Merge branch 'master' of git://git.kernel.org/pub/scm/git/git
[git/jnareb-git.git] / sha1_file.c
blob0ed23981b363a07c6f4c5b930b1a5991d5c8a622
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 "delta.h"
12 #include "pack.h"
13 #include "blob.h"
14 #include "commit.h"
15 #include "run-command.h"
16 #include "tag.h"
17 #include "tree.h"
18 #include "tree-walk.h"
19 #include "refs.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
24 #include "dir.h"
26 #ifndef O_NOATIME
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
29 #else
30 #define O_NOATIME 0
31 #endif
32 #endif
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s) { return s; }
37 const unsigned char null_sha1[20];
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
45 static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50 } *cached_objects;
51 static int cached_object_nr, cached_object_alloc;
53 static struct cached_object empty_tree = {
54 EMPTY_TREE_SHA1_BIN_LITERAL,
55 OBJ_TREE,
56 "",
60 static struct packed_git *last_found_pack;
62 static struct cached_object *find_cached_object(const unsigned char *sha1)
64 int i;
65 struct cached_object *co = cached_objects;
67 for (i = 0; i < cached_object_nr; i++, co++) {
68 if (!hashcmp(co->sha1, sha1))
69 return co;
71 if (!hashcmp(sha1, empty_tree.sha1))
72 return &empty_tree;
73 return NULL;
76 int mkdir_in_gitdir(const char *path)
78 if (mkdir(path, 0777)) {
79 int saved_errno = errno;
80 struct stat st;
81 struct strbuf sb = STRBUF_INIT;
83 if (errno != EEXIST)
84 return -1;
86 * Are we looking at a path in a symlinked worktree
87 * whose original repository does not yet have it?
88 * e.g. .git/rr-cache pointing at its original
89 * repository in which the user hasn't performed any
90 * conflict resolution yet?
92 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
93 strbuf_readlink(&sb, path, st.st_size) ||
94 !is_absolute_path(sb.buf) ||
95 mkdir(sb.buf, 0777)) {
96 strbuf_release(&sb);
97 errno = saved_errno;
98 return -1;
100 strbuf_release(&sb);
102 return adjust_shared_perm(path);
105 int safe_create_leading_directories(char *path)
107 char *pos = path + offset_1st_component(path);
108 struct stat st;
110 while (pos) {
111 pos = strchr(pos, '/');
112 if (!pos)
113 break;
114 while (*++pos == '/')
116 if (!*pos)
117 break;
118 *--pos = '\0';
119 if (!stat(path, &st)) {
120 /* path exists */
121 if (!S_ISDIR(st.st_mode)) {
122 *pos = '/';
123 return -3;
126 else if (mkdir(path, 0777)) {
127 if (errno == EEXIST &&
128 !stat(path, &st) && S_ISDIR(st.st_mode)) {
129 ; /* somebody created it since we checked */
130 } else {
131 *pos = '/';
132 return -1;
135 else if (adjust_shared_perm(path)) {
136 *pos = '/';
137 return -2;
139 *pos++ = '/';
141 return 0;
144 int safe_create_leading_directories_const(const char *path)
146 /* path points to cache entries, so xstrdup before messing with it */
147 char *buf = xstrdup(path);
148 int result = safe_create_leading_directories(buf);
149 free(buf);
150 return result;
153 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155 int i;
156 for (i = 0; i < 20; i++) {
157 static char hex[] = "0123456789abcdef";
158 unsigned int val = sha1[i];
159 char *pos = pathbuf + i*2 + (i > 0);
160 *pos++ = hex[val >> 4];
161 *pos = hex[val & 0xf];
166 * NOTE! This returns a statically allocated buffer, so you have to be
167 * careful about using it. Do an "xstrdup()" if you need to save the
168 * filename.
170 * Also note that this returns the location for creating. Reading
171 * SHA1 file can happen from any alternate directory listed in the
172 * DB_ENVIRONMENT environment variable if it is not found in
173 * the primary object database.
175 char *sha1_file_name(const unsigned char *sha1)
177 static char buf[PATH_MAX];
178 const char *objdir;
179 int len;
181 objdir = get_object_directory();
182 len = strlen(objdir);
184 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
185 if (len + 43 > PATH_MAX)
186 die("insanely long object directory %s", objdir);
187 memcpy(buf, objdir, len);
188 buf[len] = '/';
189 buf[len+3] = '/';
190 buf[len+42] = '\0';
191 fill_sha1_path(buf + len + 1, sha1);
192 return buf;
195 static char *sha1_get_pack_name(const unsigned char *sha1,
196 char **name, char **base, const char *which)
198 static const char hex[] = "0123456789abcdef";
199 char *buf;
200 int i;
202 if (!*base) {
203 const char *sha1_file_directory = get_object_directory();
204 int len = strlen(sha1_file_directory);
205 *base = xmalloc(len + 60);
206 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
207 sha1_file_directory, which);
208 *name = *base + len + 11;
211 buf = *name;
213 for (i = 0; i < 20; i++) {
214 unsigned int val = *sha1++;
215 *buf++ = hex[val >> 4];
216 *buf++ = hex[val & 0xf];
219 return *base;
222 char *sha1_pack_name(const unsigned char *sha1)
224 static char *name, *base;
226 return sha1_get_pack_name(sha1, &name, &base, "pack");
229 char *sha1_pack_index_name(const unsigned char *sha1)
231 static char *name, *base;
233 return sha1_get_pack_name(sha1, &name, &base, "idx");
236 struct alternate_object_database *alt_odb_list;
237 static struct alternate_object_database **alt_odb_tail;
239 static int git_open_noatime(const char *name);
242 * Prepare alternate object database registry.
244 * The variable alt_odb_list points at the list of struct
245 * alternate_object_database. The elements on this list come from
246 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
247 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
248 * whose contents is similar to that environment variable but can be
249 * LF separated. Its base points at a statically allocated buffer that
250 * contains "/the/directory/corresponding/to/.git/objects/...", while
251 * its name points just after the slash at the end of ".git/objects/"
252 * in the example above, and has enough space to hold 40-byte hex
253 * SHA1, an extra slash for the first level indirection, and the
254 * terminating NUL.
