Merge branch 'ah/show-ref-usage-string'
[git/debian.git] / sha1_file.c
blobd295a3225a1e083708812f99746773d0ec8fdff4
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
38 const unsigned char null_sha1[20];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git *last_found_pack;
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
71 int i;
72 struct cached_object *co = cached_objects;
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
83 int mkdir_in_gitdir(const char *path)
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
90 if (errno != EEXIST)
91 return -1;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
107 strbuf_release(&sb);
109 return adjust_shared_perm(path);
112 enum scld_error safe_create_leading_directories(char *path)
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
124 if (!*slash)
125 break;
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
159 *slash = slash_character;
161 return ret;
164 enum scld_error safe_create_leading_directories_const(const char *path)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
185 const char *sha1_file_name(const unsigned char *sha1)
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
191 objdir = get_object_directory();
192 len = strlen(objdir);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
226 buf = *name;
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
234 return *base;
237 char *sha1_pack_name(const unsigned char *sha1)
239 static char *name, *base;
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1)
246 static char *name, *base;
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
281 strbuf_addstr(&pathbuf, entry);
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
285 pfxlen = strlen(pathbuf.buf);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
335 ent->base[pfxlen] = '/';
337 return 0;
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
375 void read_info_alternates(const char * relative_base, int depth)
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 char *path;
381 int fd;
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
398 munmap(map, mapsz);
401 void add_to_alternates_file(const char *reference)
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 char *alts = git_pathdup("objects/info/alternates");
405 FILE *in, *out;
407 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
408 out = fdopen_lock_file(lock, "w");
409 if (!out)
410 die_errno("unable to fdopen alternates lockfile");
412 in = fopen(alts, "r");
413 if (in) {
414 struct strbuf line = STRBUF_INIT;
415 int found = 0;
417 while (strbuf_getline(&line, in, '\n') != EOF) {
418 if (!strcmp(reference, line.buf)) {
419 found = 1;
420 break;
422 fprintf_or_die(out, "%s\n", line.buf);
425 strbuf_release(&line);
426 fclose(in);
428 if (found) {
429 rollback_lock_file(lock);
430 lock = NULL;
433 else if (errno != ENOENT)
434 die_errno("unable to read alternates file");
436 if (lock) {
437 fprintf_or_die(out, "%s\n", reference);
438 if (commit_lock_file(lock))
439 die_errno("unable to move new alternates file into place");
440 if (alt_odb_tail)
441 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
443 free(alts);
446 int foreach_alt_odb(alt_odb_fn fn, void *cb)
448 struct alternate_object_database *ent;
449 int r = 0;
451 prepare_alt_odb();
452 for (ent = alt_odb_list; ent; ent = ent->next) {
453 r = fn(ent, cb);
454 if (r)
455 break;
457 return r;
460 void prepare_alt_odb(void)
462 const char *alt;
464 if (alt_odb_tail)
465 return;
467 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
468 if (!alt) alt = "";
470 alt_odb_tail = &alt_odb_list;
471 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
473 read_info_alternates(get_object_directory(), 0);
476 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
477 static int freshen_file(const char *fn)
479 struct utimbuf t;
480 t.actime = t.modtime = time(NULL);
481 return !utime(fn, &t);
485 * All of the check_and_freshen functions return 1 if the file exists and was
486 * freshened (if freshening was requested), 0 otherwise. If they return
487 * 0, you should not assume that it is safe to skip a write of the object (it
488 * either does not exist on disk, or has a stale mtime and may be subject to
489 * pruning).
491 static int check_and_freshen_file(const char *fn, int freshen)
493 if (access(fn, F_OK))
494 return 0;
495 if (freshen && !freshen_file(fn))
496 return 0;
497 return 1;
500 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
502 return check_and_freshen_file(sha1_file_name(sha1), freshen);
505 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
507 struct alternate_object_database *alt;
508 prepare_alt_odb();
509 for (alt = alt_odb_list; alt; alt = alt->next) {
510 fill_sha1_path(alt->name, sha1);
511 if (check_and_freshen_file(alt->base, freshen))
512 return 1;
514 return 0;
517 static int check_and_freshen(const unsigned char *sha1, int freshen)
519 return check_and_freshen_local(sha1, freshen) ||
520 check_and_freshen_nonlocal(sha1, freshen);
523 int has_loose_object_nonlocal(const unsigned char *sha1)
525 return check_and_freshen_nonlocal(sha1, 0);
528 static int has_loose_object(const unsigned char *sha1)
530 return check_and_freshen(sha1, 0);
533 static unsigned int pack_used_ctr;
534 static unsigned int pack_mmap_calls;
535 static unsigned int peak_pack_open_windows;
536 static unsigned int pack_open_windows;
537 static unsigned int pack_open_fds;
538 static unsigned int pack_max_fds;
539 static size_t peak_pack_mapped;
540 static size_t pack_mapped;
541 struct packed_git *packed_git;
543 void pack_report(void)
545 fprintf(stderr,
546 "pack_report: getpagesize() = %10" SZ_FMT "\n"
547 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
548 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
549 sz_fmt(getpagesize()),
550 sz_fmt(packed_git_window_size),
551 sz_fmt(packed_git_limit));
552 fprintf(stderr,
553 "pack_report: pack_used_ctr = %10u\n"
554 "pack_report: pack_mmap_calls = %10u\n"
555 "pack_report: pack_open_windows = %10u / %10u\n"
556 "pack_report: pack_mapped = "
557 "%10" SZ_FMT " / %10" SZ_FMT "\n",
558 pack_used_ctr,
559 pack_mmap_calls,
560 pack_open_windows, peak_pack_open_windows,
561 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
565 * Open and mmap the index file at path, perform a couple of
566 * consistency checks, then record its information to p. Return 0 on
567 * success.
569 static int check_packed_git_idx(const char *path, struct packed_git *p)
571 void *idx_map;
572 struct pack_idx_header *hdr;
573 size_t idx_size;
574 uint32_t version, nr, i, *index;
575 int fd = git_open_noatime(path);
576 struct stat st;
578 if (fd < 0)
579 return -1;
580 if (fstat(fd, &st)) {
581 close(fd);
582 return -1;
584 idx_size = xsize_t(st.st_size);
585 if (idx_size < 4 * 256 + 20 + 20) {
586 close(fd);
587 return error("index file %s is too small", path);
589 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
590 close(fd);
592 hdr = idx_map;
593 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
594 version = ntohl(hdr->idx_version);
595 if (version < 2 || version > 2) {
596 munmap(idx_map, idx_size);
597 return error("index file %s is version %"PRIu32
598 " and is not supported by this binary"
599 " (try upgrading GIT to a newer version)",
600 path, version);
602 } else
603 version = 1;
605 nr = 0;
606 index = idx_map;
607 if (version > 1)
608 index += 2; /* skip index header */
609 for (i = 0; i < 256; i++) {
610 uint32_t n = ntohl(index[i]);
611 if (n < nr) {
612 munmap(idx_map, idx_size);
613 return error("non-monotonic index %s", path);
615 nr = n;
618 if (version == 1) {
620 * Total size:
621 * - 256 index entries 4 bytes each
622 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
623 * - 20-byte SHA1 of the packfile
624 * - 20-byte SHA1 file checksum
626 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
627 munmap(idx_map, idx_size);
628 return error("wrong index v1 file size in %s", path);
630 } else if (version == 2) {
632 * Minimum size:
633 * - 8 bytes of header
634 * - 256 index entries 4 bytes each
635 * - 20-byte sha1 entry * nr
636 * - 4-byte crc entry * nr
637 * - 4-byte offset entry * nr
638 * - 20-byte SHA1 of the packfile
639 * - 20-byte SHA1 file checksum
640 * And after the 4-byte offset table might be a
641 * variable sized table containing 8-byte entries
642 * for offsets larger than 2^31.
644 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
645 unsigned long max_size = min_size;
646 if (nr)
647 max_size += (nr - 1)*8;
648 if (idx_size < min_size || idx_size > max_size) {
649 munmap(idx_map, idx_size);
650 return error("wrong index v2 file size in %s", path);
652 if (idx_size != min_size &&
654 * make sure we can deal with large pack offsets.
655 * 31-bit signed offset won't be enough, neither
656 * 32-bit unsigned one will be.
658 (sizeof(off_t) <= 4)) {
659 munmap(idx_map, idx_size);
660 return error("pack too large for current definition of off_t in %s", path);
664 p->index_version = version;
665 p->index_data = idx_map;
666 p->index_size = idx_size;
667 p->num_objects = nr;
668 return 0;
671 int open_pack_index(struct packed_git *p)
673 char *idx_name;
674 int ret;
676 if (p->index_data)
677 return 0;
679 idx_name = xstrdup(p->pack_name);
680 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
681 ret = check_packed_git_idx(idx_name, p);
682 free(idx_name);
683 return ret;
686 static void scan_windows(struct packed_git *p,
687 struct packed_git **lru_p,
688 struct pack_window **lru_w,
689 struct pack_window **lru_l)
691 struct pack_window *w, *w_l;
693 for (w_l = NULL, w = p->windows; w; w = w->next) {
694 if (!w->inuse_cnt) {
695 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
696 *lru_p = p;
697 *lru_w = w;
698 *lru_l = w_l;
701 w_l = w;
705 static int unuse_one_window(struct packed_git *current)
707 struct packed_git *p, *lru_p = NULL;
708 struct pack_window *lru_w = NULL, *lru_l = NULL;
710 if (current)
711 scan_windows(current, &lru_p, &lru_w, &lru_l);
712 for (p = packed_git; p; p = p->next)
713 scan_windows(p, &lru_p, &lru_w, &lru_l);
714 if (lru_p) {
715 munmap(lru_w->base, lru_w->len);
716 pack_mapped -= lru_w->len;
717 if (lru_l)
718 lru_l->next = lru_w->next;
719 else
720 lru_p->windows = lru_w->next;
721 free(lru_w);
722 pack_open_windows--;
723 return 1;
725 return 0;
728 void release_pack_memory(size_t need)
730 size_t cur = pack_mapped;
731 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
732 ; /* nothing */
735 static void mmap_limit_check(size_t length)
737 static size_t limit = 0;
738 if (!limit) {
739 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
740 if (!limit)
741 limit = SIZE_MAX;
743 if (length > limit)
744 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
745 (uintmax_t)length, (uintmax_t)limit);
748 void *xmmap_gently(void *start, size_t length,
749 int prot, int flags, int fd, off_t offset)
751 void *ret;
753 mmap_limit_check(length);
754 ret = mmap(start, length, prot, flags, fd, offset);
755 if (ret == MAP_FAILED) {
756 if (!length)
757 return NULL;
758 release_pack_memory(length);
759 ret = mmap(start, length, prot, flags, fd, offset);
761 return ret;
764 void *xmmap(void *start, size_t length,
765 int prot, int flags, int fd, off_t offset)
767 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
768 if (ret == MAP_FAILED)
769 die_errno("mmap failed");
770 return ret;
773 void close_pack_windows(struct packed_git *p)
775 while (p->windows) {
776 struct pack_window *w = p->windows;
778 if (w->inuse_cnt)
779 die("pack '%s' still has open windows to it",
780 p->pack_name);
781 munmap(w->base, w->len);
782 pack_mapped -= w->len;
783 pack_open_windows--;
784 p->windows = w->next;
785 free(w);
790 * The LRU pack is the one with the oldest MRU window, preferring packs
791 * with no used windows, or the oldest mtime if it has no windows allocated.
