next;
704 free(lru_w);
705 pack_open_windows--;
706 return 1;
708 return 0;
711 void release_pack_memory(size_t need)
713 size_t cur = pack_mapped;
714 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
715 ; /* nothing */
718 static void mmap_limit_check(size_t length)
720 static size_t limit = 0;
721 if (!limit) {
722 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
723 if (!limit)
724 limit = SIZE_MAX;
726 if (length > limit)
727 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
728 (uintmax_t)length, (uintmax_t)limit);
731 void *xmmap_gently(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
734 void *ret;
736 mmap_limit_check(length);
737 ret = mmap(start, length, prot, flags, fd, offset);
738 if (ret == MAP_FAILED) {
739 if (!length)
740 return NULL;
741 release_pack_memory(length);
742 ret = mmap(start, length, prot, flags, fd, offset);
744 return ret;
747 void *xmmap(void *start, size_t length,
748 int prot, int flags, int fd, off_t offset)
750 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
751 if (ret == MAP_FAILED)
752 die_errno("mmap failed");
753 return ret;
756 void close_pack_windows(struct packed_git *p)
758 while (p->windows) {
759 struct pack_window *w = p->windows;
761 if (w->inuse_cnt)
762 die("pack '%s' still has open windows to it",
763 p->pack_name);
764 munmap(w->base, w->len);
765 pack_mapped -= w->len;
766 pack_open_windows--;
767 p->windows = w->next;
768 free(w);
772 static int close_pack_fd(struct packed_git *p)
774 if (p->pack_fd < 0)
775 return 0;
777 close(p->pack_fd);
778 pack_open_fds--;
779 p->pack_fd = -1;
781 return 1;
784 static void close_pack(struct packed_git *p)
786 close_pack_windows(p);
787 close_pack_fd(p);
788 close_pack_index(p);
791 void close_all_packs(void)
793 struct packed_git *p;
795 for (p = packed_git; p; p = p->next)
796 if (p->do_not_close)
797 die("BUG! Want to close pack marked 'do-not-close'");
798 else
799 close_pack(p);
804 * The LRU pack is the one with the oldest MRU window, preferring packs
805 * with no used windows, or the oldest mtime if it has no windows allocated.
807 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
809 struct pack_window *w, *this_mru_w;
810 int has_windows_inuse = 0;
813 * Reject this pack if it has windows and the previously selected
814 * one does not. If this pack does not have windows, reject
815 * it if the pack file is newer than the previously selected one.
817 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
818 return;
820 for (w = this_mru_w = p->windows; w; w = w->next) {
822 * Reject this pack if any of its windows are in use,
823 * but the previously selected pack did not have any
824 * inuse windows. Otherwise, record that this pack
825 * has windows in use.
827 if (w->inuse_cnt) {
828 if (*accept_windows_inuse)
829 has_windows_inuse = 1;
830 else
831 return;
834 if (w->last_used > this_mru_w->last_used)
835 this_mru_w = w;
838 * Reject this pack if it has windows that have been
839 * used more recently than the previously selected pack.
840 * If the previously selected pack had windows inuse and
841 * we have not encountered a window in this pack that is
842 * inuse, skip this check since we prefer a pack with no
843 * inuse windows to one that has inuse windows.
845 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
846 this_mru_w->last_used > (*mru_w)->last_used)
847 return;
851 * Select this pack.
853 *mru_w = this_mru_w;
854 *lru_p = p;
855 *accept_windows_inuse = has_windows_inuse;
858 static int close_one_pack(void)
860 struct packed_git *p, *lru_p = NULL;
861 struct pack_window *mru_w = NULL;
862 int accept_windows_inuse = 1;
864 for (p = packed_git; p; p = p->next) {
865 if (p->pack_fd == -1)
866 continue;
867 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
870 if (lru_p)
871 return close_pack_fd(lru_p);
873 return 0;
876 void unuse_pack(struct pack_window **w_cursor)
878 struct pack_window *w = *w_cursor;
879 if (w) {
880 w->inuse_cnt--;
881 *w_cursor = NULL;
885 void close_pack_index(struct packed_git *p)
887 if (p->index_data) {
888 munmap((void *)p->index_data, p->index_size);
889 p->index_data = NULL;
894 * This is used by git-repack in case a newly created pack happens to
895 * contain the same set of objects as an existing one. In that case
896 * the resulting file might be different even if its name would be the
897 * same. It is best to close any reference to the old pack before it is
898 * replaced on disk. Of course no index pointers or windows for given pack
899 * must subsist at this point. If ever objects from this pack are requested
900 * again, the new version of the pack will be reinitialized through
901 * reprepare_packed_git().
903 void free_pack_by_name(const char *pack_name)
905 struct packed_git *p, **pp = &packed_git;
907 while (*pp) {
908 p = *pp;
909 if (strcmp(pack_name, p->pack_name) == 0) {
910 clear_delta_base_cache();
911 close_pack(p);
912 free(p->bad_object_sha1);
913 *pp = p->next;
914 if (last_found_pack == p)
915 last_found_pack = NULL;
916 free(p);
917 return;
919 pp = &p->next;
923 static unsigned int get_max_fd_limit(void)
925 #ifdef RLIMIT_NOFILE
927 struct rlimit lim;
929 if (!getrlimit(RLIMIT_NOFILE, &lim))
930 return lim.rlim_cur;
932 #endif
934 #ifdef _SC_OPEN_MAX
936 long open_max = sysconf(_SC_OPEN_MAX);
937 if (0 < open_max)
938 return open_max;
940 * Otherwise, we got -1 for one of the two
941 * reasons:
943 * (1) sysconf() did not understand _SC_OPEN_MAX
944 * and signaled an error with -1; or
945 * (2) sysconf() said there is no limit.
947 * We _could_ clear errno before calling sysconf() to
948 * tell these two cases apart and return a huge number
949 * in the latter case to let the caller cap it to a
950 * value that is not so selfish, but letting the
951 * fallback OPEN_MAX codepath take care of these cases
952 * is a lot simpler.
955 #endif
957 #ifdef OPEN_MAX
958 return OPEN_MAX;
959 #else
960 return 1; /* see the caller ;-) */
961 #endif
965 * Do not call this directly as this leaks p->pack_fd on error return;
966 * call open_packed_git() instead.
968 static int open_packed_git_1(struct packed_git *p)
970 struct stat st;
971 struct pack_header hdr;
972 unsigned char sha1[20];
973 unsigned char *idx_sha1;
974 long fd_flag;
976 if (!p->index_data && open_pack_index(p))
977 return error("packfile %s index unavailable", p->pack_name);
979 if (!pack_max_fds) {
980 unsigned int max_fds = get_max_fd_limit();
982 /* Save 3 for stdin/stdout/stderr, 22 for work */
983 if (25 < max_fds)
984 pack_max_fds = max_fds - 25;
985 else
986 pack_max_fds = 1;
989 while (pack_max_fds <= pack_open_fds && close_one_pack())
990 ; /* nothing */
992 p->pack_fd = git_open_noatime(p->pack_name);
993 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
994 return -1;
995 pack_open_fds++;
997 /* If we created the struct before we had the pack we lack size. */
998 if (!p->pack_size) {
999 if (!S_ISREG(st.st_mode))
1000 return error("packfile %s not a regular file", p->pack_name);
1001 p->pack_size = st.st_size;
1002 } else if (p->pack_size != st.st_size)
1003 return error("packfile %s size changed", p->pack_name);
1005 /* We leave these file descriptors open with sliding mmap;
1006 * there is no point keeping them open across exec(), though.
1008 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1009 if (fd_flag < 0)
1010 return error("cannot determine file descriptor flags");
1011 fd_flag |= FD_CLOEXEC;
1012 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1013 return error("cannot set FD_CLOEXEC");
1015 /* Verify we recognize this pack file format. */
1016 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1017 return error("file %s is far too short to be a packfile", p->pack_name);
1018 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1019 return error("file %s is not a GIT packfile", p->pack_name);
1020 if (!pack_version_ok(hdr.hdr_version))
1021 return error("packfile %s is version %"PRIu32" and not"
1022 " supported (try upgrading GIT to a newer version)",
1023 p->pack_name, ntohl(hdr.hdr_version));
1025 /* Verify the pack matches its index. */
1026 if (p->num_objects != ntohl(hdr.hdr_entries))
1027 return error("packfile %s claims to have %"PRIu32" objects"
1028 " while index indicates %"PRIu32" objects",
1029 p->pack_name, ntohl(hdr.hdr_entries),
1030 p->num_objects);
1031 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1032 return error("end of packfile %s is unavailable", p->pack_name);
1033 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1034 return error("packfile %s signature is unavailable", p->pack_name);
1035 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1036 if (hashcmp(sha1, idx_sha1))
1037 return error("packfile %s does not match index", p->pack_name);
1038 return 0;
1041 static int open_packed_git(struct packed_git *p)
1043 if (!open_packed_git_1(p))
1044 return 0;
1045 close_pack_fd(p);
1046 return -1;
1049 static int in_window(struct pack_window *win, off_t offset)
1051 /* We must promise at least 20 bytes (one hash) after the
1052 * offset is available from this window, otherwise the offset
1053 * is not actually in this window and a different window (which
1054 * has that one hash excess) must be used. This is to support
1055 * the object header and delta base parsing routines below.
1057 off_t win_off = win->offset;
1058 return win_off <= offset
1059 && (offset + 20) <= (win_off + win->len);
1062 unsigned char *use_pack(struct packed_git *p,
1063 struct pack_window **w_cursor,
1064 off_t offset,
1065 unsigned long *left)
1067 struct pack_window *win = *w_cursor;
1069 /* Since packfiles end in a hash of their content and it's
1070 * pointless to ask for an offset into the middle of that
1071 * hash, and the in_window function above wouldn't match
1072 * don't allow an offset too close to the end of the file.