256 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
258 const char *objdir = get_object_directory();
259 struct alternate_object_database *ent;
260 struct alternate_object_database *alt;
261 int pfxlen, entlen;
262 struct strbuf pathbuf = STRBUF_INIT;
264 if (!is_absolute_path(entry) && relative_base) {
265 strbuf_addstr(&pathbuf, real_path(relative_base));
266 strbuf_addch(&pathbuf, '/');
268 strbuf_addstr(&pathbuf, entry);
270 normalize_path_copy(pathbuf.buf, pathbuf.buf);
272 pfxlen = strlen(pathbuf.buf);
275 * The trailing slash after the directory name is given by
276 * this function at the end. Remove duplicates.
278 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
279 pfxlen -= 1;
281 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
282 ent = xmalloc(sizeof(*ent) + entlen);
283 memcpy(ent->base, pathbuf.buf, pfxlen);
284 strbuf_release(&pathbuf);
286 ent->name = ent->base + pfxlen + 1;
287 ent->base[pfxlen + 3] = '/';
288 ent->base[pfxlen] = ent->base[entlen-1] = 0;
290 /* Detect cases where alternate disappeared */
291 if (!is_directory(ent->base)) {
292 error("object directory %s does not exist; "
293 "check .git/objects/info/alternates.",
294 ent->base);
295 free(ent);
296 return -1;
299 /* Prevent the common mistake of listing the same
300 * thing twice, or object directory itself.
302 for (alt = alt_odb_list; alt; alt = alt->next) {
303 if (!memcmp(ent->base, alt->base, pfxlen)) {
304 free(ent);
305 return -1;
308 if (!strcmp(ent->base, objdir)) {
309 free(ent);
310 return -1;
313 /* add the alternate entry */
314 *alt_odb_tail = ent;
315 alt_odb_tail = &(ent->next);
316 ent->next = NULL;
318 /* recursively add alternates */
319 read_info_alternates(ent->base, depth + 1);
321 ent->base[pfxlen] = '/';
323 return 0;
326 static void link_alt_odb_entries(const char *alt, int len, int sep,
327 const char *relative_base, int depth)
329 struct string_list entries = STRING_LIST_INIT_NODUP;
330 char *alt_copy;
331 int i;
333 if (depth > 5) {
334 error("%s: ignoring alternate object stores, nesting too deep.",
335 relative_base);
336 return;
339 alt_copy = xmemdupz(alt, len);
340 string_list_split_in_place(&entries, alt_copy, sep, -1);
341 for (i = 0; i < entries.nr; i++) {
342 const char *entry = entries.items[i].string;
343 if (entry[0] == '\0' || entry[0] == '#')
344 continue;
345 if (!is_absolute_path(entry) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, entry);
348 } else {
349 link_alt_odb_entry(entry, relative_base, depth);
352 string_list_clear(&entries, 0);
353 free(alt_copy);
356 void read_info_alternates(const char * relative_base, int depth)
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
366 int fd;
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = git_open_noatime(path);
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
374 return;
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
378 close(fd);
380 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
382 munmap(map, mapsz);
385 void add_to_alternates_file(const char *reference)
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
389 char *alt = mkpath("%s\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
394 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
397 void foreach_alt_odb(alt_odb_fn fn, void *cb)
399 struct alternate_object_database *ent;
401 prepare_alt_odb();
402 for (ent = alt_odb_list; ent; ent = ent->next)
403 if (fn(ent, cb))
404 return;
407 void prepare_alt_odb(void)
409 const char *alt;
411 if (alt_odb_tail)
412 return;
414 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
415 if (!alt) alt = "";
417 alt_odb_tail = &alt_odb_list;
418 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
420 read_info_alternates(get_object_directory(), 0);
423 static int has_loose_object_local(const unsigned char *sha1)
425 char *name = sha1_file_name(sha1);
426 return !access(name, F_OK);
429 int has_loose_object_nonlocal(const unsigned char *sha1)
431 struct alternate_object_database *alt;
432 prepare_alt_odb();
433 for (alt = alt_odb_list; alt; alt = alt->next) {
434 fill_sha1_path(alt->name, sha1);
435 if (!access(alt->base, F_OK))
436 return 1;
438 return 0;
441 static int has_loose_object(const unsigned char *sha1)
443 return has_loose_object_local(sha1) ||
444 has_loose_object_nonlocal(sha1);
447 static unsigned int pack_used_ctr;
448 static unsigned int pack_mmap_calls;
449 static unsigned int peak_pack_open_windows;
450 static unsigned int pack_open_windows;
451 static unsigned int pack_open_fds;
452 static unsigned int pack_max_fds;
453 static size_t peak_pack_mapped;
454 static size_t pack_mapped;
455 struct packed_git *packed_git;
457 void pack_report(void)
459 fprintf(stderr,
460 "pack_report: getpagesize() = %10" SZ_FMT "\n"
461 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
463 sz_fmt(getpagesize()),
464 sz_fmt(packed_git_window_size),
465 sz_fmt(packed_git_limit));
466 fprintf(stderr,
467 "pack_report: pack_used_ctr = %10u\n"
468 "pack_report: pack_mmap_calls = %10u\n"
469 "pack_report: pack_open_windows = %10u / %10u\n"
470 "pack_report: pack_mapped = "
471 "%10" SZ_FMT " / %10" SZ_FMT "\n",
472 pack_used_ctr,
473 pack_mmap_calls,
474 pack_open_windows, peak_pack_open_windows,
475 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
478 static int check_packed_git_idx(const char *path, struct packed_git *p)
480 void *idx_map;
481 struct pack_idx_header *hdr;
482 size_t idx_size;
483 uint32_t version, nr, i, *index;
484 int fd = git_open_noatime(path);
485 struct stat st;
487 if (fd < 0)
488 return -1;
489 if (fstat(fd, &st)) {
490 close(fd);
491 return -1;
493 idx_size = xsize_t(st.st_size);
494 if (idx_size < 4 * 256 + 20 + 20) {
495 close(fd);
496 return error("index file %s is too small", path);
498 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
499 close(fd);
501 hdr = idx_map;
502 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
503 version = ntohl(hdr->idx_version);
504 if (version < 2 || version > 2) {
505 munmap(idx_map, idx_size);
506 return error("index file %s is version %"PRIu32
507 " and is not supported by this binary"
508 " (try upgrading GIT to a newer version)",
509 path, version);
511 } else
512 version = 1;
514 nr = 0;
515 index = idx_map;
516 if (version > 1)
517 index += 2; /* skip index header */
518 for (i = 0; i < 256; i++) {
519 uint32_t n = ntohl(index[i]);
520 if (n < nr) {
521 munmap(idx_map, idx_size);
522 return error("non-monotonic index %s", path);
524 nr = n;
527 if (version == 1) {
529 * Total size:
530 * - 256 index entries 4 bytes each
531 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
532 * - 20-byte SHA1 of the packfile
533 * - 20-byte SHA1 file checksum
535 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
536 munmap(idx_map, idx_size);
537 return error("wrong index v1 file size in %s", path);
539 } else if (version == 2) {
541 * Minimum size:
542 * - 8 bytes of header
543 * - 256 index entries 4 bytes each
544 * - 20-byte sha1 entry * nr
545 * - 4-byte crc entry * nr
546 * - 4-byte offset entry * nr
547 * - 20-byte SHA1 of the packfile
548 * - 20-byte SHA1 file checksum
549 * And after the 4-byte offset table might be a
550 * variable sized table containing 8-byte entries
551 * for offsets larger than 2^31.