793 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
795 struct pack_window *w, *this_mru_w;
796 int has_windows_inuse = 0;
799 * Reject this pack if it has windows and the previously selected
800 * one does not. If this pack does not have windows, reject
801 * it if the pack file is newer than the previously selected one.
803 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
804 return;
806 for (w = this_mru_w = p->windows; w; w = w->next) {
808 * Reject this pack if any of its windows are in use,
809 * but the previously selected pack did not have any
810 * inuse windows. Otherwise, record that this pack
811 * has windows in use.
813 if (w->inuse_cnt) {
814 if (*accept_windows_inuse)
815 has_windows_inuse = 1;
816 else
817 return;
820 if (w->last_used > this_mru_w->last_used)
821 this_mru_w = w;
824 * Reject this pack if it has windows that have been
825 * used more recently than the previously selected pack.
826 * If the previously selected pack had windows inuse and
827 * we have not encountered a window in this pack that is
828 * inuse, skip this check since we prefer a pack with no
829 * inuse windows to one that has inuse windows.
831 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
832 this_mru_w->last_used > (*mru_w)->last_used)
833 return;
837 * Select this pack.
839 *mru_w = this_mru_w;
840 *lru_p = p;
841 *accept_windows_inuse = has_windows_inuse;
844 static int close_one_pack(void)
846 struct packed_git *p, *lru_p = NULL;
847 struct pack_window *mru_w = NULL;
848 int accept_windows_inuse = 1;
850 for (p = packed_git; p; p = p->next) {
851 if (p->pack_fd == -1)
852 continue;
853 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
856 if (lru_p) {
857 close(lru_p->pack_fd);
858 pack_open_fds--;
859 lru_p->pack_fd = -1;
860 return 1;
863 return 0;
866 void unuse_pack(struct pack_window **w_cursor)
868 struct pack_window *w = *w_cursor;
869 if (w) {
870 w->inuse_cnt--;
871 *w_cursor = NULL;
875 void close_pack_index(struct packed_git *p)
877 if (p->index_data) {
878 munmap((void *)p->index_data, p->index_size);
879 p->index_data = NULL;
884 * This is used by git-repack in case a newly created pack happens to
885 * contain the same set of objects as an existing one. In that case
886 * the resulting file might be different even if its name would be the
887 * same. It is best to close any reference to the old pack before it is
888 * replaced on disk. Of course no index pointers or windows for given pack
889 * must subsist at this point. If ever objects from this pack are requested
890 * again, the new version of the pack will be reinitialized through
891 * reprepare_packed_git().
893 void free_pack_by_name(const char *pack_name)
895 struct packed_git *p, **pp = &packed_git;
897 while (*pp) {
898 p = *pp;
899 if (strcmp(pack_name, p->pack_name) == 0) {
900 clear_delta_base_cache();
901 close_pack_windows(p);
902 if (p->pack_fd != -1) {
903 close(p->pack_fd);
904 pack_open_fds--;
906 close_pack_index(p);
907 free(p->bad_object_sha1);
908 *pp = p->next;
909 if (last_found_pack == p)
910 last_found_pack = NULL;
911 free(p);
912 return;
914 pp = &p->next;
918 static unsigned int get_max_fd_limit(void)
920 #ifdef RLIMIT_NOFILE
922 struct rlimit lim;
924 if (!getrlimit(RLIMIT_NOFILE, &lim))
925 return lim.rlim_cur;
927 #endif
929 #ifdef _SC_OPEN_MAX
931 long open_max = sysconf(_SC_OPEN_MAX);
932 if (0 < open_max)
933 return open_max;
935 * Otherwise, we got -1 for one of the two
936 * reasons:
938 * (1) sysconf() did not understand _SC_OPEN_MAX
939 * and signaled an error with -1; or
940 * (2) sysconf() said there is no limit.
942 * We _could_ clear errno before calling sysconf() to
943 * tell these two cases apart and return a huge number
944 * in the latter case to let the caller cap it to a
945 * value that is not so selfish, but letting the
946 * fallback OPEN_MAX codepath take care of these cases
947 * is a lot simpler.
950 #endif
952 #ifdef OPEN_MAX
953 return OPEN_MAX;
954 #else
955 return 1; /* see the caller ;-) */
956 #endif
960 * Do not call this directly as this leaks p->pack_fd on error return;
961 * call open_packed_git() instead.
963 static int open_packed_git_1(struct packed_git *p)
965 struct stat st;
966 struct pack_header hdr;
967 unsigned char sha1[20];
968 unsigned char *idx_sha1;
969 long fd_flag;
971 if (!p->index_data && open_pack_index(p))
972 return error("packfile %s index unavailable", p->pack_name);
974 if (!pack_max_fds) {
975 unsigned int max_fds = get_max_fd_limit();
977 /* Save 3 for stdin/stdout/stderr, 22 for work */
978 if (25 < max_fds)
979 pack_max_fds = max_fds - 25;
980 else
981 pack_max_fds = 1;
984 while (pack_max_fds <= pack_open_fds && close_one_pack())
985 ; /* nothing */
987 p->pack_fd = git_open_noatime(p->pack_name);
988 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
989 return -1;
990 pack_open_fds++;
992 /* If we created the struct before we had the pack we lack size. */
993 if (!p->pack_size) {
994 if (!S_ISREG(st.st_mode))
995 return error("packfile %s not a regular file", p->pack_name);
996 p->pack_size = st.st_size;
997 } else if (p->pack_size != st.st_size)
998 return error("packfile %s size changed", p->pack_name);
1000 /* We leave these file descriptors open with sliding mmap;
1001 * there is no point keeping them open across exec(), though.
1003 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1004 if (fd_flag < 0)
1005 return error("cannot determine file descriptor flags");
1006 fd_flag |= FD_CLOEXEC;
1007 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1008 return error("cannot set FD_CLOEXEC");
1010 /* Verify we recognize this pack file format. */
1011 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1012 return error("file %s is far too short to be a packfile", p->pack_name);
1013 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1014 return error("file %s is not a GIT packfile", p->pack_name);
1015 if (!pack_version_ok(hdr.hdr_version))
1016 return error("packfile %s is version %"PRIu32" and not"
1017 " supported (try upgrading GIT to a newer version)",
1018 p->pack_name, ntohl(hdr.hdr_version));
1020 /* Verify the pack matches its index. */
1021 if (p->num_objects != ntohl(hdr.hdr_entries))
1022 return error("packfile %s claims to have %"PRIu32" objects"
1023 " while index indicates %"PRIu32" objects",
1024 p->pack_name, ntohl(hdr.hdr_entries),
1025 p->num_objects);
1026 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1027 return error("end of packfile %s is unavailable", p->pack_name);
1028 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1029 return error("packfile %s signature is unavailable", p->pack_name);
1030 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1031 if (hashcmp(sha1, idx_sha1))
1032 return error("packfile %s does not match index", p->pack_name);
1033 return 0;
1036 static int open_packed_git(struct packed_git *p)
1038 if (!open_packed_git_1(p))
1039 return 0;
1040 if (p->pack_fd != -1) {
1041 close(p->pack_fd);
1042 pack_open_fds--;
1043 p->pack_fd = -1;
1045 return -1;
1048 static int in_window(struct pack_window *win, off_t offset)
1050 /* We must promise at least 20 bytes (one hash) after the
1051 * offset is available from this window, otherwise the offset
1052 * is not actually in this window and a different window (which
1053 * has that one hash excess) must be used. This is to support
1054 * the object header and delta base parsing routines below.
1056 off_t win_off = win->offset;
1057 return win_off <= offset
1058 && (offset + 20) <= (win_off + win->len);
1061 unsigned char *use_pack(struct packed_git *p,
1062 struct pack_window **w_cursor,
1063 off_t offset,
1064 unsigned long *left)
1066 struct pack_window *win = *w_cursor;
1068 /* Since packfiles end in a hash of their content and it's
1069 * pointless to ask for an offset into the middle of that
1070 * hash, and the in_window function above wouldn't match
1071 * don't allow an offset too close to the end of the file.
1073 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1074 die("packfile %s cannot be accessed", p->pack_name);
1075 if (offset > (p->pack_size - 20))
1076 die("offset beyond end of packfile (truncated pack?)");
1078 if (!win || !in_window(win, offset)) {
1079 if (win)
1080 win->inuse_cnt--;
1081 for (win = p->windows; win; win = win->next) {
1082 if (in_window(win, offset))
1083 break;
1085 if (!win) {
1086 size_t window_align = packed_git_window_size / 2;
1087 off_t len;
1089 if (p->pack_fd == -1 && open_packed_git(p))
1090 die("packfile %s cannot be accessed", p->pack_name);
1092 win = xcalloc(1, sizeof(*win));
1093 win->offset = (offset / window_align) * window_align;
1094 len = p->pack_size - win->offset;
1095 if (len > packed_git_window_size)
1096 len = packed_git_window_size;
1097 win->len = (size_t)len;
1098 pack_mapped += win->len;
1099 while (packed_git_limit < pack_mapped
1100 && unuse_one_window(p))
1101 ; /* nothing */
1102 win->base = xmmap(NULL, win->len,
1103 PROT_READ, MAP_PRIVATE,
1104 p->pack_fd, win->offset);
1105 if (win->base == MAP_FAILED)
1106 die("packfile %s cannot be mapped: %s",
1107 p->pack_name,
1108 strerror(errno));
1109 if (!win->offset && win->len == p->pack_size
1110 && !p->do_not_close) {
1111 close(p->pack_fd);
1112 pack_open_fds--;
1113 p->pack_fd = -1;
1115 pack_mmap_calls++;
1116 pack_open_windows++;
1117 if (pack_mapped > peak_pack_mapped)
1118 peak_pack_mapped = pack_mapped;
1119 if (pack_open_windows > peak_pack_open_windows)
1120 peak_pack_open_windows = pack_open_windows;
1121 win->next = p->windows;
1122 p->windows = win;
1125 if (win != *w_cursor) {
1126 win->last_used = pack_used_ctr++;
1127 win->inuse_cnt++;
1128 *w_cursor = win;
1130 offset -= win->offset;
1131 if (left)
1132 *left = win->len - xsize_t(offset);
1133 return win->base + offset;
1136 static struct packed_git *alloc_packed_git(int extra)
1138 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1139 memset(p, 0, sizeof(*p));
1140 p->pack_fd = -1;
1141 return p;
1144 static void try_to_free_pack_memory(size_t size)
1146 release_pack_memory(size);
1149 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1151 static int have_set_try_to_free_routine;
1152 struct stat st;
1153 struct packed_git *p = alloc_packed_git(path_len + 2);
1155 if (!have_set_try_to_free_routine) {
1156 have_set_try_to_free_routine = 1;
1157 set_try_to_free_routine(try_to_free_pack_memory);
1161 * Make sure a corresponding .pack file exists and that
1162 * the index looks sane.