1074 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1075 die("packfile %s cannot be accessed", p->pack_name);
1076 if (offset > (p->pack_size - 20))
1077 die("offset beyond end of packfile (truncated pack?)");
1079 if (!win || !in_window(win, offset)) {
1080 if (win)
1081 win->inuse_cnt--;
1082 for (win = p->windows; win; win = win->next) {
1083 if (in_window(win, offset))
1084 break;
1086 if (!win) {
1087 size_t window_align = packed_git_window_size / 2;
1088 off_t len;
1090 if (p->pack_fd == -1 && open_packed_git(p))
1091 die("packfile %s cannot be accessed", p->pack_name);
1093 win = xcalloc(1, sizeof(*win));
1094 win->offset = (offset / window_align) * window_align;
1095 len = p->pack_size - win->offset;
1096 if (len > packed_git_window_size)
1097 len = packed_git_window_size;
1098 win->len = (size_t)len;
1099 pack_mapped += win->len;
1100 while (packed_git_limit < pack_mapped
1101 && unuse_one_window(p))
1102 ; /* nothing */
1103 win->base = xmmap(NULL, win->len,
1104 PROT_READ, MAP_PRIVATE,
1105 p->pack_fd, win->offset);
1106 if (win->base == MAP_FAILED)
1107 die("packfile %s cannot be mapped: %s",
1108 p->pack_name,
1109 strerror(errno));
1110 if (!win->offset && win->len == p->pack_size
1111 && !p->do_not_close)
1112 close_pack_fd(p);
1113 pack_mmap_calls++;
1114 pack_open_windows++;
1115 if (pack_mapped > peak_pack_mapped)
1116 peak_pack_mapped = pack_mapped;
1117 if (pack_open_windows > peak_pack_open_windows)
1118 peak_pack_open_windows = pack_open_windows;
1119 win->next = p->windows;
1120 p->windows = win;
1123 if (win != *w_cursor) {
1124 win->last_used = pack_used_ctr++;
1125 win->inuse_cnt++;
1126 *w_cursor = win;
1128 offset -= win->offset;
1129 if (left)
1130 *left = win->len - xsize_t(offset);
1131 return win->base + offset;
1134 static struct packed_git *alloc_packed_git(int extra)
1136 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1137 memset(p, 0, sizeof(*p));
1138 p->pack_fd = -1;
1139 return p;
1142 static void try_to_free_pack_memory(size_t size)
1144 release_pack_memory(size);
1147 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1149 static int have_set_try_to_free_routine;
1150 struct stat st;
1151 size_t alloc;
1152 struct packed_git *p;
1154 if (!have_set_try_to_free_routine) {
1155 have_set_try_to_free_routine = 1;
1156 set_try_to_free_routine(try_to_free_pack_memory);
1160 * Make sure a corresponding .pack file exists and that
1161 * the index looks sane.
1163 if (!strip_suffix_mem(path, &path_len, ".idx"))
1164 return NULL;
1167 * ".pack" is long enough to hold any suffix we're adding (and
1168 * the use xsnprintf double-checks that)
1170 alloc = path_len + strlen(".pack") + 1;
1171 p = alloc_packed_git(alloc);
1172 memcpy(p->pack_name, path, path_len);
1174 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1175 if (!access(p->pack_name, F_OK))
1176 p->pack_keep = 1;
1178 xsnprintf(p->pack_name + path_len, alloc - 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 int alloc = strlen(path) + 1;
1199 struct packed_git *p = alloc_packed_git(alloc);
1201 memcpy(p->pack_name, path, alloc); /* includes NUL */
1202 hashcpy(p->sha1, sha1);
1203 if (check_packed_git_idx(idx_path, p)) {
1204 free(p);
1205 return NULL;
1208 return p;
1211 void install_packed_git(struct packed_git *pack)
1213 if (pack->pack_fd != -1)
1214 pack_open_fds++;
1216 pack->next = packed_git;
1217 packed_git = pack;
1220 void (*report_garbage)(unsigned seen_bits, const char *path);
1222 static void report_helper(const struct string_list *list,
1223 int seen_bits, int first, int last)
1225 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1226 return;
1228 for (; first < last; first++)
1229 report_garbage(seen_bits, list->items[first].string);
1232 static void report_pack_garbage(struct string_list *list)
1234 int i, baselen = -1, first = 0, seen_bits = 0;
1236 if (!report_garbage)
1237 return;
1239 string_list_sort(list);
1241 for (i = 0; i < list->nr; i++) {
1242 const char *path = list->items[i].string;
1243 if (baselen != -1 &&
1244 strncmp(path, list->items[first].string, baselen)) {
1245 report_helper(list, seen_bits, first, i);
1246 baselen = -1;
1247 seen_bits = 0;
1249 if (baselen == -1) {
1250 const char *dot = strrchr(path, '.');
1251 if (!dot) {
1252 report_garbage(PACKDIR_FILE_GARBAGE, path);
1253 continue;
1255 baselen = dot - path + 1;
1256 first = i;
1258 if (!strcmp(path + baselen, "pack"))
1259 seen_bits |= 1;
1260 else if (!strcmp(path + baselen, "idx"))
1261 seen_bits |= 2;
1263 report_helper(list, seen_bits, first, list->nr);
1266 static void prepare_packed_git_one(char *objdir, int local)
1268 struct strbuf path = STRBUF_INIT;
1269 size_t dirnamelen;
1270 DIR *dir;
1271 struct dirent *de;
1272 struct string_list garbage = STRING_LIST_INIT_DUP;
1274 strbuf_addstr(&path, objdir);
1275 strbuf_addstr(&path, "/pack");
1276 dir = opendir(path.buf);
1277 if (!dir) {
1278 if (errno != ENOENT)
1279 error("unable to open object pack directory: %s: %s",
1280 path.buf, strerror(errno));
1281 strbuf_release(&path);
1282 return;
1284 strbuf_addch(&path, '/');
1285 dirnamelen = path.len;
1286 while ((de = readdir(dir)) != NULL) {
1287 struct packed_git *p;
1288 size_t base_len;
1290 if (is_dot_or_dotdot(de->d_name))
1291 continue;
1293 strbuf_setlen(&path, dirnamelen);
1294 strbuf_addstr(&path, de->d_name);
1296 base_len = path.len;
1297 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1298 /* Don't reopen a pack we already have. */
1299 for (p = packed_git; p; p = p->next) {
1300 size_t len;
1301 if (strip_suffix(p->pack_name, ".pack", &len) &&
1302 len == base_len &&
1303 !memcmp(p->pack_name, path.buf, len))
1304 break;
1306 if (p == NULL &&
1308 * See if it really is a valid .idx file with
1309 * corresponding .pack file that we can map.
1311 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1312 install_packed_git(p);
1315 if (!report_garbage)
1316 continue;
1318 if (ends_with(de->d_name, ".idx") ||
1319 ends_with(de->d_name, ".pack") ||
1320 ends_with(de->d_name, ".bitmap") ||
1321 ends_with(de->d_name, ".keep"))
1322 string_list_append(&garbage, path.buf);
1323 else
1324 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1326 closedir(dir);
1327 report_pack_garbage(&garbage);
1328 string_list_clear(&garbage, 0);
1329 strbuf_release(&path);
1332 static int sort_pack(const void *a_, const void *b_)
1334 struct packed_git *a = *((struct packed_git **)a_);
1335 struct packed_git *b = *((struct packed_git **)b_);
1336 int st;
1339 * Local packs tend to contain objects specific to our
1340 * variant of the project than remote ones. In addition,
1341 * remote ones could be on a network mounted filesystem.
1342 * Favor local ones for these reasons.
1344 st = a->pack_local - b->pack_local;
1345 if (st)
1346 return -st;
1349 * Younger packs tend to contain more recent objects,
1350 * and more recent objects tend to get accessed more
1351 * often.
1353 if (a->mtime < b->mtime)
1354 return 1;
1355 else if (a->mtime == b->mtime)
1356 return 0;
1357 return -1;
1360 static void rearrange_packed_git(void)
1362 struct packed_git **ary, *p;
1363 int i, n;
1365 for (n = 0, p = packed_git; p; p = p->next)
1366 n++;
1367 if (n < 2)
1368 return;
1370 /* prepare an array of packed_git for easier sorting */
1371 ary = xcalloc(n, sizeof(struct packed_git *));
1372 for (n = 0, p = packed_git; p; p = p->next)
1373 ary[n++] = p;
1375 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1377 /* link them back again */
1378 for (i = 0; i < n - 1; i++)
1379 ary[i]->next = ary[i + 1];
1380 ary[n - 1]->next = NULL;
1381 packed_git = ary[0];
1383 free(ary);
1386 static int prepare_packed_git_run_once = 0;
1387 void prepare_packed_git(void)
1389 struct alternate_object_database *alt;
1391 if (prepare_packed_git_run_once)
1392 return;
1393 prepare_packed_git_one(get_object_directory(), 1);
1394 prepare_alt_odb();
1395 for (alt = alt_odb_list; alt; alt = alt->next) {
1396 alt->name[-1] = 0;
1397 prepare_packed_git_one(alt->base, 0);
1398 alt->name[-1] = '/';
1400 rearrange_packed_git();
1401 prepare_packed_git_run_once = 1;
1404 void reprepare_packed_git(void)
1406 prepare_packed_git_run_once = 0;
1407 prepare_packed_git();
1410 static void mark_bad_packed_object(struct packed_git *p,
1411 const unsigned char *sha1)
1413 unsigned i;
1414 for (i = 0; i < p->num_bad_objects; i++)
1415 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1416 return;
1417 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1418 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1419 p->num_bad_objects++;
1422 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1424 struct packed_git *p;
1425 unsigned i;
1427 for (p = packed_git; p; p = p->next)
1428 for (i = 0; i < p->num_bad_objects; i++)
1429 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1430 return p;
1431 return NULL;
1435 * With an in-core object data in "map", rehash it to make sure the
1436 * object name actually matches "sha1" to detect object corruption.