553 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
554 unsigned long max_size = min_size;
555 if (nr)
556 max_size += (nr - 1)*8;
557 if (idx_size < min_size || idx_size > max_size) {
558 munmap(idx_map, idx_size);
559 return error("wrong index v2 file size in %s", path);
561 if (idx_size != min_size &&
563 * make sure we can deal with large pack offsets.
564 * 31-bit signed offset won't be enough, neither
565 * 32-bit unsigned one will be.
567 (sizeof(off_t) <= 4)) {
568 munmap(idx_map, idx_size);
569 return error("pack too large for current definition of off_t in %s", path);
573 p->index_version = version;
574 p->index_data = idx_map;
575 p->index_size = idx_size;
576 p->num_objects = nr;
577 return 0;
580 int open_pack_index(struct packed_git *p)
582 char *idx_name;
583 int ret;
585 if (p->index_data)
586 return 0;
588 idx_name = xstrdup(p->pack_name);
589 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
590 ret = check_packed_git_idx(idx_name, p);
591 free(idx_name);
592 return ret;
595 static void scan_windows(struct packed_git *p,
596 struct packed_git **lru_p,
597 struct pack_window **lru_w,
598 struct pack_window **lru_l)
600 struct pack_window *w, *w_l;
602 for (w_l = NULL, w = p->windows; w; w = w->next) {
603 if (!w->inuse_cnt) {
604 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
605 *lru_p = p;
606 *lru_w = w;
607 *lru_l = w_l;
610 w_l = w;
614 static int unuse_one_window(struct packed_git *current, int keep_fd)
616 struct packed_git *p, *lru_p = NULL;
617 struct pack_window *lru_w = NULL, *lru_l = NULL;
619 if (current)
620 scan_windows(current, &lru_p, &lru_w, &lru_l);
621 for (p = packed_git; p; p = p->next)
622 scan_windows(p, &lru_p, &lru_w, &lru_l);
623 if (lru_p) {
624 munmap(lru_w->base, lru_w->len);
625 pack_mapped -= lru_w->len;
626 if (lru_l)
627 lru_l->next = lru_w->next;
628 else {
629 lru_p->windows = lru_w->next;
630 if (!lru_p->windows && lru_p->pack_fd != -1
631 && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 pack_open_fds--;
634 lru_p->pack_fd = -1;
637 free(lru_w);
638 pack_open_windows--;
639 return 1;
641 return 0;
644 void release_pack_memory(size_t need, int fd)
646 size_t cur = pack_mapped;
647 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
648 ; /* nothing */
651 void *xmmap(void *start, size_t length,
652 int prot, int flags, int fd, off_t offset)
654 void *ret = mmap(start, length, prot, flags, fd, offset);
655 if (ret == MAP_FAILED) {
656 if (!length)
657 return NULL;
658 release_pack_memory(length, fd);
659 ret = mmap(start, length, prot, flags, fd, offset);
660 if (ret == MAP_FAILED)
661 die_errno("Out of memory? mmap failed");
663 return ret;
666 void close_pack_windows(struct packed_git *p)
668 while (p->windows) {
669 struct pack_window *w = p->windows;
671 if (w->inuse_cnt)
672 die("pack '%s' still has open windows to it",
673 p->pack_name);
674 munmap(w->base, w->len);
675 pack_mapped -= w->len;
676 pack_open_windows--;
677 p->windows = w->next;
678 free(w);
682 void unuse_pack(struct pack_window **w_cursor)
684 struct pack_window *w = *w_cursor;
685 if (w) {
686 w->inuse_cnt--;
687 *w_cursor = NULL;
691 void close_pack_index(struct packed_git *p)
693 if (p->index_data) {
694 munmap((void *)p->index_data, p->index_size);
695 p->index_data = NULL;
700 * This is used by git-repack in case a newly created pack happens to
701 * contain the same set of objects as an existing one. In that case
702 * the resulting file might be different even if its name would be the
703 * same. It is best to close any reference to the old pack before it is
704 * replaced on disk. Of course no index pointers nor windows for given pack
705 * must subsist at this point. If ever objects from this pack are requested
706 * again, the new version of the pack will be reinitialized through
707 * reprepare_packed_git().
709 void free_pack_by_name(const char *pack_name)
711 struct packed_git *p, **pp = &packed_git;
713 while (*pp) {
714 p = *pp;
715 if (strcmp(pack_name, p->pack_name) == 0) {
716 clear_delta_base_cache();
717 close_pack_windows(p);
718 if (p->pack_fd != -1) {
719 close(p->pack_fd);
720 pack_open_fds--;
722 close_pack_index(p);
723 free(p->bad_object_sha1);
724 *pp = p->next;
725 if (last_found_pack == p)
726 last_found_pack = NULL;
727 free(p);
728 return;
730 pp = &p->next;
734 static unsigned int get_max_fd_limit(void)
736 #ifdef RLIMIT_NOFILE
737 struct rlimit lim;
739 if (getrlimit(RLIMIT_NOFILE, &lim))
740 die_errno("cannot get RLIMIT_NOFILE");
742 return lim.rlim_cur;
743 #elif defined(_SC_OPEN_MAX)
744 return sysconf(_SC_OPEN_MAX);
745 #elif defined(OPEN_MAX)
746 return OPEN_MAX;
747 #else
748 return 1; /* see the caller ;-) */
749 #endif
753 * Do not call this directly as this leaks p->pack_fd on error return;
754 * call open_packed_git() instead.