1164 path_len -= strlen(".idx");
1165 if (path_len < 1) {
1166 free(p);
1167 return NULL;
1169 memcpy(p->pack_name, path, path_len);
1171 strcpy(p->pack_name + path_len, ".keep");
1172 if (!access(p->pack_name, F_OK))
1173 p->pack_keep = 1;
1175 strcpy(p->pack_name + path_len, ".pack");
1176 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1177 free(p);
1178 return NULL;
1181 /* ok, it looks sane as far as we can check without
1182 * actually mapping the pack file.
1184 p->pack_size = st.st_size;
1185 p->pack_local = local;
1186 p->mtime = st.st_mtime;
1187 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1188 hashclr(p->sha1);
1189 return p;
1192 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1194 const char *path = sha1_pack_name(sha1);
1195 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1197 strcpy(p->pack_name, path);
1198 hashcpy(p->sha1, sha1);
1199 if (check_packed_git_idx(idx_path, p)) {
1200 free(p);
1201 return NULL;
1204 return p;
1207 void install_packed_git(struct packed_git *pack)
1209 if (pack->pack_fd != -1)
1210 pack_open_fds++;
1212 pack->next = packed_git;
1213 packed_git = pack;
1216 void (*report_garbage)(const char *desc, const char *path);
1218 static void report_helper(const struct string_list *list,
1219 int seen_bits, int first, int last)
1221 const char *msg;
1222 switch (seen_bits) {
1223 case 0:
1224 msg = "no corresponding .idx or .pack";
1225 break;
1226 case 1:
1227 msg = "no corresponding .idx";
1228 break;
1229 case 2:
1230 msg = "no corresponding .pack";
1231 break;
1232 default:
1233 return;
1235 for (; first < last; first++)
1236 report_garbage(msg, list->items[first].string);
1239 static void report_pack_garbage(struct string_list *list)
1241 int i, baselen = -1, first = 0, seen_bits = 0;
1243 if (!report_garbage)
1244 return;
1246 string_list_sort(list);
1248 for (i = 0; i < list->nr; i++) {
1249 const char *path = list->items[i].string;
1250 if (baselen != -1 &&
1251 strncmp(path, list->items[first].string, baselen)) {
1252 report_helper(list, seen_bits, first, i);
1253 baselen = -1;
1254 seen_bits = 0;
1256 if (baselen == -1) {
1257 const char *dot = strrchr(path, '.');
1258 if (!dot) {
1259 report_garbage("garbage found", path);
1260 continue;
1262 baselen = dot - path + 1;
1263 first = i;
1265 if (!strcmp(path + baselen, "pack"))
1266 seen_bits |= 1;
1267 else if (!strcmp(path + baselen, "idx"))
1268 seen_bits |= 2;
1270 report_helper(list, seen_bits, first, list->nr);
1273 static void prepare_packed_git_one(char *objdir, int local)
1275 struct strbuf path = STRBUF_INIT;
1276 size_t dirnamelen;
1277 DIR *dir;
1278 struct dirent *de;
1279 struct string_list garbage = STRING_LIST_INIT_DUP;
1281 strbuf_addstr(&path, objdir);
1282 strbuf_addstr(&path, "/pack");
1283 dir = opendir(path.buf);
1284 if (!dir) {
1285 if (errno != ENOENT)
1286 error("unable to open object pack directory: %s: %s",
1287 path.buf, strerror(errno));
1288 strbuf_release(&path);
1289 return;
1291 strbuf_addch(&path, '/');
1292 dirnamelen = path.len;
1293 while ((de = readdir(dir)) != NULL) {
1294 struct packed_git *p;
1295 size_t base_len;
1297 if (is_dot_or_dotdot(de->d_name))
1298 continue;
1300 strbuf_setlen(&path, dirnamelen);
1301 strbuf_addstr(&path, de->d_name);
1303 base_len = path.len;
1304 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1305 /* Don't reopen a pack we already have. */
1306 for (p = packed_git; p; p = p->next) {
1307 size_t len;
1308 if (strip_suffix(p->pack_name, ".pack", &len) &&
1309 len == base_len &&
1310 !memcmp(p->pack_name, path.buf, len))
1311 break;
1313 if (p == NULL &&
1315 * See if it really is a valid .idx file with
1316 * corresponding .pack file that we can map.
1318 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1319 install_packed_git(p);
1322 if (!report_garbage)
1323 continue;
1325 if (ends_with(de->d_name, ".idx") ||
1326 ends_with(de->d_name, ".pack") ||
1327 ends_with(de->d_name, ".bitmap") ||
1328 ends_with(de->d_name, ".keep"))
1329 string_list_append(&garbage, path.buf);
1330 else
1331 report_garbage("garbage found", path.buf);
1333 closedir(dir);
1334 report_pack_garbage(&garbage);
1335 string_list_clear(&garbage, 0);
1336 strbuf_release(&path);
1339 static int sort_pack(const void *a_, const void *b_)
1341 struct packed_git *a = *((struct packed_git **)a_);
1342 struct packed_git *b = *((struct packed_git **)b_);
1343 int st;
1346 * Local packs tend to contain objects specific to our
1347 * variant of the project than remote ones. In addition,
1348 * remote ones could be on a network mounted filesystem.
1349 * Favor local ones for these reasons.
1351 st = a->pack_local - b->pack_local;
1352 if (st)
1353 return -st;
1356 * Younger packs tend to contain more recent objects,
1357 * and more recent objects tend to get accessed more
1358 * often.
1360 if (a->mtime < b->mtime)
1361 return 1;
1362 else if (a->mtime == b->mtime)
1363 return 0;
1364 return -1;
1367 static void rearrange_packed_git(void)
1369 struct packed_git **ary, *p;
1370 int i, n;
1372 for (n = 0, p = packed_git; p; p = p->next)
1373 n++;
1374 if (n < 2)
1375 return;
1377 /* prepare an array of packed_git for easier sorting */
1378 ary = xcalloc(n, sizeof(struct packed_git *));
1379 for (n = 0, p = packed_git; p; p = p->next)
1380 ary[n++] = p;
1382 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1384 /* link them back again */
1385 for (i = 0; i < n - 1; i++)
1386 ary[i]->next = ary[i + 1];
1387 ary[n - 1]->next = NULL;
1388 packed_git = ary[0];
1390 free(ary);
1393 static int prepare_packed_git_run_once = 0;
1394 void prepare_packed_git(void)
1396 struct alternate_object_database *alt;
1398 if (prepare_packed_git_run_once)
1399 return;
1400 prepare_packed_git_one(get_object_directory(), 1);
1401 prepare_alt_odb();
1402 for (alt = alt_odb_list; alt; alt = alt->next) {
1403 alt->name[-1] = 0;
1404 prepare_packed_git_one(alt->base, 0);
1405 alt->name[-1] = '/';
1407 rearrange_packed_git();
1408 prepare_packed_git_run_once = 1;
1411 void reprepare_packed_git(void)
1413 prepare_packed_git_run_once = 0;
1414 prepare_packed_git();
1417 static void mark_bad_packed_object(struct packed_git *p,
1418 const unsigned char *sha1)
1420 unsigned i;
1421 for (i = 0; i < p->num_bad_objects; i++)
1422 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1423 return;
1424 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1425 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1426 p->num_bad_objects++;
1429 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1431 struct packed_git *p;
1432 unsigned i;
1434 for (p = packed_git; p; p = p->next)
1435 for (i = 0; i < p->num_bad_objects; i++)
1436 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1437 return p;
1438 return NULL;
1442 * With an in-core object data in "map", rehash it to make sure the
1443 * object name actually matches "sha1" to detect object corruption.
1444 * With "map" == NULL, try reading the object named with "sha1" using
1445 * the streaming interface and rehash it to do the same.
1447 int check_sha1_signature(const unsigned char *sha1, void *map,
1448 unsigned long size, const char *type)
1450 unsigned char real_sha1[20];
1451 enum object_type obj_type;
1452 struct git_istream *st;
1453 git_SHA_CTX c;
1454 char hdr[32];
1455 int hdrlen;
1457 if (map) {
1458 hash_sha1_file(map, size, type, real_sha1);
1459 return hashcmp(sha1, real_sha1) ? -1 : 0;
1462 st = open_istream(sha1, &obj_type, &size, NULL);
1463 if (!st)
1464 return -1;
1466 /* Generate the header */
1467 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1469 /* Sha1.. */
1470 git_SHA1_Init(&c);
1471 git_SHA1_Update(&c, hdr, hdrlen);
1472 for (;;) {
1473 char buf[1024 * 16];
1474 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1476 if (readlen < 0) {
1477 close_istream(st);
1478 return -1;
1480 if (!readlen)
1481 break;
1482 git_SHA1_Update(&c, buf, readlen);
1484 git_SHA1_Final(real_sha1, &c);
1485 close_istream(st);
1486 return hashcmp(sha1, real_sha1) ? -1 : 0;
1489 int git_open_noatime(const char *name)
1491 static int sha1_file_open_flag = O_NOATIME;
1493 for (;;) {
1494 int fd;
1496 errno = 0;
1497 fd = open(name, O_RDONLY | sha1_file_open_flag);
1498 if (fd >= 0)
1499 return fd;
1501 /* Might the failure be due to O_NOATIME? */
1502 if (errno != ENOENT && sha1_file_open_flag) {
1503 sha1_file_open_flag = 0;
1504 continue;
1507 return -1;
1511 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1513 struct alternate_object_database *alt;
1515 if (!lstat(sha1_file_name(sha1), st))
1516 return 0;
1518 prepare_alt_odb();
1519 errno = ENOENT;
1520 for (alt = alt_odb_list; alt; alt = alt->next) {
1521 fill_sha1_path(alt->name, sha1);
1522 if (!lstat(alt->base, st))
1523 return 0;
1526 return -1;
1529 static int open_sha1_file(const unsigned char *sha1)
1531 int fd;
1532 struct alternate_object_database *alt;
1533 int most_interesting_errno;
1535 fd = git_open_noatime(sha1_file_name(sha1));
1536 if (fd >= 0)
1537 return fd;
1538 most_interesting_errno = errno;
1540 prepare_alt_odb();
1541 for (alt = alt_odb_list; alt; alt = alt->next) {
1542 fill_sha1_path(alt->name, sha1);
1543 fd = git_open_noatime(alt->base);
1544 if (fd >= 0)
1545 return fd;
1546 if (most_interesting_errno == ENOENT)
1547 most_interesting_errno = errno;
1549 errno = most_interesting_errno;
1550 return -1;
1553 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1555 void *map;
1556 int fd;
1558 fd = open_sha1_file(sha1);
1559 map = NULL;
1560 if (fd >= 0) {
1561 struct stat st;
1563 if (!fstat(fd, &st)) {
1564 *size = xsize_t(st.st_size);
1565 if (!*size) {
1566 /* mmap() is forbidden on empty files */
1567 error("object file %s is empty", sha1_file_name(sha1));
1568 return NULL;
1570 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1572 close(fd);
1574 return map;
1577 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1578 unsigned long len, enum object_type *type, unsigned long *sizep)
1580 unsigned shift;
1581 unsigned long size, c;
1582 unsigned long used = 0;
1584 c = buf[used++];
1585 *type = (c >> 4) & 7;
1586 size = c & 15;
1587 shift = 4;
1588 while (c & 0x80) {
1589 if (len <= used || bitsizeof(long) <= shift) {
1590 error("bad object header");
1591 size = used = 0;
1592 break;
1594 c = buf[used++];
1595 size += (c & 0x7f) << shift;
1596 shift += 7;
1598 *sizep = size;
1599 return used;
1602 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1604 /* Get the data stream */
1605 memset(stream, 0, sizeof(*stream));
1606 stream->next_in = map;
1607 stream->avail_in = mapsize;
1608 stream->next_out = buffer;
1609 stream->avail_out = bufsiz;
1611 git_inflate_init(stream);
1612 return git_inflate(stream, 0);
1615 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1616 unsigned long mapsize, void *buffer,
1617 unsigned long bufsiz, struct strbuf *header)
1619 int status;
1621 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1624 * Check if entire header is unpacked in the first iteration.