1437 * With "map" == NULL, try reading the object named with "sha1" using
1438 * the streaming interface and rehash it to do the same.
1440 int check_sha1_signature(const unsigned char *sha1, void *map,
1441 unsigned long size, const char *type)
1443 unsigned char real_sha1[20];
1444 enum object_type obj_type;
1445 struct git_istream *st;
1446 git_SHA_CTX c;
1447 char hdr[32];
1448 int hdrlen;
1450 if (map) {
1451 hash_sha1_file(map, size, type, real_sha1);
1452 return hashcmp(sha1, real_sha1) ? -1 : 0;
1455 st = open_istream(sha1, &obj_type, &size, NULL);
1456 if (!st)
1457 return -1;
1459 /* Generate the header */
1460 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1462 /* Sha1.. */
1463 git_SHA1_Init(&c);
1464 git_SHA1_Update(&c, hdr, hdrlen);
1465 for (;;) {
1466 char buf[1024 * 16];
1467 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1469 if (readlen < 0) {
1470 close_istream(st);
1471 return -1;
1473 if (!readlen)
1474 break;
1475 git_SHA1_Update(&c, buf, readlen);
1477 git_SHA1_Final(real_sha1, &c);
1478 close_istream(st);
1479 return hashcmp(sha1, real_sha1) ? -1 : 0;
1482 int git_open_noatime(const char *name)
1484 static int sha1_file_open_flag = O_NOATIME;
1486 for (;;) {
1487 int fd;
1489 errno = 0;
1490 fd = open(name, O_RDONLY | sha1_file_open_flag);
1491 if (fd >= 0)
1492 return fd;
1494 /* Might the failure be due to O_NOATIME? */
1495 if (errno != ENOENT && sha1_file_open_flag) {
1496 sha1_file_open_flag = 0;
1497 continue;
1500 return -1;
1504 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1506 struct alternate_object_database *alt;
1508 if (!lstat(sha1_file_name(sha1), st))
1509 return 0;
1511 prepare_alt_odb();
1512 errno = ENOENT;
1513 for (alt = alt_odb_list; alt; alt = alt->next) {
1514 fill_sha1_path(alt->name, sha1);
1515 if (!lstat(alt->base, st))
1516 return 0;
1519 return -1;
1522 static int open_sha1_file(const unsigned char *sha1)
1524 int fd;
1525 struct alternate_object_database *alt;
1526 int most_interesting_errno;
1528 fd = git_open_noatime(sha1_file_name(sha1));
1529 if (fd >= 0)
1530 return fd;
1531 most_interesting_errno = errno;
1533 prepare_alt_odb();
1534 for (alt = alt_odb_list; alt; alt = alt->next) {
1535 fill_sha1_path(alt->name, sha1);
1536 fd = git_open_noatime(alt->base);
1537 if (fd >= 0)
1538 return fd;
1539 if (most_interesting_errno == ENOENT)
1540 most_interesting_errno = errno;
1542 errno = most_interesting_errno;
1543 return -1;
1546 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1548 void *map;
1549 int fd;
1551 fd = open_sha1_file(sha1);
1552 map = NULL;
1553 if (fd >= 0) {
1554 struct stat st;
1556 if (!fstat(fd, &st)) {
1557 *size = xsize_t(st.st_size);
1558 if (!*size) {
1559 /* mmap() is forbidden on empty files */
1560 error("object file %s is empty", sha1_file_name(sha1));
1561 return NULL;
1563 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1565 close(fd);
1567 return map;
1570 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1571 unsigned long len, enum object_type *type, unsigned long *sizep)
1573 unsigned shift;
1574 unsigned long size, c;
1575 unsigned long used = 0;
1577 c = buf[used++];
1578 *type = (c >> 4) & 7;
1579 size = c & 15;
1580 shift = 4;
1581 while (c & 0x80) {
1582 if (len <= used || bitsizeof(long) <= shift) {
1583 error("bad object header");
1584 size = used = 0;
1585 break;
1587 c = buf[used++];
1588 size += (c & 0x7f) << shift;
1589 shift += 7;
1591 *sizep = size;
1592 return used;
1595 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1597 /* Get the data stream */
1598 memset(stream, 0, sizeof(*stream));
1599 stream->next_in = map;
1600 stream->avail_in = mapsize;
1601 stream->next_out = buffer;
1602 stream->avail_out = bufsiz;
1604 git_inflate_init(stream);
1605 return git_inflate(stream, 0);
1608 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1609 unsigned long mapsize, void *buffer,
1610 unsigned long bufsiz, struct strbuf *header)
1612 int status;
1614 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1617 * Check if entire header is unpacked in the first iteration.
1619 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1620 return 0;
1623 * buffer[0..bufsiz] was not large enough. Copy the partial
1624 * result out to header, and then append the result of further
1625 * reading the stream.
1627 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1628 stream->next_out = buffer;
1629 stream->avail_out = bufsiz;
1631 do {
1632 status = git_inflate(stream, 0);
1633 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1634 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1635 return 0;
1636 stream->next_out = buffer;
1637 stream->avail_out = bufsiz;
1638 } while (status != Z_STREAM_END);
1639 return -1;
1642 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1644 int bytes = strlen(buffer) + 1;
1645 unsigned char *buf = xmallocz(size);
1646 unsigned long n;
1647 int status = Z_OK;
1649 n = stream->total_out - bytes;
1650 if (n > size)
1651 n = size;
1652 memcpy(buf, (char *) buffer + bytes, n);
1653 bytes = n;
1654 if (bytes <= size) {
1656 * The above condition must be (bytes <= size), not
1657 * (bytes < size). In other words, even though we
1658 * expect no more output and set avail_out to zero,
1659 * the input zlib stream may have bytes that express
1660 * "this concludes the stream", and we *do* want to
1661 * eat that input.
1663 * Otherwise we would not be able to test that we
1664 * consumed all the input to reach the expected size;
1665 * we also want to check that zlib tells us that all
1666 * went well with status == Z_STREAM_END at the end.
1668 stream->next_out = buf + bytes;
1669 stream->avail_out = size - bytes;
1670 while (status == Z_OK)
1671 status = git_inflate(stream, Z_FINISH);
1673 if (status == Z_STREAM_END && !stream->avail_in) {
1674 git_inflate_end(stream);
1675 return buf;
1678 if (status < 0)
1679 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1680 else if (stream->avail_in)
1681 error("garbage at end of loose object '%s'",
1682 sha1_to_hex(sha1));
1683 free(buf);
1684 return NULL;
1688 * We used to just use "sscanf()", but that's actually way
1689 * too permissive for what we want to check. So do an anal
1690 * object header parse by hand.
1692 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1693 unsigned int flags)
1695 const char *type_buf = hdr;
1696 unsigned long size;
1697 int type, type_len = 0;
1700 * The type can be of any size but is followed by
1701 * a space.
1703 for (;;) {
1704 char c = *hdr++;
1705 if (c == ' ')
1706 break;
1707 type_len++;
1710 type = type_from_string_gently(type_buf, type_len, 1);
1711 if (oi->typename)
1712 strbuf_add(oi->typename, type_buf, type_len);
1714 * Set type to 0 if its an unknown object and
1715 * we're obtaining the type using '--allow-unkown-type'
1716 * option.
1718 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1719 type = 0;
1720 else if (type < 0)
1721 die("invalid object type");
1722 if (oi->typep)
1723 *oi->typep = type;
1726 * The length must follow immediately, and be in canonical
1727 * decimal format (ie "010" is not valid).
1729 size = *hdr++ - '0';
1730 if (size > 9)
1731 return -1;
1732 if (size) {
1733 for (;;) {
1734 unsigned long c = *hdr - '0';
1735 if (c > 9)
1736 break;
1737 hdr++;
1738 size = size * 10 + c;
1742 if (oi->sizep)
1743 *oi->sizep = size;
1746 * The length must be followed by a zero byte
1748 return *hdr ? -1 : type;
1751 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1753 struct object_info oi;
1755 oi.sizep = sizep;
1756 oi.typename = NULL;
1757 oi.typep = NULL;
1758 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1761 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1763 int ret;
1764 git_zstream stream;
1765 char hdr[8192];
1767 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1768 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1769 return NULL;
1771 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1774 unsigned long get_size_from_delta(struct packed_git *p,
1775 struct pack_window **w_curs,
1776 off_t curpos)
1778 const unsigned char *data;
1779 unsigned char delta_head[20], *in;
1780 git_zstream stream;
1781 int st;
1783 memset(&stream, 0, sizeof(stream));
1784 stream.next_out = delta_head;
1785 stream.avail_out = sizeof(delta_head);
1787 git_inflate_init(&stream);
1788 do {
1789 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1790 stream.next_in = in;
1791 st = git_inflate(&stream, Z_FINISH);
1792 curpos += stream.next_in - in;
1793 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1794 stream.total_out < sizeof(delta_head));
1795 git_inflate_end(&stream);
1796 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1797 error("delta data unpack-initial failed");
1798 return 0;
1801 /* Examine the initial part of the delta to figure out
1802 * the result size.
1804 data = delta_head;
1806 /* ignore base size */
1807 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1809 /* Read the result size */
1810 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1813 static off_t get_delta_base(struct packed_git *p,
1814 struct pack_window **w_curs,
1815 off_t *curpos,
1816 enum object_type type,
1817 off_t delta_obj_offset)
1819 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1820 off_t base_offset;
1822 /* use_pack() assured us we have [base_info, base_info + 20)
1823 * as a range that we can look at without walking off the
1824 * end of the mapped window. Its actually the hash size
1825 * that is assured. An OFS_DELTA longer than the hash size
1826 * is stupid, as then a REF_DELTA would be smaller to store.