756 static int open_packed_git_1(struct packed_git *p)
758 struct stat st;
759 struct pack_header hdr;
760 unsigned char sha1[20];
761 unsigned char *idx_sha1;
762 long fd_flag;
764 if (!p->index_data && open_pack_index(p))
765 return error("packfile %s index unavailable", p->pack_name);
767 if (!pack_max_fds) {
768 unsigned int max_fds = get_max_fd_limit();
770 /* Save 3 for stdin/stdout/stderr, 22 for work */
771 if (25 < max_fds)
772 pack_max_fds = max_fds - 25;
773 else
774 pack_max_fds = 1;
777 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
778 ; /* nothing */
780 p->pack_fd = git_open_noatime(p->pack_name);
781 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
782 return -1;
783 pack_open_fds++;
785 /* If we created the struct before we had the pack we lack size. */
786 if (!p->pack_size) {
787 if (!S_ISREG(st.st_mode))
788 return error("packfile %s not a regular file", p->pack_name);
789 p->pack_size = st.st_size;
790 } else if (p->pack_size != st.st_size)
791 return error("packfile %s size changed", p->pack_name);
793 /* We leave these file descriptors open with sliding mmap;
794 * there is no point keeping them open across exec(), though.
796 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
797 if (fd_flag < 0)
798 return error("cannot determine file descriptor flags");
799 fd_flag |= FD_CLOEXEC;
800 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
801 return error("cannot set FD_CLOEXEC");
803 /* Verify we recognize this pack file format. */
804 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
805 return error("file %s is far too short to be a packfile", p->pack_name);
806 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
807 return error("file %s is not a GIT packfile", p->pack_name);
808 if (!pack_version_ok(hdr.hdr_version))
809 return error("packfile %s is version %"PRIu32" and not"
810 " supported (try upgrading GIT to a newer version)",
811 p->pack_name, ntohl(hdr.hdr_version));
813 /* Verify the pack matches its index. */
814 if (p->num_objects != ntohl(hdr.hdr_entries))
815 return error("packfile %s claims to have %"PRIu32" objects"
816 " while index indicates %"PRIu32" objects",
817 p->pack_name, ntohl(hdr.hdr_entries),
818 p->num_objects);
819 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
820 return error("end of packfile %s is unavailable", p->pack_name);
821 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
822 return error("packfile %s signature is unavailable", p->pack_name);
823 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
824 if (hashcmp(sha1, idx_sha1))
825 return error("packfile %s does not match index", p->pack_name);
826 return 0;
829 static int open_packed_git(struct packed_git *p)
831 if (!open_packed_git_1(p))
832 return 0;
833 if (p->pack_fd != -1) {
834 close(p->pack_fd);
835 pack_open_fds--;
836 p->pack_fd = -1;
838 return -1;
841 static int in_window(struct pack_window *win, off_t offset)
843 /* We must promise at least 20 bytes (one hash) after the
844 * offset is available from this window, otherwise the offset
845 * is not actually in this window and a different window (which
846 * has that one hash excess) must be used. This is to support
847 * the object header and delta base parsing routines below.
849 off_t win_off = win->offset;
850 return win_off <= offset
851 && (offset + 20) <= (win_off + win->len);
854 unsigned char *use_pack(struct packed_git *p,
855 struct pack_window **w_cursor,
856 off_t offset,
857 unsigned long *left)
859 struct pack_window *win = *w_cursor;
861 /* Since packfiles end in a hash of their content and it's
862 * pointless to ask for an offset into the middle of that
863 * hash, and the in_window function above wouldn't match
864 * don't allow an offset too close to the end of the file.
866 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
867 die("packfile %s cannot be accessed", p->pack_name);
868 if (offset > (p->pack_size - 20))
869 die("offset beyond end of packfile (truncated pack?)");
871 if (!win || !in_window(win, offset)) {
872 if (win)
873 win->inuse_cnt--;
874 for (win = p->windows; win; win = win->next) {
875 if (in_window(win, offset))
876 break;
878 if (!win) {
879 size_t window_align = packed_git_window_size / 2;
880 off_t len;
882 if (p->pack_fd == -1 && open_packed_git(p))
883 die("packfile %s cannot be accessed", p->pack_name);
885 win = xcalloc(1, sizeof(*win));
886 win->offset = (offset / window_align) * window_align;
887 len = p->pack_size - win->offset;
888 if (len > packed_git_window_size)
889 len = packed_git_window_size;
890 win->len = (size_t)len;
891 pack_mapped += win->len;
892 while (packed_git_limit < pack_mapped
893 && unuse_one_window(p, p->pack_fd))
894 ; /* nothing */
895 win->base = xmmap(NULL, win->len,
896 PROT_READ, MAP_PRIVATE,
897 p->pack_fd, win->offset);
898 if (win->base == MAP_FAILED)
899 die("packfile %s cannot be mapped: %s",
900 p->pack_name,
901 strerror(errno));
902 if (!win->offset && win->len == p->pack_size
903 && !p->do_not_close) {
904 close(p->pack_fd);
905 pack_open_fds--;
906 p->pack_fd = -1;
908 pack_mmap_calls++;
909 pack_open_windows++;
910 if (pack_mapped > peak_pack_mapped)
911 peak_pack_mapped = pack_mapped;
912 if (pack_open_windows > peak_pack_open_windows)
913 peak_pack_open_windows = pack_open_windows;
914 win->next = p->windows;
915 p->windows = win;
918 if (win != *w_cursor) {
919 win->last_used = pack_used_ctr++;
920 win->inuse_cnt++;
921 *w_cursor = win;
923 offset -= win->offset;
924 if (left)
925 *left = win->len - xsize_t(offset);
926 return win->base + offset;
929 static struct packed_git *alloc_packed_git(int extra)
931 struct packed_git *p = xmalloc(sizeof(*p) + extra);
932 memset(p, 0, sizeof(*p));
933 p->pack_fd = -1;
934 return p;
937 static void try_to_free_pack_memory(size_t size)
939 release_pack_memory(size, -1);
942 struct packed_git *add_packed_git(const char *path, int path_len, int local)
944 static int have_set_try_to_free_routine;
945 struct stat st;
946 struct packed_git *p = alloc_packed_git(path_len + 2);
948 if (!have_set_try_to_free_routine) {
949 have_set_try_to_free_routine = 1;
950 set_try_to_free_routine(try_to_free_pack_memory);
954 * Make sure a corresponding .pack file exists and that
955 * the index looks sane.