1626 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1627 return 0;
1630 * buffer[0..bufsiz] was not large enough. Copy the partial
1631 * result out to header, and then append the result of further
1632 * reading the stream.
1634 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1635 stream->next_out = buffer;
1636 stream->avail_out = bufsiz;
1638 do {
1639 status = git_inflate(stream, 0);
1640 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1641 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1642 return 0;
1643 stream->next_out = buffer;
1644 stream->avail_out = bufsiz;
1645 } while (status != Z_STREAM_END);
1646 return -1;
1649 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1651 int bytes = strlen(buffer) + 1;
1652 unsigned char *buf = xmallocz(size);
1653 unsigned long n;
1654 int status = Z_OK;
1656 n = stream->total_out - bytes;
1657 if (n > size)
1658 n = size;
1659 memcpy(buf, (char *) buffer + bytes, n);
1660 bytes = n;
1661 if (bytes <= size) {
1663 * The above condition must be (bytes <= size), not
1664 * (bytes < size). In other words, even though we
1665 * expect no more output and set avail_out to zero,
1666 * the input zlib stream may have bytes that express
1667 * "this concludes the stream", and we *do* want to
1668 * eat that input.
1670 * Otherwise we would not be able to test that we
1671 * consumed all the input to reach the expected size;
1672 * we also want to check that zlib tells us that all
1673 * went well with status == Z_STREAM_END at the end.
1675 stream->next_out = buf + bytes;
1676 stream->avail_out = size - bytes;
1677 while (status == Z_OK)
1678 status = git_inflate(stream, Z_FINISH);
1680 if (status == Z_STREAM_END && !stream->avail_in) {
1681 git_inflate_end(stream);
1682 return buf;
1685 if (status < 0)
1686 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1687 else if (stream->avail_in)
1688 error("garbage at end of loose object '%s'",
1689 sha1_to_hex(sha1));
1690 free(buf);
1691 return NULL;
1695 * We used to just use "sscanf()", but that's actually way
1696 * too permissive for what we want to check. So do an anal
1697 * object header parse by hand.
1699 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1700 unsigned int flags)
1702 const char *type_buf = hdr;
1703 unsigned long size;
1704 int type, type_len = 0;
1707 * The type can be of any size but is followed by
1708 * a space.
1710 for (;;) {
1711 char c = *hdr++;
1712 if (c == ' ')
1713 break;
1714 type_len++;
1717 type = type_from_string_gently(type_buf, type_len, 1);
1718 if (oi->typename)
1719 strbuf_add(oi->typename, type_buf, type_len);
1721 * Set type to 0 if its an unknown object and
1722 * we're obtaining the type using '--allow-unkown-type'
1723 * option.
1725 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1726 type = 0;
1727 else if (type < 0)
1728 die("invalid object type");
1729 if (oi->typep)
1730 *oi->typep = type;
1733 * The length must follow immediately, and be in canonical
1734 * decimal format (ie "010" is not valid).
1736 size = *hdr++ - '0';
1737 if (size > 9)
1738 return -1;
1739 if (size) {
1740 for (;;) {
1741 unsigned long c = *hdr - '0';
1742 if (c > 9)
1743 break;
1744 hdr++;
1745 size = size * 10 + c;
1749 if (oi->sizep)
1750 *oi->sizep = size;
1753 * The length must be followed by a zero byte
1755 return *hdr ? -1 : type;
1758 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1760 struct object_info oi;
1762 oi.sizep = sizep;
1763 oi.typename = NULL;
1764 oi.typep = NULL;
1765 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1768 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1770 int ret;
1771 git_zstream stream;
1772 char hdr[8192];
1774 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1775 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1776 return NULL;
1778 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1781 unsigned long get_size_from_delta(struct packed_git *p,
1782 struct pack_window **w_curs,
1783 off_t curpos)
1785 const unsigned char *data;
1786 unsigned char delta_head[20], *in;
1787 git_zstream stream;
1788 int st;
1790 memset(&stream, 0, sizeof(stream));
1791 stream.next_out = delta_head;
1792 stream.avail_out = sizeof(delta_head);
1794 git_inflate_init(&stream);
1795 do {
1796 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1797 stream.next_in = in;
1798 st = git_inflate(&stream, Z_FINISH);
1799 curpos += stream.next_in - in;
1800 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1801 stream.total_out < sizeof(delta_head));
1802 git_inflate_end(&stream);
1803 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1804 error("delta data unpack-initial failed");
1805 return 0;
1808 /* Examine the initial part of the delta to figure out
1809 * the result size.
1811 data = delta_head;
1813 /* ignore base size */
1814 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1816 /* Read the result size */
1817 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1820 static off_t get_delta_base(struct packed_git *p,
1821 struct pack_window **w_curs,
1822 off_t *curpos,
1823 enum object_type type,
1824 off_t delta_obj_offset)
1826 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1827 off_t base_offset;
1829 /* use_pack() assured us we have [base_info, base_info + 20)
1830 * as a range that we can look at without walking off the
1831 * end of the mapped window. Its actually the hash size
1832 * that is assured. An OFS_DELTA longer than the hash size
1833 * is stupid, as then a REF_DELTA would be smaller to store.
1835 if (type == OBJ_OFS_DELTA) {
1836 unsigned used = 0;
1837 unsigned char c = base_info[used++];
1838 base_offset = c & 127;
1839 while (c & 128) {
1840 base_offset += 1;
1841 if (!base_offset || MSB(base_offset, 7))
1842 return 0; /* overflow */
1843 c = base_info[used++];
1844 base_offset = (base_offset << 7) + (c & 127);
1846 base_offset = delta_obj_offset - base_offset;
1847 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1848 return 0; /* out of bound */
1849 *curpos += used;
1850 } else if (type == OBJ_REF_DELTA) {
1851 /* The base entry _must_ be in the same pack */
1852 base_offset = find_pack_entry_one(base_info, p);
1853 *curpos += 20;
1854 } else
1855 die("I am totally screwed");
1856 return base_offset;
1860 * Like get_delta_base above, but we return the sha1 instead of the pack
1861 * offset. This means it is cheaper for REF deltas (we do not have to do
1862 * the final object lookup), but more expensive for OFS deltas (we
1863 * have to load the revidx to convert the offset back into a sha1).
1865 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1866 struct pack_window **w_curs,
1867 off_t curpos,
1868 enum object_type type,
1869 off_t delta_obj_offset)
1871 if (type == OBJ_REF_DELTA) {
1872 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1873 return base;
1874 } else if (type == OBJ_OFS_DELTA) {
1875 struct revindex_entry *revidx;
1876 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1877 type, delta_obj_offset);
1879 if (!base_offset)
1880 return NULL;
1882 revidx = find_pack_revindex(p, base_offset);
1883 if (!revidx)
1884 return NULL;
1886 return nth_packed_object_sha1(p, revidx->nr);
1887 } else
1888 return NULL;
1891 int unpack_object_header(struct packed_git *p,
1892 struct pack_window **w_curs,
1893 off_t *curpos,
1894 unsigned long *sizep)
1896 unsigned char *base;
1897 unsigned long left;
1898 unsigned long used;
1899 enum object_type type;
1901 /* use_pack() assures us we have [base, base + 20) available
1902 * as a range that we can look at. (Its actually the hash
1903 * size that is assured.) With our object header encoding
1904 * the maximum deflated object size is 2^137, which is just
1905 * insane, so we know won't exceed what we have been given.
1907 base = use_pack(p, w_curs, *curpos, &left);
1908 used = unpack_object_header_buffer(base, left, &type, sizep);
1909 if (!used) {
1910 type = OBJ_BAD;
1911 } else
1912 *curpos += used;
1914 return type;
1917 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1919 int type;
1920 struct revindex_entry *revidx;
1921 const unsigned char *sha1;
1922 revidx = find_pack_revindex(p, obj_offset);
1923 if (!revidx)
1924 return OBJ_BAD;
1925 sha1 = nth_packed_object_sha1(p, revidx->nr);
1926 mark_bad_packed_object(p, sha1);
1927 type = sha1_object_info(sha1, NULL);
1928 if (type <= OBJ_NONE)
1929 return OBJ_BAD;
1930 return type;
1933 #define POI_STACK_PREALLOC 64
1935 static enum object_type packed_to_object_type(struct packed_git *p,
1936 off_t obj_offset,
1937 enum object_type type,
1938 struct pack_window **w_curs,
1939 off_t curpos)
1941 off_t small_poi_stack[POI_STACK_PREALLOC];
1942 off_t *poi_stack = small_poi_stack;
1943 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1945 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1946 off_t base_offset;
1947 unsigned long size;
1948 /* Push the object we're going to leave behind */
1949 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1950 poi_stack_alloc = alloc_nr(poi_stack_nr);
1951 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1952 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1953 } else {
1954 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1956 poi_stack[poi_stack_nr++] = obj_offset;
1957 /* If parsing the base offset fails, just unwind */
1958 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1959 if (!base_offset)
1960 goto unwind;
1961 curpos = obj_offset = base_offset;
1962 type = unpack_object_header(p, w_curs, &curpos, &size);
1963 if (type <= OBJ_NONE) {
1964 /* If getting the base itself fails, we first
1965 * retry the base, otherwise unwind */
1966 type = retry_bad_packed_offset(p, base_offset);
1967 if (type > OBJ_NONE)
1968 goto out;
1969 goto unwind;
1973 switch (type) {
1974 case OBJ_BAD:
1975 case OBJ_COMMIT:
1976 case OBJ_TREE:
1977 case OBJ_BLOB:
1978 case OBJ_TAG:
1979 break;
1980 default:
1981 error("unknown object type %i at offset %"PRIuMAX" in %s",
1982 type, (uintmax_t)obj_offset, p->pack_name);
1983 type = OBJ_BAD;
1986 out:
1987 if (poi_stack != small_poi_stack)
1988 free(poi_stack);
1989 return type;
1991 unwind:
1992 while (poi_stack_nr) {
1993 obj_offset = poi_stack[--poi_stack_nr];
1994 type = retry_bad_packed_offset(p, obj_offset);
1995 if (type > OBJ_NONE)
1996 goto out;
1998 type = OBJ_BAD;
1999 goto out;
2002 static int packed_object_info(struct packed_git *p, off_t obj_offset,
2003 struct object_info *oi)
2005 struct pack_window *w_curs = NULL;
2006 unsigned long size;
2007 off_t curpos = obj_offset;
2008 enum object_type type;
2011 * We always get the representation type, but only convert it to
2012 * a "real" type later if the caller is interested.