1828 if (type == OBJ_OFS_DELTA) {
1829 unsigned used = 0;
1830 unsigned char c = base_info[used++];
1831 base_offset = c & 127;
1832 while (c & 128) {
1833 base_offset += 1;
1834 if (!base_offset || MSB(base_offset, 7))
1835 return 0; /* overflow */
1836 c = base_info[used++];
1837 base_offset = (base_offset << 7) + (c & 127);
1839 base_offset = delta_obj_offset - base_offset;
1840 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1841 return 0; /* out of bound */
1842 *curpos += used;
1843 } else if (type == OBJ_REF_DELTA) {
1844 /* The base entry _must_ be in the same pack */
1845 base_offset = find_pack_entry_one(base_info, p);
1846 *curpos += 20;
1847 } else
1848 die("I am totally screwed");
1849 return base_offset;
1853 * Like get_delta_base above, but we return the sha1 instead of the pack
1854 * offset. This means it is cheaper for REF deltas (we do not have to do
1855 * the final object lookup), but more expensive for OFS deltas (we
1856 * have to load the revidx to convert the offset back into a sha1).
1858 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1859 struct pack_window **w_curs,
1860 off_t curpos,
1861 enum object_type type,
1862 off_t delta_obj_offset)
1864 if (type == OBJ_REF_DELTA) {
1865 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1866 return base;
1867 } else if (type == OBJ_OFS_DELTA) {
1868 struct revindex_entry *revidx;
1869 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1870 type, delta_obj_offset);
1872 if (!base_offset)
1873 return NULL;
1875 revidx = find_pack_revindex(p, base_offset);
1876 if (!revidx)
1877 return NULL;
1879 return nth_packed_object_sha1(p, revidx->nr);
1880 } else
1881 return NULL;
1884 int unpack_object_header(struct packed_git *p,
1885 struct pack_window **w_curs,
1886 off_t *curpos,
1887 unsigned long *sizep)
1889 unsigned char *base;
1890 unsigned long left;
1891 unsigned long used;
1892 enum object_type type;
1894 /* use_pack() assures us we have [base, base + 20) available
1895 * as a range that we can look at. (Its actually the hash
1896 * size that is assured.) With our object header encoding
1897 * the maximum deflated object size is 2^137, which is just
1898 * insane, so we know won't exceed what we have been given.
1900 base = use_pack(p, w_curs, *curpos, &left);
1901 used = unpack_object_header_buffer(base, left, &type, sizep);
1902 if (!used) {
1903 type = OBJ_BAD;
1904 } else
1905 *curpos += used;
1907 return type;
1910 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1912 int type;
1913 struct revindex_entry *revidx;
1914 const unsigned char *sha1;
1915 revidx = find_pack_revindex(p, obj_offset);
1916 if (!revidx)
1917 return OBJ_BAD;
1918 sha1 = nth_packed_object_sha1(p, revidx->nr);
1919 mark_bad_packed_object(p, sha1);
1920 type = sha1_object_info(sha1, NULL);
1921 if (type <= OBJ_NONE)
1922 return OBJ_BAD;
1923 return type;
1926 #define POI_STACK_PREALLOC 64
1928 static enum object_type packed_to_object_type(struct packed_git *p,
1929 off_t obj_offset,
1930 enum object_type type,
1931 struct pack_window **w_curs,
1932 off_t curpos)
1934 off_t small_poi_stack[POI_STACK_PREALLOC];
1935 off_t *poi_stack = small_poi_stack;
1936 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1938 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1939 off_t base_offset;
1940 unsigned long size;
1941 /* Push the object we're going to leave behind */
1942 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1943 poi_stack_alloc = alloc_nr(poi_stack_nr);
1944 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1945 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1946 } else {
1947 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1949 poi_stack[poi_stack_nr++] = obj_offset;
1950 /* If parsing the base offset fails, just unwind */
1951 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1952 if (!base_offset)
1953 goto unwind;
1954 curpos = obj_offset = base_offset;
1955 type = unpack_object_header(p, w_curs, &curpos, &size);
1956 if (type <= OBJ_NONE) {
1957 /* If getting the base itself fails, we first
1958 * retry the base, otherwise unwind */
1959 type = retry_bad_packed_offset(p, base_offset);
1960 if (type > OBJ_NONE)
1961 goto out;
1962 goto unwind;
1966 switch (type) {
1967 case OBJ_BAD:
1968 case OBJ_COMMIT:
1969 case OBJ_TREE:
1970 case OBJ_BLOB:
1971 case OBJ_TAG:
1972 break;
1973 default:
1974 error("unknown object type %i at offset %"PRIuMAX" in %s",
1975 type, (uintmax_t)obj_offset, p->pack_name);
1976 type = OBJ_BAD;
1979 out:
1980 if (poi_stack != small_poi_stack)
1981 free(poi_stack);
1982 return type;
1984 unwind:
1985 while (poi_stack_nr) {
1986 obj_offset = poi_stack[--poi_stack_nr];
1987 type = retry_bad_packed_offset(p, obj_offset);
1988 if (type > OBJ_NONE)
1989 goto out;
1991 type = OBJ_BAD;
1992 goto out;
1995 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1996 struct object_info *oi)
1998 struct pack_window *w_curs = NULL;
1999 unsigned long size;
2000 off_t curpos = obj_offset;
2001 enum object_type type;
2004 * We always get the representation type, but only convert it to
2005 * a "real" type later if the caller is interested.
2007 type = unpack_object_header(p, &w_curs, &curpos, &size);
2009 if (oi->sizep) {
2010 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2011 off_t tmp_pos = curpos;
2012 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2013 type, obj_offset);
2014 if (!base_offset) {
2015 type = OBJ_BAD;
2016 goto out;
2018 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2019 if (*oi->sizep == 0) {
2020 type = OBJ_BAD;
2021 goto out;
2023 } else {
2024 *oi->sizep = size;
2028 if (oi->disk_sizep) {
2029 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2030 *oi->disk_sizep = revidx[1].offset - obj_offset;
2033 if (oi->typep) {
2034 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2035 if (*oi->typep < 0) {
2036 type = OBJ_BAD;
2037 goto out;
2041 if (oi->delta_base_sha1) {
2042 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2043 const unsigned char *base;
2045 base = get_delta_base_sha1(p, &w_curs, curpos,
2046 type, obj_offset);
2047 if (!base) {
2048 type = OBJ_BAD;
2049 goto out;
2052 hashcpy(oi->delta_base_sha1, base);
2053 } else
2054 hashclr(oi->delta_base_sha1);
2057 out:
2058 unuse_pack(&w_curs);
2059 return type;
2062 static void *unpack_compressed_entry(struct packed_git *p,
2063 struct pack_window **w_curs,
2064 off_t curpos,
2065 unsigned long size)
2067 int st;
2068 git_zstream stream;
2069 unsigned char *buffer, *in;
2071 buffer = xmallocz_gently(size);
2072 if (!buffer)
2073 return NULL;
2074 memset(&stream, 0, sizeof(stream));
2075 stream.next_out = buffer;
2076 stream.avail_out = size + 1;
2078 git_inflate_init(&stream);
2079 do {
2080 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2081 stream.next_in = in;
2082 st = git_inflate(&stream, Z_FINISH);
2083 if (!stream.avail_out)
2084 break; /* the payload is larger than it should be */
2085 curpos += stream.next_in - in;
2086 } while (st == Z_OK || st == Z_BUF_ERROR);
2087 git_inflate_end(&stream);
2088 if ((st != Z_STREAM_END) || stream.total_out != size) {
2089 free(buffer);
2090 return NULL;
2093 return buffer;
2096 #define MAX_DELTA_CACHE (256)
2098 static size_t delta_base_cached;
2100 static struct delta_base_cache_lru_list {
2101 struct delta_base_cache_lru_list *prev;
2102 struct delta_base_cache_lru_list *next;
2103 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2105 static struct delta_base_cache_entry {
2106 struct delta_base_cache_lru_list lru;
2107 void *data;
2108 struct packed_git *p;
2109 off_t base_offset;
2110 unsigned long size;
2111 enum object_type type;
2112 } delta_base_cache[MAX_DELTA_CACHE];
2114 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2116 unsigned long hash;
2118 hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
2119 hash += (hash >> 8) + (hash >> 16);
2120 return hash % MAX_DELTA_CACHE;
2123 static struct delta_base_cache_entry *
2124 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2126 unsigned long hash = pack_entry_hash(p, base_offset);
2127 return delta_base_cache + hash;
2130 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2131 struct packed_git *p, off_t base_offset)
2133 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2136 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2138 struct delta_base_cache_entry *ent;
2139 ent = get_delta_base_cache_entry(p, base_offset);
2140 return eq_delta_base_cache_entry(ent, p, base_offset);
2143 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2145 ent->data = NULL;
2146 ent->lru.next->prev = ent->lru.prev;
2147 ent->lru.prev->next = ent->lru.next;
2148 delta_base_cached -= ent->size;
2151 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2152 unsigned long *base_size, enum object_type *type, int keep_cache)
2154 struct delta_base_cache_entry *ent;
2155 void *ret;
2157 ent = get_delta_base_cache_entry(p, base_offset);
2159 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2160 return unpack_entry(p, base_offset, type, base_size);
2162 ret = ent->data;
2164 if (!keep_cache)
2165 clear_delta_base_cache_entry(ent);
2166 else
2167 ret = xmemdupz(ent->data, ent->size);
2168 *type = ent->type;
2169 *base_size = ent->size;
2170 return ret;
2173 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2175 if (ent->data) {
2176 free(ent->data);
2177 ent->data = NULL;
2178 ent->lru.next->prev = ent->lru.prev;
2179 ent->lru.prev->next = ent->lru.next;
2180 delta_base_cached -= ent->size;
2184 void clear_delta_base_cache(void)
2186 unsigned long p;
2187 for (p = 0; p < MAX_DELTA_CACHE; p++)