957 path_len -= strlen(".idx");
958 if (path_len < 1) {
959 free(p);
960 return NULL;
962 memcpy(p->pack_name, path, path_len);
964 strcpy(p->pack_name + path_len, ".keep");
965 if (!access(p->pack_name, F_OK))
966 p->pack_keep = 1;
968 strcpy(p->pack_name + path_len, ".pack");
969 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
970 free(p);
971 return NULL;
974 /* ok, it looks sane as far as we can check without
975 * actually mapping the pack file.
977 p->pack_size = st.st_size;
978 p->pack_local = local;
979 p->mtime = st.st_mtime;
980 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
981 hashclr(p->sha1);
982 return p;
985 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
987 const char *path = sha1_pack_name(sha1);
988 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
990 strcpy(p->pack_name, path);
991 hashcpy(p->sha1, sha1);
992 if (check_packed_git_idx(idx_path, p)) {
993 free(p);
994 return NULL;
997 return p;
1000 void install_packed_git(struct packed_git *pack)
1002 if (pack->pack_fd != -1)
1003 pack_open_fds++;
1005 pack->next = packed_git;
1006 packed_git = pack;
1009 void (*report_garbage)(const char *desc, const char *path);
1011 static void report_helper(const struct string_list *list,
1012 int seen_bits, int first, int last)
1014 const char *msg;
1015 switch (seen_bits) {
1016 case 0:
1017 msg = "no corresponding .idx nor .pack";
1018 break;
1019 case 1:
1020 msg = "no corresponding .idx";
1021 break;
1022 case 2:
1023 msg = "no corresponding .pack";
1024 break;
1025 default:
1026 return;
1028 for (; first < last; first++)
1029 report_garbage(msg, list->items[first].string);
1032 static void report_pack_garbage(struct string_list *list)
1034 int i, baselen = -1, first = 0, seen_bits = 0;
1036 if (!report_garbage)
1037 return;
1039 sort_string_list(list);
1041 for (i = 0; i < list->nr; i++) {
1042 const char *path = list->items[i].string;
1043 if (baselen != -1 &&
1044 strncmp(path, list->items[first].string, baselen)) {
1045 report_helper(list, seen_bits, first, i);
1046 baselen = -1;
1047 seen_bits = 0;
1049 if (baselen == -1) {
1050 const char *dot = strrchr(path, '.');
1051 if (!dot) {
1052 report_garbage("garbage found", path);
1053 continue;
1055 baselen = dot - path + 1;
1056 first = i;
1058 if (!strcmp(path + baselen, "pack"))
1059 seen_bits |= 1;
1060 else if (!strcmp(path + baselen, "idx"))
1061 seen_bits |= 2;
1063 report_helper(list, seen_bits, first, list->nr);
1066 static void prepare_packed_git_one(char *objdir, int local)
1068 /* Ensure that this buffer is large enough so that we can
1069 append "/pack/" without clobbering the stack even if
1070 strlen(objdir) were PATH_MAX. */
1071 char path[PATH_MAX + 1 + 4 + 1 + 1];
1072 int len;
1073 DIR *dir;
1074 struct dirent *de;
1075 struct string_list garbage = STRING_LIST_INIT_DUP;
1077 sprintf(path, "%s/pack", objdir);
1078 len = strlen(path);
1079 dir = opendir(path);
1080 if (!dir) {
1081 if (errno != ENOENT)
1082 error("unable to open object pack directory: %s: %s",
1083 path, strerror(errno));
1084 return;
1086 path[len++] = '/';
1087 while ((de = readdir(dir)) != NULL) {
1088 int namelen = strlen(de->d_name);
1089 struct packed_git *p;
1091 if (len + namelen + 1 > sizeof(path)) {
1092 if (report_garbage) {
1093 struct strbuf sb = STRBUF_INIT;
1094 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1095 report_garbage("path too long", sb.buf);
1096 strbuf_release(&sb);
1098 continue;
1101 if (is_dot_or_dotdot(de->d_name))
1102 continue;
1104 strcpy(path + len, de->d_name);
1106 if (has_extension(de->d_name, ".idx")) {
1107 /* Don't reopen a pack we already have. */
1108 for (p = packed_git; p; p = p->next) {
1109 if (!memcmp(path, p->pack_name, len + namelen - 4))
1110 break;
1112 if (p == NULL &&
1114 * See if it really is a valid .idx file with
1115 * corresponding .pack file that we can map.
1117 (p = add_packed_git(path, len + namelen, local)) != NULL)
1118 install_packed_git(p);
1121 if (!report_garbage)
1122 continue;
1124 if (has_extension(de->d_name, ".idx") ||
1125 has_extension(de->d_name, ".pack") ||
1126 has_extension(de->d_name, ".keep"))
1127 string_list_append(&garbage, path);
1128 else
1129 report_garbage("garbage found", path);
1131 closedir(dir);
1132 report_pack_garbage(&garbage);
1133 string_list_clear(&garbage, 0);
1136 static int sort_pack(const void *a_, const void *b_)
1138 struct packed_git *a = *((struct packed_git **)a_);
1139 struct packed_git *b = *((struct packed_git **)b_);
1140 int st;
1143 * Local packs tend to contain objects specific to our
1144 * variant of the project than remote ones. In addition,
1145 * remote ones could be on a network mounted filesystem.
1146 * Favor local ones for these reasons.
1148 st = a->pack_local - b->pack_local;
1149 if (st)
1150 return -st;
1153 * Younger packs tend to contain more recent objects,
1154 * and more recent objects tend to get accessed more
1155 * often.
1157 if (a->mtime < b->mtime)
1158 return 1;
1159 else if (a->mtime == b->mtime)
1160 return 0;
1161 return -1;
1164 static void rearrange_packed_git(void)
1166 struct packed_git **ary, *p;
1167 int i, n;
1169 for (n = 0, p = packed_git; p; p = p->next)
1170 n++;
1171 if (n < 2)
1172 return;
1174 /* prepare an array of packed_git for easier sorting */
1175 ary = xcalloc(n, sizeof(struct packed_git *));
1176 for (n = 0, p = packed_git; p; p = p->next)
1177 ary[n++] = p;
1179 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1181 /* link them back again */
1182 for (i = 0; i < n - 1; i++)
1183 ary[i]->next = ary[i + 1];
1184 ary[n - 1]->next = NULL;
1185 packed_git = ary[0];
1187 free(ary);
1190 static int prepare_packed_git_run_once = 0;
1191 void prepare_packed_git(void)
1193 struct alternate_object_database *alt;
1195 if (prepare_packed_git_run_once)
1196 return;
1197 prepare_packed_git_one(get_object_directory(), 1);
1198 prepare_alt_odb();
1199 for (alt = alt_odb_list; alt; alt = alt->next) {
1200 alt->name[-1] = 0;
1201 prepare_packed_git_one(alt->base, 0);
1202 alt->name[-1] = '/';
1204 rearrange_packed_git();
1205 prepare_packed_git_run_once = 1;
1208 void reprepare_packed_git(void)
1210 discard_revindex();
1211 prepare_packed_git_run_once = 0;
1212 prepare_packed_git();
1215 static void mark_bad_packed_object(struct packed_git *p,
1216 const unsigned char *sha1)
1218 unsigned i;
1219 for (i = 0; i < p->num_bad_objects; i++)
1220 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1221 return;
1222 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1223 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1224 p->num_bad_objects++;
1227 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1229 struct packed_git *p;
1230 unsigned i;
1232 for (p = packed_git; p; p = p->next)
1233 for (i = 0; i < p->num_bad_objects; i++)
1234 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1235 return p;
1236 return NULL;
1240 * With an in-core object data in "map", rehash it to make sure the
1241 * object name actually matches "sha1" to detect object corruption.