2014 type = unpack_object_header(p, &w_curs, &curpos, &size);
2016 if (oi->sizep) {
2017 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2018 off_t tmp_pos = curpos;
2019 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2020 type, obj_offset);
2021 if (!base_offset) {
2022 type = OBJ_BAD;
2023 goto out;
2025 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2026 if (*oi->sizep == 0) {
2027 type = OBJ_BAD;
2028 goto out;
2030 } else {
2031 *oi->sizep = size;
2035 if (oi->disk_sizep) {
2036 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2037 *oi->disk_sizep = revidx[1].offset - obj_offset;
2040 if (oi->typep) {
2041 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2042 if (*oi->typep < 0) {
2043 type = OBJ_BAD;
2044 goto out;
2048 if (oi->delta_base_sha1) {
2049 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2050 const unsigned char *base;
2052 base = get_delta_base_sha1(p, &w_curs, curpos,
2053 type, obj_offset);
2054 if (!base) {
2055 type = OBJ_BAD;
2056 goto out;
2059 hashcpy(oi->delta_base_sha1, base);
2060 } else
2061 hashclr(oi->delta_base_sha1);
2064 out:
2065 unuse_pack(&w_curs);
2066 return type;
2069 static void *unpack_compressed_entry(struct packed_git *p,
2070 struct pack_window **w_curs,
2071 off_t curpos,
2072 unsigned long size)
2074 int st;
2075 git_zstream stream;
2076 unsigned char *buffer, *in;
2078 buffer = xmallocz_gently(size);
2079 if (!buffer)
2080 return NULL;
2081 memset(&stream, 0, sizeof(stream));
2082 stream.next_out = buffer;
2083 stream.avail_out = size + 1;
2085 git_inflate_init(&stream);
2086 do {
2087 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2088 stream.next_in = in;
2089 st = git_inflate(&stream, Z_FINISH);
2090 if (!stream.avail_out)
2091 break; /* the payload is larger than it should be */
2092 curpos += stream.next_in - in;
2093 } while (st == Z_OK || st == Z_BUF_ERROR);
2094 git_inflate_end(&stream);
2095 if ((st != Z_STREAM_END) || stream.total_out != size) {
2096 free(buffer);
2097 return NULL;
2100 return buffer;
2103 #define MAX_DELTA_CACHE (256)
2105 static size_t delta_base_cached;
2107 static struct delta_base_cache_lru_list {
2108 struct delta_base_cache_lru_list *prev;
2109 struct delta_base_cache_lru_list *next;
2110 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2112 static struct delta_base_cache_entry {
2113 struct delta_base_cache_lru_list lru;
2114 void *data;
2115 struct packed_git *p;
2116 off_t base_offset;
2117 unsigned long size;
2118 enum object_type type;
2119 } delta_base_cache[MAX_DELTA_CACHE];
2121 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2123 unsigned long hash;
2125 hash = (unsigned long)p + (unsigned long)base_offset;
2126 hash += (hash >> 8) + (hash >> 16);
2127 return hash % MAX_DELTA_CACHE;
2130 static struct delta_base_cache_entry *
2131 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2133 unsigned long hash = pack_entry_hash(p, base_offset);
2134 return delta_base_cache + hash;
2137 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2138 struct packed_git *p, off_t base_offset)
2140 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2143 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2145 struct delta_base_cache_entry *ent;
2146 ent = get_delta_base_cache_entry(p, base_offset);
2147 return eq_delta_base_cache_entry(ent, p, base_offset);
2150 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2152 ent->data = NULL;
2153 ent->lru.next->prev = ent->lru.prev;
2154 ent->lru.prev->next = ent->lru.next;
2155 delta_base_cached -= ent->size;
2158 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2159 unsigned long *base_size, enum object_type *type, int keep_cache)
2161 struct delta_base_cache_entry *ent;
2162 void *ret;
2164 ent = get_delta_base_cache_entry(p, base_offset);
2166 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2167 return unpack_entry(p, base_offset, type, base_size);
2169 ret = ent->data;
2171 if (!keep_cache)
2172 clear_delta_base_cache_entry(ent);
2173 else
2174 ret = xmemdupz(ent->data, ent->size);
2175 *type = ent->type;
2176 *base_size = ent->size;
2177 return ret;
2180 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2182 if (ent->data) {
2183 free(ent->data);
2184 ent->data = NULL;
2185 ent->lru.next->prev = ent->lru.prev;
2186 ent->lru.prev->next = ent->lru.next;
2187 delta_base_cached -= ent->size;
2191 void clear_delta_base_cache(void)
2193 unsigned long p;
2194 for (p = 0; p < MAX_DELTA_CACHE; p++)
2195 release_delta_base_cache(&delta_base_cache[p]);
2198 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2199 void *base, unsigned long base_size, enum object_type type)
2201 unsigned long hash = pack_entry_hash(p, base_offset);
2202 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2203 struct delta_base_cache_lru_list *lru;
2205 release_delta_base_cache(ent);
2206 delta_base_cached += base_size;
2208 for (lru = delta_base_cache_lru.next;
2209 delta_base_cached > delta_base_cache_limit
2210 && lru != &delta_base_cache_lru;
2211 lru = lru->next) {
2212 struct delta_base_cache_entry *f = (void *)lru;
2213 if (f->type == OBJ_BLOB)
2214 release_delta_base_cache(f);
2216 for (lru = delta_base_cache_lru.next;
2217 delta_base_cached > delta_base_cache_limit
2218 && lru != &delta_base_cache_lru;
2219 lru = lru->next) {
2220 struct delta_base_cache_entry *f = (void *)lru;
2221 release_delta_base_cache(f);
2224 ent->p = p;
2225 ent->base_offset = base_offset;
2226 ent->type = type;
2227 ent->data = base;
2228 ent->size = base_size;
2229 ent->lru.next = &delta_base_cache_lru;
2230 ent->lru.prev = delta_base_cache_lru.prev;
2231 delta_base_cache_lru.prev->next = &ent->lru;
2232 delta_base_cache_lru.prev = &ent->lru;
2235 static void *read_object(const unsigned char *sha1, enum object_type *type,
2236 unsigned long *size);
2238 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2240 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2241 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2242 p->pack_name, (uintmax_t)obj_offset);
2245 int do_check_packed_object_crc;
2247 #define UNPACK_ENTRY_STACK_PREALLOC 64
2248 struct unpack_entry_stack_ent {
2249 off_t obj_offset;
2250 off_t curpos;
2251 unsigned long size;
2254 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2255 enum object_type *final_type, unsigned long *final_size)
2257 struct pack_window *w_curs = NULL;
2258 off_t curpos = obj_offset;
2259 void *data = NULL;
2260 unsigned long size;
2261 enum object_type type;
2262 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2263 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2264 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2265 int base_from_cache = 0;
2267 write_pack_access_log(p, obj_offset);
2269 /* PHASE 1: drill down to the innermost base object */
2270 for (;;) {
2271 off_t base_offset;
2272 int i;
2273 struct delta_base_cache_entry *ent;
2275 ent = get_delta_base_cache_entry(p, curpos);
2276 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2277 type = ent->type;
2278 data = ent->data;
2279 size = ent->size;
2280 clear_delta_base_cache_entry(ent);
2281 base_from_cache = 1;
2282 break;
2285 if (do_check_packed_object_crc && p->index_version > 1) {
2286 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2287 unsigned long len = revidx[1].offset - obj_offset;
2288 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2289 const unsigned char *sha1 =
2290 nth_packed_object_sha1(p, revidx->nr);
2291 error("bad packed object CRC for %s",
2292 sha1_to_hex(sha1));
2293 mark_bad_packed_object(p, sha1);
2294 unuse_pack(&w_curs);
2295 return NULL;
2299 type = unpack_object_header(p, &w_curs, &curpos, &size);
2300 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2301 break;
2303 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2304 if (!base_offset) {
2305 error("failed to validate delta base reference "
2306 "at offset %"PRIuMAX" from %s",
2307 (uintmax_t)curpos, p->pack_name);
2308 /* bail to phase 2, in hopes of recovery */
2309 data = NULL;
2310 break;
2313 /* push object, proceed to base */
2314 if (delta_stack_nr >= delta_stack_alloc
2315 && delta_stack == small_delta_stack) {
2316 delta_stack_alloc = alloc_nr(delta_stack_nr);
2317 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2318 memcpy(delta_stack, small_delta_stack,
2319 sizeof(*delta_stack)*delta_stack_nr);
2320 } else {
2321 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2323 i = delta_stack_nr++;
2324 delta_stack[i].obj_offset = obj_offset;
2325 delta_stack[i].curpos = curpos;
2326 delta_stack[i].size = size;
2328 curpos = obj_offset = base_offset;
2331 /* PHASE 2: handle the base */
2332 switch (type) {
2333 case OBJ_OFS_DELTA:
2334 case OBJ_REF_DELTA:
2335 if (data)
2336 die("BUG in unpack_entry: left loop at a valid delta");
2337 break;
2338 case OBJ_COMMIT:
2339 case OBJ_TREE:
2340 case OBJ_BLOB:
2341 case OBJ_TAG:
2342 if (!base_from_cache)
2343 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2344 break;
2345 default:
2346 data = NULL;
2347 error("unknown object type %i at offset %"PRIuMAX" in %s",
2348 type, (uintmax_t)obj_offset, p->pack_name);
2351 /* PHASE 3: apply deltas in order */
2353 /* invariants:
2354 * 'data' holds the base data, or NULL if there was corruption
2356 while (delta_stack_nr) {
2357 void *delta_data;
2358 void *base = data;
2359 unsigned long delta_size, base_size = size;
2360 int i;
2362 data = NULL;
2364 if (base)
2365 add_delta_base_cache(p, obj_offset, base, base_size, type);
2367 if (!base) {
2369 * We're probably in deep shit, but let's try to fetch
2370 * the required base anyway from another pack or loose.
2371 * This is costly but should happen only in the presence
2372 * of a corrupted pack, and is better than failing outright.