2188 release_delta_base_cache(&delta_base_cache[p]);
2191 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2192 void *base, unsigned long base_size, enum object_type type)
2194 unsigned long hash = pack_entry_hash(p, base_offset);
2195 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2196 struct delta_base_cache_lru_list *lru;
2198 release_delta_base_cache(ent);
2199 delta_base_cached += base_size;
2201 for (lru = delta_base_cache_lru.next;
2202 delta_base_cached > delta_base_cache_limit
2203 && lru != &delta_base_cache_lru;
2204 lru = lru->next) {
2205 struct delta_base_cache_entry *f = (void *)lru;
2206 if (f->type == OBJ_BLOB)
2207 release_delta_base_cache(f);
2209 for (lru = delta_base_cache_lru.next;
2210 delta_base_cached > delta_base_cache_limit
2211 && lru != &delta_base_cache_lru;
2212 lru = lru->next) {
2213 struct delta_base_cache_entry *f = (void *)lru;
2214 release_delta_base_cache(f);
2217 ent->p = p;
2218 ent->base_offset = base_offset;
2219 ent->type = type;
2220 ent->data = base;
2221 ent->size = base_size;
2222 ent->lru.next = &delta_base_cache_lru;
2223 ent->lru.prev = delta_base_cache_lru.prev;
2224 delta_base_cache_lru.prev->next = &ent->lru;
2225 delta_base_cache_lru.prev = &ent->lru;
2228 static void *read_object(const unsigned char *sha1, enum object_type *type,
2229 unsigned long *size);
2231 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2233 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2234 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2235 p->pack_name, (uintmax_t)obj_offset);
2238 int do_check_packed_object_crc;
2240 #define UNPACK_ENTRY_STACK_PREALLOC 64
2241 struct unpack_entry_stack_ent {
2242 off_t obj_offset;
2243 off_t curpos;
2244 unsigned long size;
2247 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2248 enum object_type *final_type, unsigned long *final_size)
2250 struct pack_window *w_curs = NULL;
2251 off_t curpos = obj_offset;
2252 void *data = NULL;
2253 unsigned long size;
2254 enum object_type type;
2255 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2256 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2257 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2258 int base_from_cache = 0;
2260 write_pack_access_log(p, obj_offset);
2262 /* PHASE 1: drill down to the innermost base object */
2263 for (;;) {
2264 off_t base_offset;
2265 int i;
2266 struct delta_base_cache_entry *ent;
2268 ent = get_delta_base_cache_entry(p, curpos);
2269 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2270 type = ent->type;
2271 data = ent->data;
2272 size = ent->size;
2273 clear_delta_base_cache_entry(ent);
2274 base_from_cache = 1;
2275 break;
2278 if (do_check_packed_object_crc && p->index_version > 1) {
2279 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2280 unsigned long len = revidx[1].offset - obj_offset;
2281 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2282 const unsigned char *sha1 =
2283 nth_packed_object_sha1(p, revidx->nr);
2284 error("bad packed object CRC for %s",
2285 sha1_to_hex(sha1));
2286 mark_bad_packed_object(p, sha1);
2287 unuse_pack(&w_curs);
2288 return NULL;
2292 type = unpack_object_header(p, &w_curs, &curpos, &size);
2293 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2294 break;
2296 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2297 if (!base_offset) {
2298 error("failed to validate delta base reference "
2299 "at offset %"PRIuMAX" from %s",
2300 (uintmax_t)curpos, p->pack_name);
2301 /* bail to phase 2, in hopes of recovery */
2302 data = NULL;
2303 break;
2306 /* push object, proceed to base */
2307 if (delta_stack_nr >= delta_stack_alloc
2308 && delta_stack == small_delta_stack) {
2309 delta_stack_alloc = alloc_nr(delta_stack_nr);
2310 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2311 memcpy(delta_stack, small_delta_stack,
2312 sizeof(*delta_stack)*delta_stack_nr);
2313 } else {
2314 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2316 i = delta_stack_nr++;
2317 delta_stack[i].obj_offset = obj_offset;
2318 delta_stack[i].curpos = curpos;
2319 delta_stack[i].size = size;
2321 curpos = obj_offset = base_offset;
2324 /* PHASE 2: handle the base */
2325 switch (type) {
2326 case OBJ_OFS_DELTA:
2327 case OBJ_REF_DELTA:
2328 if (data)
2329 die("BUG in unpack_entry: left loop at a valid delta");
2330 break;
2331 case OBJ_COMMIT:
2332 case OBJ_TREE:
2333 case OBJ_BLOB:
2334 case OBJ_TAG:
2335 if (!base_from_cache)
2336 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2337 break;
2338 default:
2339 data = NULL;
2340 error("unknown object type %i at offset %"PRIuMAX" in %s",
2341 type, (uintmax_t)obj_offset, p->pack_name);
2344 /* PHASE 3: apply deltas in order */
2346 /* invariants:
2347 * 'data' holds the base data, or NULL if there was corruption
2349 while (delta_stack_nr) {
2350 void *delta_data;
2351 void *base = data;
2352 unsigned long delta_size, base_size = size;
2353 int i;
2355 data = NULL;
2357 if (base)
2358 add_delta_base_cache(p, obj_offset, base, base_size, type);
2360 if (!base) {
2362 * We're probably in deep shit, but let's try to fetch
2363 * the required base anyway from another pack or loose.
2364 * This is costly but should happen only in the presence
2365 * of a corrupted pack, and is better than failing outright.
2367 struct revindex_entry *revidx;
2368 const unsigned char *base_sha1;
2369 revidx = find_pack_revindex(p, obj_offset);
2370 if (revidx) {
2371 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2372 error("failed to read delta base object %s"
2373 " at offset %"PRIuMAX" from %s",
2374 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2375 p->pack_name);
2376 mark_bad_packed_object(p, base_sha1);
2377 base = read_object(base_sha1, &type, &base_size);
2381 i = --delta_stack_nr;
2382 obj_offset = delta_stack[i].obj_offset;
2383 curpos = delta_stack[i].curpos;
2384 delta_size = delta_stack[i].size;
2386 if (!base)
2387 continue;
2389 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2391 if (!delta_data) {
2392 error("failed to unpack compressed delta "
2393 "at offset %"PRIuMAX" from %s",
2394 (uintmax_t)curpos, p->pack_name);
2395 data = NULL;
2396 continue;
2399 data = patch_delta(base, base_size,
2400 delta_data, delta_size,
2401 &size);
2404 * We could not apply the delta; warn the user, but keep going.
2405 * Our failure will be noticed either in the next iteration of
2406 * the loop, or if this is the final delta, in the caller when
2407 * we return NULL. Those code paths will take care of making
2408 * a more explicit warning and retrying with another copy of
2409 * the object.
2411 if (!data)
2412 error("failed to apply delta");
2414 free(delta_data);
2417 *final_type = type;
2418 *final_size = size;
2420 unuse_pack(&w_curs);
2422 if (delta_stack != small_delta_stack)
2423 free(delta_stack);
2425 return data;
2428 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2429 uint32_t n)
2431 const unsigned char *index = p->index_data;
2432 if (!index) {
2433 if (open_pack_index(p))
2434 return NULL;
2435 index = p->index_data;
2437 if (n >= p->num_objects)
2438 return NULL;
2439 index += 4 * 256;
2440 if (p->index_version == 1) {
2441 return index + 24 * n + 4;
2442 } else {
2443 index += 8;
2444 return index + 20 * n;
2448 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2450 const unsigned char *index = p->index_data;
2451 index += 4 * 256;
2452 if (p->index_version == 1) {
2453 return ntohl(*((uint32_t *)(index + 24 * n)));
2454 } else {
2455 uint32_t off;
2456 index += 8 + p->num_objects * (20 + 4);
2457 off = ntohl(*((uint32_t *)(index + 4 * n)));
2458 if (!(off & 0x80000000))
2459 return off;
2460 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2461 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2462 ntohl(*((uint32_t *)(index + 4)));
2466 off_t find_pack_entry_one(const unsigned char *sha1,
2467 struct packed_git *p)
2469 const uint32_t *level1_ofs = p->index_data;
2470 const unsigned char *index = p->index_data;
2471 unsigned hi, lo, stride;
2472 static int use_lookup = -1;
2473 static int debug_lookup = -1;
2475 if (debug_lookup < 0)
2476 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2478 if (!index) {
2479 if (open_pack_index(p))
2480 return 0;
2481 level1_ofs = p->index_data;
2482 index = p->index_data;
2484 if (p->index_version > 1) {
2485 level1_ofs += 2;
2486 index += 8;
2488 index += 4 * 256;
2489 hi = ntohl(level1_ofs[*sha1]);
2490 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2491 if (p->index_version > 1) {
2492 stride = 20;
2493 } else {
2494 stride = 24;
2495 index += 4;
2498 if (debug_lookup)
2499 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2500 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2502 if (use_lookup < 0)
2503 use_lookup = !!getenv("GIT_USE_LOOKUP");
2504 if (use_lookup) {
2505 int pos = sha1_entry_pos(index, stride, 0,
2506 lo, hi, p->num_objects, sha1);
2507 if (pos < 0)
2508 return 0;
2509 return nth_packed_object_offset(p, pos);
2512 do {
2513 unsigned mi = (lo + hi) / 2;
2514 int cmp = hashcmp(index + mi * stride, sha1);
2516 if (debug_lookup)
2517 printf("lo %u hi %u rg %u mi %u\n",
2518 lo, hi, hi - lo, mi);
2519 if (!cmp)
2520 return nth_packed_object_offset(p, mi);
2521 if (cmp > 0)
2522 hi = mi;
2523 else
2524 lo = mi+1;
2525 } while (lo < hi);
2526 return 0;
2529 int is_pack_valid(struct packed_git *p)
2531 /* An already open pack is known to be valid. */
2532 if (p->pack_fd != -1)
2533 return 1;
2535 /* If the pack has one window completely covering the
2536 * file size, the pack is known to be valid even if
2537 * the descriptor is not currently open.