1242 * With "map" == NULL, try reading the object named with "sha1" using
1243 * the streaming interface and rehash it to do the same.
1245 int check_sha1_signature(const unsigned char *sha1, void *map,
1246 unsigned long size, const char *type)
1248 unsigned char real_sha1[20];
1249 enum object_type obj_type;
1250 struct git_istream *st;
1251 git_SHA_CTX c;
1252 char hdr[32];
1253 int hdrlen;
1255 if (map) {
1256 hash_sha1_file(map, size, type, real_sha1);
1257 return hashcmp(sha1, real_sha1) ? -1 : 0;
1260 st = open_istream(sha1, &obj_type, &size, NULL);
1261 if (!st)
1262 return -1;
1264 /* Generate the header */
1265 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1267 /* Sha1.. */
1268 git_SHA1_Init(&c);
1269 git_SHA1_Update(&c, hdr, hdrlen);
1270 for (;;) {
1271 char buf[1024 * 16];
1272 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1274 if (readlen < 0) {
1275 close_istream(st);
1276 return -1;
1278 if (!readlen)
1279 break;
1280 git_SHA1_Update(&c, buf, readlen);
1282 git_SHA1_Final(real_sha1, &c);
1283 close_istream(st);
1284 return hashcmp(sha1, real_sha1) ? -1 : 0;
1287 static int git_open_noatime(const char *name)
1289 static int sha1_file_open_flag = O_NOATIME;
1291 for (;;) {
1292 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1293 if (fd >= 0)
1294 return fd;
1296 /* Might the failure be due to O_NOATIME? */
1297 if (errno != ENOENT && sha1_file_open_flag) {
1298 sha1_file_open_flag = 0;
1299 continue;
1302 return -1;
1306 static int open_sha1_file(const unsigned char *sha1)
1308 int fd;
1309 char *name = sha1_file_name(sha1);
1310 struct alternate_object_database *alt;
1312 fd = git_open_noatime(name);
1313 if (fd >= 0)
1314 return fd;
1316 prepare_alt_odb();
1317 errno = ENOENT;
1318 for (alt = alt_odb_list; alt; alt = alt->next) {
1319 name = alt->name;
1320 fill_sha1_path(name, sha1);
1321 fd = git_open_noatime(alt->base);
1322 if (fd >= 0)
1323 return fd;
1325 return -1;
1328 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1330 void *map;
1331 int fd;
1333 fd = open_sha1_file(sha1);
1334 map = NULL;
1335 if (fd >= 0) {
1336 struct stat st;
1338 if (!fstat(fd, &st)) {
1339 *size = xsize_t(st.st_size);
1340 if (!*size) {
1341 /* mmap() is forbidden on empty files */
1342 error("object file %s is empty", sha1_file_name(sha1));
1343 return NULL;
1345 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1347 close(fd);
1349 return map;
1353 * There used to be a second loose object header format which
1354 * was meant to mimic the in-pack format, allowing for direct
1355 * copy of the object data. This format turned up not to be
1356 * really worth it and we no longer write loose objects in that
1357 * format.
1359 static int experimental_loose_object(unsigned char *map)
1361 unsigned int word;
1364 * We must determine if the buffer contains the standard
1365 * zlib-deflated stream or the experimental format based
1366 * on the in-pack object format. Compare the header byte
1367 * for each format:
1369 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1370 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1372 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1373 * in standard loose-object format, UNLESS it is a Git-pack
1374 * format object *exactly* 8 bytes in size when inflated.
1376 * However, RFC1950 also specifies that the 1st 16-bit word
1377 * must be divisible by 31 - this checksum tells us our buffer
1378 * is in the standard format, giving a false positive only if
1379 * the 1st word of the Git-pack format object happens to be
1380 * divisible by 31, ie:
1381 * ((byte0 * 256) + byte1) % 31 = 0
1382 * => 0ttt10000www1000 % 31 = 0
1384 * As it happens, this case can only arise for www=3 & ttt=1
1385 * - ie, a Commit object, which would have to be 8 bytes in
1386 * size. As no Commit can be that small, we find that the
1387 * combination of these two criteria (bitmask & checksum)
1388 * can always correctly determine the buffer format.
1390 word = (map[0] << 8) + map[1];
1391 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1392 return 0;
1393 else
1394 return 1;
1397 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1398 unsigned long len, enum object_type *type, unsigned long *sizep)
1400 unsigned shift;
1401 unsigned long size, c;
1402 unsigned long used = 0;
1404 c = buf[used++];
1405 *type = (c >> 4) & 7;
1406 size = c & 15;
1407 shift = 4;
1408 while (c & 0x80) {
1409 if (len <= used || bitsizeof(long) <= shift) {
1410 error("bad object header");
1411 size = used = 0;
1412 break;
1414 c = buf[used++];
1415 size += (c & 0x7f) << shift;
1416 shift += 7;
1418 *sizep = size;
1419 return used;
1422 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1424 unsigned long size, used;
1425 static const char valid_loose_object_type[8] = {
1426 0, /* OBJ_EXT */
1427 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1428 0, /* "delta" and others are invalid in a loose object */
1430 enum object_type type;
1432 /* Get the data stream */
1433 memset(stream, 0, sizeof(*stream));
1434 stream->next_in = map;
1435 stream->avail_in = mapsize;
1436 stream->next_out = buffer;
1437 stream->avail_out = bufsiz;
1439 if (experimental_loose_object(map)) {
1441 * The old experimental format we no longer produce;
1442 * we can still read it.