2374 struct revindex_entry *revidx;
2375 const unsigned char *base_sha1;
2376 revidx = find_pack_revindex(p, obj_offset);
2377 if (revidx) {
2378 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2379 error("failed to read delta base object %s"
2380 " at offset %"PRIuMAX" from %s",
2381 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2382 p->pack_name);
2383 mark_bad_packed_object(p, base_sha1);
2384 base = read_object(base_sha1, &type, &base_size);
2388 i = --delta_stack_nr;
2389 obj_offset = delta_stack[i].obj_offset;
2390 curpos = delta_stack[i].curpos;
2391 delta_size = delta_stack[i].size;
2393 if (!base)
2394 continue;
2396 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2398 if (!delta_data) {
2399 error("failed to unpack compressed delta "
2400 "at offset %"PRIuMAX" from %s",
2401 (uintmax_t)curpos, p->pack_name);
2402 data = NULL;
2403 continue;
2406 data = patch_delta(base, base_size,
2407 delta_data, delta_size,
2408 &size);
2411 * We could not apply the delta; warn the user, but keep going.
2412 * Our failure will be noticed either in the next iteration of
2413 * the loop, or if this is the final delta, in the caller when
2414 * we return NULL. Those code paths will take care of making
2415 * a more explicit warning and retrying with another copy of
2416 * the object.
2418 if (!data)
2419 error("failed to apply delta");
2421 free(delta_data);
2424 *final_type = type;
2425 *final_size = size;
2427 unuse_pack(&w_curs);
2429 if (delta_stack != small_delta_stack)
2430 free(delta_stack);
2432 return data;
2435 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2436 uint32_t n)
2438 const unsigned char *index = p->index_data;
2439 if (!index) {
2440 if (open_pack_index(p))
2441 return NULL;
2442 index = p->index_data;
2444 if (n >= p->num_objects)
2445 return NULL;
2446 index += 4 * 256;
2447 if (p->index_version == 1) {
2448 return index + 24 * n + 4;
2449 } else {
2450 index += 8;
2451 return index + 20 * n;
2455 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2457 const unsigned char *index = p->index_data;
2458 index += 4 * 256;
2459 if (p->index_version == 1) {
2460 return ntohl(*((uint32_t *)(index + 24 * n)));
2461 } else {
2462 uint32_t off;
2463 index += 8 + p->num_objects * (20 + 4);
2464 off = ntohl(*((uint32_t *)(index + 4 * n)));
2465 if (!(off & 0x80000000))
2466 return off;
2467 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2468 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2469 ntohl(*((uint32_t *)(index + 4)));
2473 off_t find_pack_entry_one(const unsigned char *sha1,
2474 struct packed_git *p)
2476 const uint32_t *level1_ofs = p->index_data;
2477 const unsigned char *index = p->index_data;
2478 unsigned hi, lo, stride;
2479 static int use_lookup = -1;
2480 static int debug_lookup = -1;
2482 if (debug_lookup < 0)
2483 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2485 if (!index) {
2486 if (open_pack_index(p))
2487 return 0;
2488 level1_ofs = p->index_data;
2489 index = p->index_data;
2491 if (p->index_version > 1) {
2492 level1_ofs += 2;
2493 index += 8;
2495 index += 4 * 256;
2496 hi = ntohl(level1_ofs[*sha1]);
2497 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2498 if (p->index_version > 1) {
2499 stride = 20;
2500 } else {
2501 stride = 24;
2502 index += 4;
2505 if (debug_lookup)
2506 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2507 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2509 if (use_lookup < 0)
2510 use_lookup = !!getenv("GIT_USE_LOOKUP");
2511 if (use_lookup) {
2512 int pos = sha1_entry_pos(index, stride, 0,
2513 lo, hi, p->num_objects, sha1);
2514 if (pos < 0)
2515 return 0;
2516 return nth_packed_object_offset(p, pos);
2519 do {
2520 unsigned mi = (lo + hi) / 2;
2521 int cmp = hashcmp(index + mi * stride, sha1);
2523 if (debug_lookup)
2524 printf("lo %u hi %u rg %u mi %u\n",
2525 lo, hi, hi - lo, mi);
2526 if (!cmp)
2527 return nth_packed_object_offset(p, mi);
2528 if (cmp > 0)
2529 hi = mi;
2530 else
2531 lo = mi+1;
2532 } while (lo < hi);
2533 return 0;
2536 int is_pack_valid(struct packed_git *p)
2538 /* An already open pack is known to be valid. */
2539 if (p->pack_fd != -1)
2540 return 1;
2542 /* If the pack has one window completely covering the
2543 * file size, the pack is known to be valid even if
2544 * the descriptor is not currently open.
2546 if (p->windows) {
2547 struct pack_window *w = p->windows;
2549 if (!w->offset && w->len == p->pack_size)
2550 return 1;
2553 /* Force the pack to open to prove its valid. */
2554 return !open_packed_git(p);
2557 static int fill_pack_entry(const unsigned char *sha1,
2558 struct pack_entry *e,
2559 struct packed_git *p)
2561 off_t offset;
2563 if (p->num_bad_objects) {
2564 unsigned i;
2565 for (i = 0; i < p->num_bad_objects; i++)
2566 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2567 return 0;
2570 offset = find_pack_entry_one(sha1, p);
2571 if (!offset)
2572 return 0;
2575 * We are about to tell the caller where they can locate the
2576 * requested object. We better make sure the packfile is
2577 * still here and can be accessed before supplying that
2578 * answer, as it may have been deleted since the index was
2579 * loaded!
2581 if (!is_pack_valid(p))
2582 return 0;
2583 e->offset = offset;
2584 e->p = p;
2585 hashcpy(e->sha1, sha1);
2586 return 1;
2590 * Iff a pack file contains the object named by sha1, return true and
2591 * store its location to e.
2593 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2595 struct packed_git *p;
2597 prepare_packed_git();
2598 if (!packed_git)
2599 return 0;
2601 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2602 return 1;
2604 for (p = packed_git; p; p = p->next) {
2605 if (p == last_found_pack)
2606 continue; /* we already checked this one */
2608 if (fill_pack_entry(sha1, e, p)) {
2609 last_found_pack = p;
2610 return 1;
2613 return 0;
2616 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2617 struct packed_git *packs)
2619 struct packed_git *p;
2621 for (p = packs; p; p = p->next) {
2622 if (find_pack_entry_one(sha1, p))
2623 return p;
2625 return NULL;
2629 static int sha1_loose_object_info(const unsigned char *sha1,
2630 struct object_info *oi,
2631 int flags)
2633 int status = 0;
2634 unsigned long mapsize;
2635 void *map;
2636 git_zstream stream;
2637 char hdr[32];
2638 struct strbuf hdrbuf = STRBUF_INIT;
2640 if (oi->delta_base_sha1)
2641 hashclr(oi->delta_base_sha1);
2644 * If we don't care about type or size, then we don't
2645 * need to look inside the object at all. Note that we
2646 * do not optimize out the stat call, even if the
2647 * caller doesn't care about the disk-size, since our
2648 * return value implicitly indicates whether the
2649 * object even exists.
2651 if (!oi->typep && !oi->typename && !oi->sizep) {
2652 struct stat st;
2653 if (stat_sha1_file(sha1, &st) < 0)
2654 return -1;
2655 if (oi->disk_sizep)
2656 *oi->disk_sizep = st.st_size;
2657 return 0;
2660 map = map_sha1_file(sha1, &mapsize);
2661 if (!map)
2662 return -1;
2663 if (oi->disk_sizep)
2664 *oi->disk_sizep = mapsize;
2665 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2666 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2667 status = error("unable to unpack %s header with --allow-unknown-type",
2668 sha1_to_hex(sha1));
2669 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2670 status = error("unable to unpack %s header",
2671 sha1_to_hex(sha1));
2672 if (status < 0)
2673 ; /* Do nothing */
2674 else if (hdrbuf.len) {
2675 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2676 status = error("unable to parse %s header with --allow-unknown-type",
2677 sha1_to_hex(sha1));
2678 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2679 status = error("unable to parse %s header", sha1_to_hex(sha1));
2680 git_inflate_end(&stream);
2681 munmap(map, mapsize);
2682 if (status && oi->typep)
2683 *oi->typep = status;
2684 strbuf_release(&hdrbuf);
2685 return 0;
2688 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2690 struct cached_object *co;
2691 struct pack_entry e;
2692 int rtype;
2693 enum object_type real_type;
2694 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2696 co = find_cached_object(real);
2697 if (co) {
2698 if (oi->typep)
2699 *(oi->typep) = co->type;
2700 if (oi->sizep)
2701 *(oi->sizep) = co->size;
2702 if (oi->disk_sizep)
2703 *(oi->disk_sizep) = 0;
2704 if (oi->delta_base_sha1)
2705 hashclr(oi->delta_base_sha1);
2706 if (oi->typename)
2707 strbuf_addstr(oi->typename, typename(co->type));
2708 oi->whence = OI_CACHED;
2709 return 0;
2712 if (!find_pack_entry(real, &e)) {
2713 /* Most likely it's a loose object. */
2714 if (!sha1_loose_object_info(real, oi, flags)) {
2715 oi->whence = OI_LOOSE;
2716 return 0;
2719 /* Not a loose object; someone else may have just packed it. */
2720 reprepare_packed_git();
2721 if (!find_pack_entry(real, &e))
2722 return -1;
2726 * packed_object_info() does not follow the delta chain to
2727 * find out the real type, unless it is given oi->typep.
2729 if (oi->typename && !oi->typep)
2730 oi->typep = &real_type;
2732 rtype = packed_object_info(e.p, e.offset, oi);
2733 if (rtype < 0) {
2734 mark_bad_packed_object(e.p, real);
2735 if (oi->typep == &real_type)
2736 oi->typep = NULL;
2737 return sha1_object_info_extended(real, oi, 0);
2738 } else if (in_delta_base_cache(e.p, e.offset)) {
2739 oi->whence = OI_DBCACHED;
2740 } else {
2741 oi->whence = OI_PACKED;
2742 oi->u.packed.offset = e.offset;
2743 oi->u.packed.pack = e.p;
2744 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2745 rtype == OBJ_OFS_DELTA);
2747 if (oi->typename)
2748 strbuf_addstr(oi->typename, typename(*oi->typep));
2749 if (oi->typep == &real_type)
2750 oi->typep = NULL;
2752 return 0;
2755 /* returns enum object_type or negative */
2756 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2758 enum object_type type;
2759 struct object_info oi = {NULL};
2761 oi.typep = &type;
2762 oi.sizep = sizep;
2763 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2764 return -1;
2765 return type;
2768 static void *read_packed_sha1(const unsigned char *sha1,
2769 enum object_type *type, unsigned long *size)
2771 struct pack_entry e;
2772 void *data;
2774 if (!find_pack_entry(sha1, &e))
2775 return NULL;
2776 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2777 if (!data) {
2779 * We're probably in deep shit, but let's try to fetch
2780 * the required object anyway from another pack or loose.
2781 * This should happen only in the presence of a corrupted
2782 * pack, and is better than failing outright.