2539 if (p->windows) {
2540 struct pack_window *w = p->windows;
2542 if (!w->offset && w->len == p->pack_size)
2543 return 1;
2546 /* Force the pack to open to prove its valid. */
2547 return !open_packed_git(p);
2550 static int fill_pack_entry(const unsigned char *sha1,
2551 struct pack_entry *e,
2552 struct packed_git *p)
2554 off_t offset;
2556 if (p->num_bad_objects) {
2557 unsigned i;
2558 for (i = 0; i < p->num_bad_objects; i++)
2559 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2560 return 0;
2563 offset = find_pack_entry_one(sha1, p);
2564 if (!offset)
2565 return 0;
2568 * We are about to tell the caller where they can locate the
2569 * requested object. We better make sure the packfile is
2570 * still here and can be accessed before supplying that
2571 * answer, as it may have been deleted since the index was
2572 * loaded!
2574 if (!is_pack_valid(p))
2575 return 0;
2576 e->offset = offset;
2577 e->p = p;
2578 hashcpy(e->sha1, sha1);
2579 return 1;
2583 * Iff a pack file contains the object named by sha1, return true and
2584 * store its location to e.
2586 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2588 struct packed_git *p;
2590 prepare_packed_git();
2591 if (!packed_git)
2592 return 0;
2594 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2595 return 1;
2597 for (p = packed_git; p; p = p->next) {
2598 if (p == last_found_pack)
2599 continue; /* we already checked this one */
2601 if (fill_pack_entry(sha1, e, p)) {
2602 last_found_pack = p;
2603 return 1;
2606 return 0;
2609 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2610 struct packed_git *packs)
2612 struct packed_git *p;
2614 for (p = packs; p; p = p->next) {
2615 if (find_pack_entry_one(sha1, p))
2616 return p;
2618 return NULL;
2622 static int sha1_loose_object_info(const unsigned char *sha1,
2623 struct object_info *oi,
2624 int flags)
2626 int status = 0;
2627 unsigned long mapsize;
2628 void *map;
2629 git_zstream stream;
2630 char hdr[32];
2631 struct strbuf hdrbuf = STRBUF_INIT;
2633 if (oi->delta_base_sha1)
2634 hashclr(oi->delta_base_sha1);
2637 * If we don't care about type or size, then we don't
2638 * need to look inside the object at all. Note that we
2639 * do not optimize out the stat call, even if the
2640 * caller doesn't care about the disk-size, since our
2641 * return value implicitly indicates whether the
2642 * object even exists.
2644 if (!oi->typep && !oi->typename && !oi->sizep) {
2645 struct stat st;
2646 if (stat_sha1_file(sha1, &st) < 0)
2647 return -1;
2648 if (oi->disk_sizep)
2649 *oi->disk_sizep = st.st_size;
2650 return 0;
2653 map = map_sha1_file(sha1, &mapsize);
2654 if (!map)
2655 return -1;
2656 if (oi->disk_sizep)
2657 *oi->disk_sizep = mapsize;
2658 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2659 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2660 status = error("unable to unpack %s header with --allow-unknown-type",
2661 sha1_to_hex(sha1));
2662 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2663 status = error("unable to unpack %s header",
2664 sha1_to_hex(sha1));
2665 if (status < 0)
2666 ; /* Do nothing */
2667 else if (hdrbuf.len) {
2668 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2669 status = error("unable to parse %s header with --allow-unknown-type",
2670 sha1_to_hex(sha1));
2671 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2672 status = error("unable to parse %s header", sha1_to_hex(sha1));
2673 git_inflate_end(&stream);
2674 munmap(map, mapsize);
2675 if (status && oi->typep)
2676 *oi->typep = status;
2677 strbuf_release(&hdrbuf);
2678 return 0;
2681 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2683 struct cached_object *co;
2684 struct pack_entry e;
2685 int rtype;
2686 enum object_type real_type;
2687 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2689 co = find_cached_object(real);
2690 if (co) {
2691 if (oi->typep)
2692 *(oi->typep) = co->type;
2693 if (oi->sizep)
2694 *(oi->sizep) = co->size;
2695 if (oi->disk_sizep)
2696 *(oi->disk_sizep) = 0;
2697 if (oi->delta_base_sha1)
2698 hashclr(oi->delta_base_sha1);
2699 if (oi->typename)
2700 strbuf_addstr(oi->typename, typename(co->type));
2701 oi->whence = OI_CACHED;
2702 return 0;
2705 if (!find_pack_entry(real, &e)) {
2706 /* Most likely it's a loose object. */
2707 if (!sha1_loose_object_info(real, oi, flags)) {
2708 oi->whence = OI_LOOSE;
2709 return 0;
2712 /* Not a loose object; someone else may have just packed it. */
2713 reprepare_packed_git();
2714 if (!find_pack_entry(real, &e))
2715 return -1;
2719 * packed_object_info() does not follow the delta chain to
2720 * find out the real type, unless it is given oi->typep.
2722 if (oi->typename && !oi->typep)
2723 oi->typep = &real_type;
2725 rtype = packed_object_info(e.p, e.offset, oi);
2726 if (rtype < 0) {
2727 mark_bad_packed_object(e.p, real);
2728 if (oi->typep == &real_type)
2729 oi->typep = NULL;
2730 return sha1_object_info_extended(real, oi, 0);
2731 } else if (in_delta_base_cache(e.p, e.offset)) {
2732 oi->whence = OI_DBCACHED;
2733 } else {
2734 oi->whence = OI_PACKED;
2735 oi->u.packed.offset = e.offset;
2736 oi->u.packed.pack = e.p;
2737 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2738 rtype == OBJ_OFS_DELTA);
2740 if (oi->typename)
2741 strbuf_addstr(oi->typename, typename(*oi->typep));
2742 if (oi->typep == &real_type)
2743 oi->typep = NULL;
2745 return 0;
2748 /* returns enum object_type or negative */
2749 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2751 enum object_type type;
2752 struct object_info oi = {NULL};
2754 oi.typep = &type;
2755 oi.sizep = sizep;
2756 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2757 return -1;
2758 return type;
2761 static void *read_packed_sha1(const unsigned char *sha1,
2762 enum object_type *type, unsigned long *size)
2764 struct pack_entry e;
2765 void *data;
2767 if (!find_pack_entry(sha1, &e))
2768 return NULL;
2769 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2770 if (!data) {
2772 * We're probably in deep shit, but let's try to fetch
2773 * the required object anyway from another pack or loose.
2774 * This should happen only in the presence of a corrupted
2775 * pack, and is better than failing outright.
2777 error("failed to read object %s at offset %"PRIuMAX" from %s",
2778 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2779 mark_bad_packed_object(e.p, sha1);
2780 data = read_object(sha1, type, size);
2782 return data;
2785 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2786 unsigned char *sha1)
2788 struct cached_object *co;
2790 hash_sha1_file(buf, len, typename(type), sha1);
2791 if (has_sha1_file(sha1) || find_cached_object(sha1))
2792 return 0;
2793 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2794 co = &cached_objects[cached_object_nr++];
2795 co->size = len;
2796 co->type = type;
2797 co->buf = xmalloc(len);
2798 memcpy(co->buf, buf, len);
2799 hashcpy(co->sha1, sha1);
2800 return 0;
2803 static void *read_object(const unsigned char *sha1, enum object_type *type,
2804 unsigned long *size)
2806 unsigned long mapsize;
2807 void *map, *buf;
2808 struct cached_object *co;
2810 co = find_cached_object(sha1);
2811 if (co) {
2812 *type = co->type;
2813 *size = co->size;
2814 return xmemdupz(co->buf, co->size);
2817 buf = read_packed_sha1(sha1, type, size);
2818 if (buf)
2819 return buf;
2820 map = map_sha1_file(sha1, &mapsize);
2821 if (map) {
2822 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2823 munmap(map, mapsize);
2824 return buf;
2826 reprepare_packed_git();
2827 return read_packed_sha1(sha1, type, size);
2831 * This function dies on corrupt objects; the callers who want to
2832 * deal with them should arrange to call read_object() and give error
2833 * messages themselves.
2835 void *read_sha1_file_extended(const unsigned char *sha1,
2836 enum object_type *type,
2837 unsigned long *size,
2838 unsigned flag)
2840 void *data;
2841 const struct packed_git *p;
2842 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2844 errno = 0;
2845 data = read_object(repl, type, size);
2846 if (data)
2847 return data;
2849 if (errno && errno != ENOENT)
2850 die_errno("failed to read object %s", sha1_to_hex(sha1));
2852 /* die if we replaced an object with one that does not exist */
2853 if (repl != sha1)
2854 die("replacement %s not found for %s",
2855 sha1_to_hex(repl), sha1_to_hex(sha1));
2857 if (has_loose_object(repl)) {
2858 const char *path = sha1_file_name(sha1);
2860 die("loose object %s (stored in %s) is corrupt",
2861 sha1_to_hex(repl), path);
2864 if ((p = has_packed_and_bad(repl)) != NULL)
2865 die("packed object %s (stored in %s) is corrupt",
2866 sha1_to_hex(repl), p->pack_name);
2868 return NULL;
2871 void *read_object_with_reference(const unsigned char *sha1,
2872 const char *required_type_name,
2873 unsigned long *size,
2874 unsigned char *actual_sha1_return)
2876 enum object_type type, required_type;
2877 void *buffer;
2878 unsigned long isize;
2879 unsigned char actual_sha1[20];
2881 required_type = type_from_string(required_type_name);
2882 hashcpy(actual_sha1, sha1);
2883 while (1) {
2884 int ref_length = -1;
2885 const char *ref_type = NULL;
2887 buffer = read_sha1_file(actual_sha1, &type, &isize);
2888 if (!buffer)
2889 return NULL;
2890 if (type == required_type) {
2891 *size = isize;
2892 if (actual_sha1_return)
2893 hashcpy(actual_sha1_return, actual_sha1);
2894 return buffer;
2896 /* Handle references */
2897 else if (type == OBJ_COMMIT)
2898 ref_type = "tree ";
2899 else if (type == OBJ_TAG)
2900 ref_type = "object ";
2901 else {
2902 free(buffer);
2903 return NULL;
2905 ref_length = strlen(ref_type);
2907 if (ref_length + 40 > isize ||
2908 memcmp(buffer, ref_type, ref_length) ||
2909 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2910 free(buffer);
2911 return NULL;
2913 free(buffer);
2914 /* Now we have the ID of the referred-to object in
2915 * actual_sha1. Check again. */
2919 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2920 const char *type, unsigned char *sha1,
2921 char *hdr, int *hdrlen)
2923 git_SHA_CTX c;
2925 /* Generate the header */
2926 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2928 /* Sha1.. */
2929 git_SHA1_Init(&c);
2930 git_SHA1_Update(&c, hdr, *hdrlen);
2931 git_SHA1_Update(&c, buf, len);
2932 git_SHA1_Final(sha1, &c);
2936 * Move the just written object into its final resting place.