1444 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1445 if (!used || !valid_loose_object_type[type])
1446 return -1;
1447 map += used;
1448 mapsize -= used;
1450 /* Set up the stream for the rest.. */
1451 stream->next_in = map;
1452 stream->avail_in = mapsize;
1453 git_inflate_init(stream);
1455 /* And generate the fake traditional header */
1456 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1457 typename(type), size);
1458 return 0;
1460 git_inflate_init(stream);
1461 return git_inflate(stream, 0);
1464 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1466 int bytes = strlen(buffer) + 1;
1467 unsigned char *buf = xmallocz(size);
1468 unsigned long n;
1469 int status = Z_OK;
1471 n = stream->total_out - bytes;
1472 if (n > size)
1473 n = size;
1474 memcpy(buf, (char *) buffer + bytes, n);
1475 bytes = n;
1476 if (bytes <= size) {
1478 * The above condition must be (bytes <= size), not
1479 * (bytes < size). In other words, even though we
1480 * expect no more output and set avail_out to zero,
1481 * the input zlib stream may have bytes that express
1482 * "this concludes the stream", and we *do* want to
1483 * eat that input.
1485 * Otherwise we would not be able to test that we
1486 * consumed all the input to reach the expected size;
1487 * we also want to check that zlib tells us that all
1488 * went well with status == Z_STREAM_END at the end.
1490 stream->next_out = buf + bytes;
1491 stream->avail_out = size - bytes;
1492 while (status == Z_OK)
1493 status = git_inflate(stream, Z_FINISH);
1495 if (status == Z_STREAM_END && !stream->avail_in) {
1496 git_inflate_end(stream);
1497 return buf;
1500 if (status < 0)
1501 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1502 else if (stream->avail_in)
1503 error("garbage at end of loose object '%s'",
1504 sha1_to_hex(sha1));
1505 free(buf);
1506 return NULL;
1510 * We used to just use "sscanf()", but that's actually way
1511 * too permissive for what we want to check. So do an anal
1512 * object header parse by hand.
1514 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1516 char type[10];
1517 int i;
1518 unsigned long size;
1521 * The type can be at most ten bytes (including the
1522 * terminating '\0' that we add), and is followed by
1523 * a space.
1525 i = 0;
1526 for (;;) {
1527 char c = *hdr++;
1528 if (c == ' ')
1529 break;
1530 type[i++] = c;
1531 if (i >= sizeof(type))
1532 return -1;
1534 type[i] = 0;
1537 * The length must follow immediately, and be in canonical
1538 * decimal format (ie "010" is not valid).
1540 size = *hdr++ - '0';
1541 if (size > 9)
1542 return -1;
1543 if (size) {
1544 for (;;) {
1545 unsigned long c = *hdr - '0';
1546 if (c > 9)
1547 break;
1548 hdr++;
1549 size = size * 10 + c;
1552 *sizep = size;
1555 * The length must be followed by a zero byte
1557 return *hdr ? -1 : type_from_string(type);
1560 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1562 int ret;
1563 git_zstream stream;
1564 char hdr[8192];
1566 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1567 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1568 return NULL;
1570 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1573 unsigned long get_size_from_delta(struct packed_git *p,
1574 struct pack_window **w_curs,
1575 off_t curpos)
1577 const unsigned char *data;
1578 unsigned char delta_head[20], *in;
1579 git_zstream stream;
1580 int st;
1582 memset(&stream, 0, sizeof(stream));
1583 stream.next_out = delta_head;
1584 stream.avail_out = sizeof(delta_head);
1586 git_inflate_init(&stream);
1587 do {
1588 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1589 stream.next_in = in;
1590 st = git_inflate(&stream, Z_FINISH);
1591 curpos += stream.next_in - in;
1592 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1593 stream.total_out < sizeof(delta_head));
1594 git_inflate_end(&stream);
1595 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1596 error("delta data unpack-initial failed");
1597 return 0;
1600 /* Examine the initial part of the delta to figure out
1601 * the result size.
1603 data = delta_head;
1605 /* ignore base size */
1606 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1608 /* Read the result size */
1609 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1612 static off_t get_delta_base(struct packed_git *p,
1613 struct pack_window **w_curs,
1614 off_t *curpos,
1615 enum object_type type,
1616 off_t delta_obj_offset)
1618 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1619 off_t base_offset;
1621 /* use_pack() assured us we have [base_info, base_info + 20)
1622 * as a range that we can look at without walking off the
1623 * end of the mapped window. Its actually the hash size
1624 * that is assured. An OFS_DELTA longer than the hash size
1625 * is stupid, as then a REF_DELTA would be smaller to store.
1627 if (type == OBJ_OFS_DELTA) {
1628 unsigned used = 0;
1629 unsigned char c = base_info[used++];
1630 base_offset = c & 127;
1631 while (c & 128) {
1632 base_offset += 1;
1633 if (!base_offset || MSB(base_offset, 7))
1634 return 0; /* overflow */
1635 c = base_info[used++];
1636 base_offset = (base_offset << 7) + (c & 127);
1638 base_offset = delta_obj_offset - base_offset;
1639 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1640 return 0; /* out of bound */
1641 *curpos += used;
1642 } else if (type == OBJ_REF_DELTA) {
1643 /* The base entry _must_ be in the same pack */
1644 base_offset = find_pack_entry_one(base_info, p);
1645 *curpos += 20;
1646 } else
1647 die("I am totally screwed");
1648 return base_offset;
1651 /* forward declaration for a mutually recursive function */
1652 static int packed_object_info(struct packed_git *p, off_t offset,
1653 unsigned long *sizep, int *rtype);
1655 static int packed_delta_info(struct packed_git *p,
1656 struct pack_window **w_curs,
1657 off_t curpos,
1658 enum object_type type,
1659 off_t obj_offset,
1660 unsigned long *sizep)
1662 off_t base_offset;
1664 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1665 if (!base_offset)
1666 return OBJ_BAD;
1667 type = packed_object_info(p, base_offset, NULL, NULL);
1668 if (type <= OBJ_NONE) {
1669 struct revindex_entry *revidx;
1670 const unsigned char *base_sha1;
1671 revidx = find_pack_revindex(p, base_offset);
1672 if (!revidx)
1673 return OBJ_BAD;
1674 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1675 mark_bad_packed_object(p, base_sha1);
1676 type = sha1_object_info(base_sha1, NULL);
1677 if (type <= OBJ_NONE)
1678 return OBJ_BAD;
1681 /* We choose to only get the type of the base object and
1682 * ignore potentially corrupt pack file that expects the delta
1683 * based on a base with a wrong size. This saves tons of
1684 * inflate() calls.