2784 error("failed to read object %s at offset %"PRIuMAX" from %s",
2785 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2786 mark_bad_packed_object(e.p, sha1);
2787 data = read_object(sha1, type, size);
2789 return data;
2792 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2793 unsigned char *sha1)
2795 struct cached_object *co;
2797 hash_sha1_file(buf, len, typename(type), sha1);
2798 if (has_sha1_file(sha1) || find_cached_object(sha1))
2799 return 0;
2800 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2801 co = &cached_objects[cached_object_nr++];
2802 co->size = len;
2803 co->type = type;
2804 co->buf = xmalloc(len);
2805 memcpy(co->buf, buf, len);
2806 hashcpy(co->sha1, sha1);
2807 return 0;
2810 static void *read_object(const unsigned char *sha1, enum object_type *type,
2811 unsigned long *size)
2813 unsigned long mapsize;
2814 void *map, *buf;
2815 struct cached_object *co;
2817 co = find_cached_object(sha1);
2818 if (co) {
2819 *type = co->type;
2820 *size = co->size;
2821 return xmemdupz(co->buf, co->size);
2824 buf = read_packed_sha1(sha1, type, size);
2825 if (buf)
2826 return buf;
2827 map = map_sha1_file(sha1, &mapsize);
2828 if (map) {
2829 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2830 munmap(map, mapsize);
2831 return buf;
2833 reprepare_packed_git();
2834 return read_packed_sha1(sha1, type, size);
2838 * This function dies on corrupt objects; the callers who want to
2839 * deal with them should arrange to call read_object() and give error
2840 * messages themselves.
2842 void *read_sha1_file_extended(const unsigned char *sha1,
2843 enum object_type *type,
2844 unsigned long *size,
2845 unsigned flag)
2847 void *data;
2848 const struct packed_git *p;
2849 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2851 errno = 0;
2852 data = read_object(repl, type, size);
2853 if (data)
2854 return data;
2856 if (errno && errno != ENOENT)
2857 die_errno("failed to read object %s", sha1_to_hex(sha1));
2859 /* die if we replaced an object with one that does not exist */
2860 if (repl != sha1)
2861 die("replacement %s not found for %s",
2862 sha1_to_hex(repl), sha1_to_hex(sha1));
2864 if (has_loose_object(repl)) {
2865 const char *path = sha1_file_name(sha1);
2867 die("loose object %s (stored in %s) is corrupt",
2868 sha1_to_hex(repl), path);
2871 if ((p = has_packed_and_bad(repl)) != NULL)
2872 die("packed object %s (stored in %s) is corrupt",
2873 sha1_to_hex(repl), p->pack_name);
2875 return NULL;
2878 void *read_object_with_reference(const unsigned char *sha1,
2879 const char *required_type_name,
2880 unsigned long *size,
2881 unsigned char *actual_sha1_return)
2883 enum object_type type, required_type;
2884 void *buffer;
2885 unsigned long isize;
2886 unsigned char actual_sha1[20];
2888 required_type = type_from_string(required_type_name);
2889 hashcpy(actual_sha1, sha1);
2890 while (1) {
2891 int ref_length = -1;
2892 const char *ref_type = NULL;
2894 buffer = read_sha1_file(actual_sha1, &type, &isize);
2895 if (!buffer)
2896 return NULL;
2897 if (type == required_type) {
2898 *size = isize;
2899 if (actual_sha1_return)
2900 hashcpy(actual_sha1_return, actual_sha1);
2901 return buffer;
2903 /* Handle references */
2904 else if (type == OBJ_COMMIT)
2905 ref_type = "tree ";
2906 else if (type == OBJ_TAG)
2907 ref_type = "object ";
2908 else {
2909 free(buffer);
2910 return NULL;
2912 ref_length = strlen(ref_type);
2914 if (ref_length + 40 > isize ||
2915 memcmp(buffer, ref_type, ref_length) ||
2916 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2917 free(buffer);
2918 return NULL;
2920 free(buffer);
2921 /* Now we have the ID of the referred-to object in
2922 * actual_sha1. Check again. */
2926 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2927 const char *type, unsigned char *sha1,
2928 char *hdr, int *hdrlen)
2930 git_SHA_CTX c;
2932 /* Generate the header */
2933 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2935 /* Sha1.. */
2936 git_SHA1_Init(&c);
2937 git_SHA1_Update(&c, hdr, *hdrlen);
2938 git_SHA1_Update(&c, buf, len);
2939 git_SHA1_Final(sha1, &c);
2943 * Move the just written object into its final resting place.
2945 int finalize_object_file(const char *tmpfile, const char *filename)
2947 int ret = 0;
2949 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2950 goto try_rename;
2951 else if (link(tmpfile, filename))
2952 ret = errno;
2955 * Coda hack - coda doesn't like cross-directory links,
2956 * so we fall back to a rename, which will mean that it
2957 * won't be able to check collisions, but that's not a
2958 * big deal.
2960 * The same holds for FAT formatted media.
2962 * When this succeeds, we just return. We have nothing
2963 * left to unlink.
2965 if (ret && ret != EEXIST) {
2966 try_rename:
2967 if (!rename(tmpfile, filename))
2968 goto out;
2969 ret = errno;
2971 unlink_or_warn(tmpfile);
2972 if (ret) {
2973 if (ret != EEXIST) {
2974 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2976 /* FIXME!!! Collision check here ? */
2979 out:
2980 if (adjust_shared_perm(filename))
2981 return error("unable to set permission to '%s'", filename);
2982 return 0;
2985 static int write_buffer(int fd, const void *buf, size_t len)
2987 if (write_in_full(fd, buf, len) < 0)
2988 return error("file write error (%s)", strerror(errno));
2989 return 0;
2992 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2993 unsigned char *sha1)
2995 char hdr[32];
2996 int hdrlen;
2997 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2998 return 0;
3001 /* Finalize a file on disk, and close it. */
3002 static void close_sha1_file(int fd)
3004 if (fsync_object_files)
3005 fsync_or_die(fd, "sha1 file");
3006 if (close(fd) != 0)
3007 die_errno("error when closing sha1 file");
3010 /* Size of directory component, including the ending '/' */
3011 static inline int directory_size(const char *filename)
3013 const char *s = strrchr(filename, '/');
3014 if (!s)
3015 return 0;
3016 return s - filename + 1;
3020 * This creates a temporary file in the same directory as the final
3021 * 'filename'
3023 * We want to avoid cross-directory filename renames, because those
3024 * can have problems on various filesystems (FAT, NFS, Coda).
3026 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3028 int fd, dirlen = directory_size(filename);
3030 if (dirlen + 20 > bufsiz) {
3031 errno = ENAMETOOLONG;
3032 return -1;
3034 memcpy(buffer, filename, dirlen);
3035 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3036 fd = git_mkstemp_mode(buffer, 0444);
3037 if (fd < 0 && dirlen && errno == ENOENT) {
3038 /* Make sure the directory exists */
3039 memcpy(buffer, filename, dirlen);
3040 buffer[dirlen-1] = 0;
3041 if (mkdir(buffer, 0777) && errno != EEXIST)
3042 return -1;
3043 if (adjust_shared_perm(buffer))
3044 return -1;
3046 /* Try again */
3047 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3048 fd = git_mkstemp_mode(buffer, 0444);
3050 return fd;
3053 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3054 const void *buf, unsigned long len, time_t mtime)
3056 int fd, ret;
3057 unsigned char compressed[4096];
3058 git_zstream stream;
3059 git_SHA_CTX c;
3060 unsigned char parano_sha1[20];
3061 static char tmp_file[PATH_MAX];
3062 const char *filename = sha1_file_name(sha1);
3064 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3065 if (fd < 0) {
3066 if (errno == EACCES)
3067 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3068 else
3069 return error("unable to create temporary file: %s", strerror(errno));
3072 /* Set it up */
3073 git_deflate_init(&stream, zlib_compression_level);
3074 stream.next_out = compressed;
3075 stream.avail_out = sizeof(compressed);
3076 git_SHA1_Init(&c);
3078 /* First header.. */
3079 stream.next_in = (unsigned char *)hdr;
3080 stream.avail_in = hdrlen;
3081 while (git_deflate(&stream, 0) == Z_OK)
3082 ; /* nothing */
3083 git_SHA1_Update(&c, hdr, hdrlen);
3085 /* Then the data itself.. */
3086 stream.next_in = (void *)buf;
3087 stream.avail_in = len;
3088 do {
3089 unsigned char *in0 = stream.next_in;
3090 ret = git_deflate(&stream, Z_FINISH);
3091 git_SHA1_Update(&c, in0, stream.next_in - in0);
3092 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3093 die("unable to write sha1 file");
3094 stream.next_out = compressed;
3095 stream.avail_out = sizeof(compressed);
3096 } while (ret == Z_OK);
3098 if (ret != Z_STREAM_END)
3099 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3100 ret = git_deflate_end_gently(&stream);
3101 if (ret != Z_OK)
3102 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3103 git_SHA1_Final(parano_sha1, &c);
3104 if (hashcmp(sha1, parano_sha1) != 0)
3105 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3107 close_sha1_file(fd);
3109 if (mtime) {
3110 struct utimbuf utb;
3111 utb.actime = mtime;
3112 utb.modtime = mtime;
3113 if (utime(tmp_file, &utb) < 0)
3114 warning("failed utime() on %s: %s",
3115 tmp_file, strerror(errno));
3118 return finalize_object_file(tmp_file, filename);
3121 static int freshen_loose_object(const unsigned char *sha1)
3123 return check_and_freshen(sha1, 1);
3126 static int freshen_packed_object(const unsigned char *sha1)
3128 struct pack_entry e;
3129 if (!find_pack_entry(sha1, &e))
3130 return 0;
3131 if (e.p->freshened)
3132 return 1;
3133 if (!freshen_file(e.p->pack_name))
3134 return 0;
3135 e.p->freshened = 1;
3136 return 1;
3139 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3141 char hdr[32];
3142 int hdrlen;
3144 /* Normally if we have it in the pack then we do not bother writing
3145 * it out into .git/objects/??/?{38} file.