2938 int finalize_object_file(const char *tmpfile, const char *filename)
2940 int ret = 0;
2942 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2943 goto try_rename;
2944 else if (link(tmpfile, filename))
2945 ret = errno;
2948 * Coda hack - coda doesn't like cross-directory links,
2949 * so we fall back to a rename, which will mean that it
2950 * won't be able to check collisions, but that's not a
2951 * big deal.
2953 * The same holds for FAT formatted media.
2955 * When this succeeds, we just return. We have nothing
2956 * left to unlink.
2958 if (ret && ret != EEXIST) {
2959 try_rename:
2960 if (!rename(tmpfile, filename))
2961 goto out;
2962 ret = errno;
2964 unlink_or_warn(tmpfile);
2965 if (ret) {
2966 if (ret != EEXIST) {
2967 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2969 /* FIXME!!! Collision check here ? */
2972 out:
2973 if (adjust_shared_perm(filename))
2974 return error("unable to set permission to '%s'", filename);
2975 return 0;
2978 static int write_buffer(int fd, const void *buf, size_t len)
2980 if (write_in_full(fd, buf, len) < 0)
2981 return error("file write error (%s)", strerror(errno));
2982 return 0;
2985 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2986 unsigned char *sha1)
2988 char hdr[32];
2989 int hdrlen = sizeof(hdr);
2990 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2991 return 0;
2994 /* Finalize a file on disk, and close it. */
2995 static void close_sha1_file(int fd)
2997 if (fsync_object_files)
2998 fsync_or_die(fd, "sha1 file");
2999 if (close(fd) != 0)
3000 die_errno("error when closing sha1 file");
3003 /* Size of directory component, including the ending '/' */
3004 static inline int directory_size(const char *filename)
3006 const char *s = strrchr(filename, '/');
3007 if (!s)
3008 return 0;
3009 return s - filename + 1;
3013 * This creates a temporary file in the same directory as the final
3014 * 'filename'
3016 * We want to avoid cross-directory filename renames, because those
3017 * can have problems on various filesystems (FAT, NFS, Coda).
3019 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3021 int fd, dirlen = directory_size(filename);
3023 strbuf_reset(tmp);
3024 strbuf_add(tmp, filename, dirlen);
3025 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3026 fd = git_mkstemp_mode(tmp->buf, 0444);
3027 if (fd < 0 && dirlen && errno == ENOENT) {
3029 * Make sure the directory exists; note that the contents
3030 * of the buffer are undefined after mkstemp returns an
3031 * error, so we have to rewrite the whole buffer from
3032 * scratch.
3034 strbuf_reset(tmp);
3035 strbuf_add(tmp, filename, dirlen - 1);
3036 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3037 return -1;
3038 if (adjust_shared_perm(tmp->buf))
3039 return -1;
3041 /* Try again */
3042 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3043 fd = git_mkstemp_mode(tmp->buf, 0444);
3045 return fd;
3048 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3049 const void *buf, unsigned long len, time_t mtime)
3051 int fd, ret;
3052 unsigned char compressed[4096];
3053 git_zstream stream;
3054 git_SHA_CTX c;
3055 unsigned char parano_sha1[20];
3056 static struct strbuf tmp_file = STRBUF_INIT;
3057 const char *filename = sha1_file_name(sha1);
3059 fd = create_tmpfile(&tmp_file, filename);
3060 if (fd < 0) {
3061 if (errno == EACCES)
3062 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3063 else
3064 return error("unable to create temporary file: %s", strerror(errno));
3067 /* Set it up */
3068 git_deflate_init(&stream, zlib_compression_level);
3069 stream.next_out = compressed;
3070 stream.avail_out = sizeof(compressed);
3071 git_SHA1_Init(&c);
3073 /* First header.. */
3074 stream.next_in = (unsigned char *)hdr;
3075 stream.avail_in = hdrlen;
3076 while (git_deflate(&stream, 0) == Z_OK)
3077 ; /* nothing */
3078 git_SHA1_Update(&c, hdr, hdrlen);
3080 /* Then the data itself.. */
3081 stream.next_in = (void *)buf;
3082 stream.avail_in = len;
3083 do {
3084 unsigned char *in0 = stream.next_in;
3085 ret = git_deflate(&stream, Z_FINISH);
3086 git_SHA1_Update(&c, in0, stream.next_in - in0);
3087 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3088 die("unable to write sha1 file");
3089 stream.next_out = compressed;
3090 stream.avail_out = sizeof(compressed);
3091 } while (ret == Z_OK);
3093 if (ret != Z_STREAM_END)
3094 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3095 ret = git_deflate_end_gently(&stream);
3096 if (ret != Z_OK)
3097 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3098 git_SHA1_Final(parano_sha1, &c);
3099 if (hashcmp(sha1, parano_sha1) != 0)
3100 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3102 close_sha1_file(fd);
3104 if (mtime) {
3105 struct utimbuf utb;
3106 utb.actime = mtime;
3107 utb.modtime = mtime;
3108 if (utime(tmp_file.buf, &utb) < 0)
3109 warning("failed utime() on %s: %s",
3110 tmp_file.buf, strerror(errno));
3113 return finalize_object_file(tmp_file.buf, filename);
3116 static int freshen_loose_object(const unsigned char *sha1)
3118 return check_and_freshen(sha1, 1);
3121 static int freshen_packed_object(const unsigned char *sha1)
3123 struct pack_entry e;
3124 if (!find_pack_entry(sha1, &e))
3125 return 0;
3126 if (e.p->freshened)
3127 return 1;
3128 if (!freshen_file(e.p->pack_name))
3129 return 0;
3130 e.p->freshened = 1;
3131 return 1;
3134 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3136 char hdr[32];
3137 int hdrlen = sizeof(hdr);
3139 /* Normally if we have it in the pack then we do not bother writing
3140 * it out into .git/objects/??/?{38} file.
3142 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3143 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3144 return 0;
3145 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3148 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3149 unsigned char *sha1, unsigned flags)
3151 char *header;
3152 int hdrlen, status = 0;
3154 /* type string, SP, %lu of the length plus NUL must fit this */
3155 hdrlen = strlen(type) + 32;
3156 header = xmalloc(hdrlen);
3157 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3159 if (!(flags & HASH_WRITE_OBJECT))
3160 goto cleanup;
3161 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3162 goto cleanup;
3163 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3165 cleanup:
3166 free(header);
3167 return status;
3170 int force_object_loose(const unsigned char *sha1, time_t mtime)
3172 void *buf;
3173 unsigned long len;
3174 enum object_type type;
3175 char hdr[32];
3176 int hdrlen;
3177 int ret;
3179 if (has_loose_object(sha1))
3180 return 0;
3181 buf = read_packed_sha1(sha1, &type, &len);
3182 if (!buf)
3183 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3184 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3185 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3186 free(buf);
3188 return ret;
3191 int has_pack_index(const unsigned char *sha1)
3193 struct stat st;
3194 if (stat(sha1_pack_index_name(sha1), &st))
3195 return 0;
3196 return 1;
3199 int has_sha1_pack(const unsigned char *sha1)
3201 struct pack_entry e;
3202 return find_pack_entry(sha1, &e);
3205 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3207 struct pack_entry e;
3209 if (find_pack_entry(sha1, &e))
3210 return 1;
3211 if (has_loose_object(sha1))
3212 return 1;
3213 if (flags & HAS_SHA1_QUICK)
3214 return 0;
3215 reprepare_packed_git();
3216 return find_pack_entry(sha1, &e);
3219 static void check_tree(const void *buf, size_t size)
3221 struct tree_desc desc;
3222 struct name_entry entry;
3224 init_tree_desc(&desc, buf, size);
3225 while (tree_entry(&desc, &entry))
3226 /* do nothing
3227 * tree_entry() will die() on malformed entries */
3231 static void check_commit(const void *buf, size_t size)
3233 struct commit c;
3234 memset(&c, 0, sizeof(c));
3235 if (parse_commit_buffer(&c, buf, size))
3236 die("corrupt commit");
3239 static void check_tag(const void *buf, size_t size)
3241 struct tag t;
3242 memset(&t, 0, sizeof(t));
3243 if (parse_tag_buffer(&t, buf, size))
3244 die("corrupt tag");
3247 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3248 enum object_type type,
3249 const char *path, unsigned flags)
3251 int ret, re_allocated = 0;
3252 int write_object = flags & HASH_WRITE_OBJECT;
3254 if (!type)
3255 type = OBJ_BLOB;
3258 * Convert blobs to git internal format
3260 if ((type == OBJ_BLOB) && path) {
3261 struct strbuf nbuf = STRBUF_INIT;
3262 if (convert_to_git(path, buf, size, &nbuf,
3263 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3264 buf = strbuf_detach(&nbuf, &size);
3265 re_allocated = 1;
3268 if (flags & HASH_FORMAT_CHECK) {
3269 if (type == OBJ_TREE)
3270 check_tree(buf, size);
3271 if (type == OBJ_COMMIT)
3272 check_commit(buf, size);
3273 if (type == OBJ_TAG)
3274 check_tag(buf, size);
3277 if (write_object)
3278 ret = write_sha1_file(buf, size, typename(type), sha1);
3279 else
3280 ret = hash_sha1_file(buf, size, typename(type), sha1);
3281 if (re_allocated)
3282 free(buf);
3283 return ret;
3286 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3287 const char *path, unsigned flags)
3289 int ret;
3290 const int write_object = flags & HASH_WRITE_OBJECT;
3291 struct strbuf sbuf = STRBUF_INIT;
3293 assert(path);
3294 assert(would_convert_to_git_filter_fd(path));
3296 convert_to_git_filter_fd(path, fd, &sbuf,
3297 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3299 if (write_object)
3300 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3301 sha1);
3302 else
3303 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3304 sha1);
3305 strbuf_release(&sbuf);
3306 return ret;
3309 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3310 const char *path, unsigned flags)
3312 struct strbuf sbuf = STRBUF_INIT;
3313 int ret;
3315 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3316 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3317 else
3318 ret = -1;
3319 strbuf_release(&sbuf);
3320 return ret;
3323 #define SMALL_FILE_SIZE (32*1024)
3325 static int index_core(unsigned char *sha1, int fd, size_t size,
3326 enum object_type type, const char *path,
3327 unsigned flags)
3329 int ret;
3331 if (!size) {
3332 ret = index_mem(sha1, "", size, type, path, flags);
3333 } else if (size <= SMALL_FILE_SIZE) {
3334 char *buf = xmalloc(size);
3335 if (size == read_in_full(fd, buf, size))
3336 ret = index_mem(sha1, buf, size, type, path, flags);
3337 else
3338 ret = error("short read %s", strerror(errno));
3339 free(buf);
3340 } else {
3341 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3342 ret = index_mem(sha1, buf, size, type, path, flags);
3343 munmap(buf, size);
3345 return ret;
3349 * This creates one packfile per large blob unless bulk-checkin
3350 * machinery is "plugged".