3147 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3148 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3149 return 0;
3150 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3153 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3154 unsigned char *sha1, unsigned flags)
3156 char *header;
3157 int hdrlen, status = 0;
3159 /* type string, SP, %lu of the length plus NUL must fit this */
3160 header = xmalloc(strlen(type) + 32);
3161 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3163 if (!(flags & HASH_WRITE_OBJECT))
3164 goto cleanup;
3165 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3166 goto cleanup;
3167 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3169 cleanup:
3170 free(header);
3171 return status;
3174 int force_object_loose(const unsigned char *sha1, time_t mtime)
3176 void *buf;
3177 unsigned long len;
3178 enum object_type type;
3179 char hdr[32];
3180 int hdrlen;
3181 int ret;
3183 if (has_loose_object(sha1))
3184 return 0;
3185 buf = read_packed_sha1(sha1, &type, &len);
3186 if (!buf)
3187 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3188 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3189 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3190 free(buf);
3192 return ret;
3195 int has_pack_index(const unsigned char *sha1)
3197 struct stat st;
3198 if (stat(sha1_pack_index_name(sha1), &st))
3199 return 0;
3200 return 1;
3203 int has_sha1_pack(const unsigned char *sha1)
3205 struct pack_entry e;
3206 return find_pack_entry(sha1, &e);
3209 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3211 struct pack_entry e;
3213 if (find_pack_entry(sha1, &e))
3214 return 1;
3215 if (has_loose_object(sha1))
3216 return 1;
3217 if (flags & HAS_SHA1_QUICK)
3218 return 0;
3219 reprepare_packed_git();
3220 return find_pack_entry(sha1, &e);
3223 static void check_tree(const void *buf, size_t size)
3225 struct tree_desc desc;
3226 struct name_entry entry;
3228 init_tree_desc(&desc, buf, size);
3229 while (tree_entry(&desc, &entry))
3230 /* do nothing
3231 * tree_entry() will die() on malformed entries */
3235 static void check_commit(const void *buf, size_t size)
3237 struct commit c;
3238 memset(&c, 0, sizeof(c));
3239 if (parse_commit_buffer(&c, buf, size))
3240 die("corrupt commit");
3243 static void check_tag(const void *buf, size_t size)
3245 struct tag t;
3246 memset(&t, 0, sizeof(t));
3247 if (parse_tag_buffer(&t, buf, size))
3248 die("corrupt tag");
3251 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3252 enum object_type type,
3253 const char *path, unsigned flags)
3255 int ret, re_allocated = 0;
3256 int write_object = flags & HASH_WRITE_OBJECT;
3258 if (!type)
3259 type = OBJ_BLOB;
3262 * Convert blobs to git internal format
3264 if ((type == OBJ_BLOB) && path) {
3265 struct strbuf nbuf = STRBUF_INIT;
3266 if (convert_to_git(path, buf, size, &nbuf,
3267 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3268 buf = strbuf_detach(&nbuf, &size);
3269 re_allocated = 1;
3272 if (flags & HASH_FORMAT_CHECK) {
3273 if (type == OBJ_TREE)
3274 check_tree(buf, size);
3275 if (type == OBJ_COMMIT)
3276 check_commit(buf, size);
3277 if (type == OBJ_TAG)
3278 check_tag(buf, size);
3281 if (write_object)
3282 ret = write_sha1_file(buf, size, typename(type), sha1);
3283 else
3284 ret = hash_sha1_file(buf, size, typename(type), sha1);
3285 if (re_allocated)
3286 free(buf);
3287 return ret;
3290 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3291 const char *path, unsigned flags)
3293 int ret;
3294 const int write_object = flags & HASH_WRITE_OBJECT;
3295 struct strbuf sbuf = STRBUF_INIT;
3297 assert(path);
3298 assert(would_convert_to_git_filter_fd(path));
3300 convert_to_git_filter_fd(path, fd, &sbuf,
3301 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3303 if (write_object)
3304 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3305 sha1);
3306 else
3307 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3308 sha1);
3309 strbuf_release(&sbuf);
3310 return ret;
3313 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3314 const char *path, unsigned flags)
3316 struct strbuf sbuf = STRBUF_INIT;
3317 int ret;
3319 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3320 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3321 else
3322 ret = -1;
3323 strbuf_release(&sbuf);
3324 return ret;
3327 #define SMALL_FILE_SIZE (32*1024)
3329 static int index_core(unsigned char *sha1, int fd, size_t size,
3330 enum object_type type, const char *path,
3331 unsigned flags)
3333 int ret;
3335 if (!size) {
3336 ret = index_mem(sha1, "", size, type, path, flags);
3337 } else if (size <= SMALL_FILE_SIZE) {
3338 char *buf = xmalloc(size);
3339 if (size == read_in_full(fd, buf, size))
3340 ret = index_mem(sha1, buf, size, type, path, flags);
3341 else
3342 ret = error("short read %s", strerror(errno));
3343 free(buf);
3344 } else {
3345 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3346 ret = index_mem(sha1, buf, size, type, path, flags);
3347 munmap(buf, size);
3349 return ret;
3353 * This creates one packfile per large blob unless bulk-checkin
3354 * machinery is "plugged".
3356 * This also bypasses the usual "convert-to-git" dance, and that is on
3357 * purpose. We could write a streaming version of the converting
3358 * functions and insert that before feeding the data to fast-import
3359 * (or equivalent in-core API described above). However, that is
3360 * somewhat complicated, as we do not know the size of the filter
3361 * result, which we need to know beforehand when writing a git object.
3362 * Since the primary motivation for trying to stream from the working
3363 * tree file and to avoid mmaping it in core is to deal with large
3364 * binary blobs, they generally do not want to get any conversion, and
3365 * callers should avoid this code path when filters are requested.
3367 static int index_stream(unsigned char *sha1, int fd, size_t size,
3368 enum object_type type, const char *path,
3369 unsigned flags)
3371 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3374 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3375 enum object_type type, const char *path, unsigned flags)
3377 int ret;
3380 * Call xsize_t() only when needed to avoid potentially unnecessary
3381 * die() for large files.
3383 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3384 ret = index_stream_convert_blob(sha1, fd, path, flags);
3385 else if (!S_ISREG(st->st_mode))
3386 ret = index_pipe(sha1, fd, type, path, flags);
3387 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3388 (path && would_convert_to_git(path)))
3389 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3390 flags);
3391 else
3392 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3393 flags);
3394 close(fd);
3395 return ret;
3398 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3400 int fd;
3401 struct strbuf sb = STRBUF_INIT;
3403 switch (st->st_mode & S_IFMT) {
3404 case S_IFREG:
3405 fd = open(path, O_RDONLY);
3406 if (fd < 0)
3407 return error("open(\"%s\"): %s", path,
3408 strerror(errno));
3409 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3410 return error("%s: failed to insert into database",
3411 path);
3412 break;
3413 case S_IFLNK:
3414 if (strbuf_readlink(&sb, path, st->st_size)) {
3415 char *errstr = strerror(errno);
3416 return error("readlink(\"%s\"): %s", path,
3417 errstr);
3419 if (!(flags & HASH_WRITE_OBJECT))
3420 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3421 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3422 return error("%s: failed to insert into database",
3423 path);
3424 strbuf_release(&sb);
3425 break;
3426 case S_IFDIR:
3427 return resolve_gitlink_ref(path, "HEAD", sha1);
3428 default:
3429 return error("%s: unsupported file type", path);
3431 return 0;
3434 int read_pack_header(int fd, struct pack_header *header)
3436 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3437 /* "eof before pack header was fully read" */
3438 return PH_ERROR_EOF;
3440 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3441 /* "protocol error (pack signature mismatch detected)" */
3442 return PH_ERROR_PACK_SIGNATURE;
3443 if (!pack_version_ok(header->hdr_version))
3444 /* "protocol error (pack version unsupported)" */
3445 return PH_ERROR_PROTOCOL;
3446 return 0;
3449 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3451 enum object_type type = sha1_object_info(sha1, NULL);
3452 if (type < 0)
3453 die("%s is not a valid object", sha1_to_hex(sha1));
3454 if (type != expect)
3455 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3456 typename(expect));
3459 static int for_each_file_in_obj_subdir(int subdir_nr,
3460 struct strbuf *path,
3461 each_loose_object_fn obj_cb,
3462 each_loose_cruft_fn cruft_cb,
3463 each_loose_subdir_fn subdir_cb,
3464 void *data)
3466 size_t baselen = path->len;
3467 DIR *dir = opendir(path->buf);
3468 struct dirent *de;
3469 int r = 0;
3471 if (!dir) {
3472 if (errno == ENOENT)
3473 return 0;
3474 return error("unable to open %s: %s", path->buf, strerror(errno));
3477 while ((de = readdir(dir))) {
3478 if (is_dot_or_dotdot(de->d_name))
3479 continue;
3481 strbuf_setlen(path, baselen);
3482 strbuf_addf(path, "/%s", de->d_name);
3484 if (strlen(de->d_name) == 38) {
3485 char hex[41];
3486 unsigned char sha1[20];
3488 snprintf(hex, sizeof(hex), "%02x%s",
3489 subdir_nr, de->d_name);
3490 if (!get_sha1_hex(hex, sha1)) {
3491 if (obj_cb) {
3492 r = obj_cb(sha1, path->buf, data);
3493 if (r)
3494 break;
3496 continue;
3500 if (cruft_cb) {
3501 r = cruft_cb(de->d_name, path->buf, data);
3502 if (r)
3503 break;
3506 strbuf_setlen(path, baselen);
3508 if (!r && subdir_cb)
3509 r = subdir_cb(subdir_nr, path->buf, data);
3511 closedir(dir);
3512 return r;
3515 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3516 each_loose_object_fn obj_cb,
3517 each_loose_cruft_fn cruft_cb,
3518 each_loose_subdir_fn subdir_cb,
3519 void *data)
3521 size_t baselen = path->len;
3522 int r = 0;
3523 int i;
3525 for (i = 0; i < 256; i++) {
3526 strbuf_addf(path, "/%02x", i);
3527 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3528 subdir_cb, data);
3529 strbuf_setlen(path, baselen);
3530 if (r)
3531 break;
3534 return r;
3537 int for_each_loose_file_in_objdir(const char *path,
3538 each_loose_object_fn obj_cb,
3539 each_loose_cruft_fn cruft_cb,
3540 each_loose_subdir_fn subdir_cb,
3541 void *data)
3543 struct strbuf buf = STRBUF_INIT;
3544 int r;
3546 strbuf_addstr(&buf, path);
3547 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3548 subdir_cb, data);
3549 strbuf_release(&buf);
3551 return r;
3554 struct loose_alt_odb_data {
3555 each_loose_object_fn *cb;
3556 void *data;
3559 static int loose_from_alt_odb(struct alternate_object_database *alt,
3560 void *vdata)
3562 struct loose_alt_odb_data *data = vdata;
3563 struct strbuf buf = STRBUF_INIT;
3564 int r;
3566 /* copy base not including trailing '/' */
3567 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3568 r = for_each_loose_file_in_objdir_buf(&buf,
3569 data->cb, NULL, NULL,
3570 data->data);
3571 strbuf_release(&buf);
3572 return r;
3575 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3577 struct loose_alt_odb_data alt;
3578 int r;
3580 r = for_each_loose_file_in_objdir(get_object_directory(),
3581 cb, NULL, NULL, data);
3582 if (r)
3583 return r;
3585 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3586 return 0;
3588 alt.cb = cb;
3589 alt.data = data;
3590 return foreach_alt_odb(loose_from_alt_odb, &alt);
3593 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3595 uint32_t i;
3596 int r = 0;
3598 for (i = 0; i < p->num_objects; i++) {
3599 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3601 if (!sha1)
3602 return error("unable to get sha1 of object %u in %s",
3603 i, p->pack_name);
3605 r = cb(sha1, p, i, data);
3606 if (r)
3607 break;
3609 return r;
3612 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3614 struct packed_git *p;
3615 int r = 0;
3616 int pack_errors = 0;
3618 prepare_packed_git();
3619 for (p = packed_git; p; p = p->next) {
3620 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3621 continue;
3622 if (open_pack_index(p)) {
3623 pack_errors = 1;
3624 continue;
3626 r = for_each_object_in_pack(p, cb, data);
3627 if (r)
3628 break;
3630 return r ? r : pack_errors;