3352 * This also bypasses the usual "convert-to-git" dance, and that is on
3353 * purpose. We could write a streaming version of the converting
3354 * functions and insert that before feeding the data to fast-import
3355 * (or equivalent in-core API described above). However, that is
3356 * somewhat complicated, as we do not know the size of the filter
3357 * result, which we need to know beforehand when writing a git object.
3358 * Since the primary motivation for trying to stream from the working
3359 * tree file and to avoid mmaping it in core is to deal with large
3360 * binary blobs, they generally do not want to get any conversion, and
3361 * callers should avoid this code path when filters are requested.
3363 static int index_stream(unsigned char *sha1, int fd, size_t size,
3364 enum object_type type, const char *path,
3365 unsigned flags)
3367 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3370 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3371 enum object_type type, const char *path, unsigned flags)
3373 int ret;
3376 * Call xsize_t() only when needed to avoid potentially unnecessary
3377 * die() for large files.
3379 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3380 ret = index_stream_convert_blob(sha1, fd, path, flags);
3381 else if (!S_ISREG(st->st_mode))
3382 ret = index_pipe(sha1, fd, type, path, flags);
3383 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3384 (path && would_convert_to_git(path)))
3385 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3386 flags);
3387 else
3388 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3389 flags);
3390 close(fd);
3391 return ret;
3394 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3396 int fd;
3397 struct strbuf sb = STRBUF_INIT;
3399 switch (st->st_mode & S_IFMT) {
3400 case S_IFREG:
3401 fd = open(path, O_RDONLY);
3402 if (fd < 0)
3403 return error("open(\"%s\"): %s", path,
3404 strerror(errno));
3405 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3406 return error("%s: failed to insert into database",
3407 path);
3408 break;
3409 case S_IFLNK:
3410 if (strbuf_readlink(&sb, path, st->st_size)) {
3411 char *errstr = strerror(errno);
3412 return error("readlink(\"%s\"): %s", path,
3413 errstr);
3415 if (!(flags & HASH_WRITE_OBJECT))
3416 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3417 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3418 return error("%s: failed to insert into database",
3419 path);
3420 strbuf_release(&sb);
3421 break;
3422 case S_IFDIR:
3423 return resolve_gitlink_ref(path, "HEAD", sha1);
3424 default:
3425 return error("%s: unsupported file type", path);
3427 return 0;
3430 int read_pack_header(int fd, struct pack_header *header)
3432 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3433 /* "eof before pack header was fully read" */
3434 return PH_ERROR_EOF;
3436 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3437 /* "protocol error (pack signature mismatch detected)" */
3438 return PH_ERROR_PACK_SIGNATURE;
3439 if (!pack_version_ok(header->hdr_version))
3440 /* "protocol error (pack version unsupported)" */
3441 return PH_ERROR_PROTOCOL;
3442 return 0;
3445 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3447 enum object_type type = sha1_object_info(sha1, NULL);
3448 if (type < 0)
3449 die("%s is not a valid object", sha1_to_hex(sha1));
3450 if (type != expect)
3451 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3452 typename(expect));
3455 static int for_each_file_in_obj_subdir(int subdir_nr,
3456 struct strbuf *path,
3457 each_loose_object_fn obj_cb,
3458 each_loose_cruft_fn cruft_cb,
3459 each_loose_subdir_fn subdir_cb,
3460 void *data)
3462 size_t baselen = path->len;
3463 DIR *dir = opendir(path->buf);
3464 struct dirent *de;
3465 int r = 0;
3467 if (!dir) {
3468 if (errno == ENOENT)
3469 return 0;
3470 return error("unable to open %s: %s", path->buf, strerror(errno));
3473 while ((de = readdir(dir))) {
3474 if (is_dot_or_dotdot(de->d_name))
3475 continue;
3477 strbuf_setlen(path, baselen);
3478 strbuf_addf(path, "/%s", de->d_name);
3480 if (strlen(de->d_name) == 38) {
3481 char hex[41];
3482 unsigned char sha1[20];
3484 snprintf(hex, sizeof(hex), "%02x%s",
3485 subdir_nr, de->d_name);
3486 if (!get_sha1_hex(hex, sha1)) {
3487 if (obj_cb) {
3488 r = obj_cb(sha1, path->buf, data);
3489 if (r)
3490 break;
3492 continue;
3496 if (cruft_cb) {
3497 r = cruft_cb(de->d_name, path->buf, data);
3498 if (r)
3499 break;
3502 strbuf_setlen(path, baselen);
3504 if (!r && subdir_cb)
3505 r = subdir_cb(subdir_nr, path->buf, data);
3507 closedir(dir);
3508 return r;
3511 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3512 each_loose_object_fn obj_cb,
3513 each_loose_cruft_fn cruft_cb,
3514 each_loose_subdir_fn subdir_cb,
3515 void *data)
3517 size_t baselen = path->len;
3518 int r = 0;
3519 int i;
3521 for (i = 0; i < 256; i++) {
3522 strbuf_addf(path, "/%02x", i);
3523 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3524 subdir_cb, data);
3525 strbuf_setlen(path, baselen);
3526 if (r)
3527 break;
3530 return r;
3533 int for_each_loose_file_in_objdir(const char *path,
3534 each_loose_object_fn obj_cb,
3535 each_loose_cruft_fn cruft_cb,
3536 each_loose_subdir_fn subdir_cb,
3537 void *data)
3539 struct strbuf buf = STRBUF_INIT;
3540 int r;
3542 strbuf_addstr(&buf, path);
3543 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3544 subdir_cb, data);
3545 strbuf_release(&buf);
3547 return r;
3550 struct loose_alt_odb_data {
3551 each_loose_object_fn *cb;
3552 void *data;
3555 static int loose_from_alt_odb(struct alternate_object_database *alt,
3556 void *vdata)
3558 struct loose_alt_odb_data *data = vdata;
3559 struct strbuf buf = STRBUF_INIT;
3560 int r;
3562 /* copy base not including trailing '/' */
3563 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3564 r = for_each_loose_file_in_objdir_buf(&buf,
3565 data->cb, NULL, NULL,
3566 data->data);
3567 strbuf_release(&buf);
3568 return r;
3571 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3573 struct loose_alt_odb_data alt;
3574 int r;
3576 r = for_each_loose_file_in_objdir(get_object_directory(),
3577 cb, NULL, NULL, data);
3578 if (r)
3579 return r;
3581 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3582 return 0;
3584 alt.cb = cb;
3585 alt.data = data;
3586 return foreach_alt_odb(loose_from_alt_odb, &alt);
3589 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3591 uint32_t i;
3592 int r = 0;
3594 for (i = 0; i < p->num_objects; i++) {
3595 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3597 if (!sha1)
3598 return error("unable to get sha1 of object %u in %s",
3599 i, p->pack_name);
3601 r = cb(sha1, p, i, data);
3602 if (r)
3603 break;
3605 return r;
3608 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3610 struct packed_git *p;
3611 int r = 0;
3612 int pack_errors = 0;
3614 prepare_packed_git();
3615 for (p = packed_git; p; p = p->next) {
3616 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3617 continue;
3618 if (open_pack_index(p)) {
3619 pack_errors = 1;
3620 continue;
3622 r = for_each_object_in_pack(p, cb, data);
3623 if (r)
3624 break;
3626 return r ? r : pack_errors;