1686 if (sizep) {
1687 *sizep = get_size_from_delta(p, w_curs, curpos);
1688 if (*sizep == 0)
1689 type = OBJ_BAD;
1692 return type;
1695 int unpack_object_header(struct packed_git *p,
1696 struct pack_window **w_curs,
1697 off_t *curpos,
1698 unsigned long *sizep)
1700 unsigned char *base;
1701 unsigned long left;
1702 unsigned long used;
1703 enum object_type type;
1705 /* use_pack() assures us we have [base, base + 20) available
1706 * as a range that we can look at. (Its actually the hash
1707 * size that is assured.) With our object header encoding
1708 * the maximum deflated object size is 2^137, which is just
1709 * insane, so we know won't exceed what we have been given.
1711 base = use_pack(p, w_curs, *curpos, &left);
1712 used = unpack_object_header_buffer(base, left, &type, sizep);
1713 if (!used) {
1714 type = OBJ_BAD;
1715 } else
1716 *curpos += used;
1718 return type;
1721 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1722 unsigned long *sizep, int *rtype)
1724 struct pack_window *w_curs = NULL;
1725 unsigned long size;
1726 off_t curpos = obj_offset;
1727 enum object_type type;
1729 type = unpack_object_header(p, &w_curs, &curpos, &size);
1730 if (rtype)
1731 *rtype = type; /* representation type */
1733 switch (type) {
1734 case OBJ_OFS_DELTA:
1735 case OBJ_REF_DELTA:
1736 type = packed_delta_info(p, &w_curs, curpos,
1737 type, obj_offset, sizep);
1738 break;
1739 case OBJ_COMMIT:
1740 case OBJ_TREE:
1741 case OBJ_BLOB:
1742 case OBJ_TAG:
1743 if (sizep)
1744 *sizep = size;
1745 break;
1746 default:
1747 error("unknown object type %i at offset %"PRIuMAX" in %s",
1748 type, (uintmax_t)obj_offset, p->pack_name);
1749 type = OBJ_BAD;
1751 unuse_pack(&w_curs);
1752 return type;
1755 static void *unpack_compressed_entry(struct packed_git *p,
1756 struct pack_window **w_curs,
1757 off_t curpos,
1758 unsigned long size)
1760 int st;
1761 git_zstream stream;
1762 unsigned char *buffer, *in;
1764 buffer = xmallocz(size);
1765 memset(&stream, 0, sizeof(stream));
1766 stream.next_out = buffer;
1767 stream.avail_out = size + 1;
1769 git_inflate_init(&stream);
1770 do {
1771 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1772 stream.next_in = in;
1773 st = git_inflate(&stream, Z_FINISH);
1774 if (!stream.avail_out)
1775 break; /* the payload is larger than it should be */
1776 curpos += stream.next_in - in;
1777 } while (st == Z_OK || st == Z_BUF_ERROR);
1778 git_inflate_end(&stream);
1779 if ((st != Z_STREAM_END) || stream.total_out != size) {
1780 free(buffer);
1781 return NULL;
1784 return buffer;
1787 #define MAX_DELTA_CACHE (256)
1789 static size_t delta_base_cached;
1791 static struct delta_base_cache_lru_list {
1792 struct delta_base_cache_lru_list *prev;
1793 struct delta_base_cache_lru_list *next;
1794 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1796 static struct delta_base_cache_entry {
1797 struct delta_base_cache_lru_list lru;
1798 void *data;
1799 struct packed_git *p;
1800 off_t base_offset;
1801 unsigned long size;
1802 enum object_type type;
1803 } delta_base_cache[MAX_DELTA_CACHE];
1805 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1807 unsigned long hash;
1809 hash = (unsigned long)p + (unsigned long)base_offset;
1810 hash += (hash >> 8) + (hash >> 16);
1811 return hash % MAX_DELTA_CACHE;
1814 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1816 unsigned long hash = pack_entry_hash(p, base_offset);
1817 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1818 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1821 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1822 unsigned long *base_size, enum object_type *type, int keep_cache)
1824 void *ret;
1825 unsigned long hash = pack_entry_hash(p, base_offset);
1826 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1828 ret = ent->data;
1829 if (!ret || ent->p != p || ent->base_offset != base_offset)
1830 return unpack_entry(p, base_offset, type, base_size);
1832 if (!keep_cache) {
1833 ent->data = NULL;
1834 ent->lru.next->prev = ent->lru.prev;
1835 ent->lru.prev->next = ent->lru.next;
1836 delta_base_cached -= ent->size;
1837 } else {
1838 ret = xmemdupz(ent->data, ent->size);
1840 *type = ent->type;
1841 *base_size = ent->size;
1842 return ret;
1845 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1847 if (ent->data) {
1848 free(ent->data);
1849 ent->data = NULL;
1850 ent->lru.next->prev = ent->lru.prev;
1851 ent->lru.prev->next = ent->lru.next;
1852 delta_base_cached -= ent->size;
1856 void clear_delta_base_cache(void)
1858 unsigned long p;
1859 for (p = 0; p < MAX_DELTA_CACHE; p++)
1860 release_delta_base_cache(&delta_base_cache[p]);
1863 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1864 void *base, unsigned long base_size, enum object_type type)
1866 unsigned long hash = pack_entry_hash(p, base_offset);
1867 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1868 struct delta_base_cache_lru_list *lru;
1870 release_delta_base_cache(ent);
1871 delta_base_cached += base_size;
1873 for (lru = delta_base_cache_lru.next;
1874 delta_base_cached > delta_base_cache_limit
1875 && lru != &delta_base_cache_lru;
1876 lru = lru->next) {
1877 struct delta_base_cache_entry *f = (void *)lru;
1878 if (f->type == OBJ_BLOB)
1879 release_delta_base_cache(f);
1881 for (lru = delta_base_cache_lru.next;
1882 delta_base_cached > delta_base_cache_limit
1883 && lru != &delta_base_cache_lru;
1884 lru = lru->next) {
1885 struct delta_base_cache_entry *f = (void *)lru;
1886 release_delta_base_cache(f);
1889 ent->p = p;
1890 ent->base_offset = base_offset;
1891 ent->type = type;
1892 ent->data = base;
1893 ent->size = base_size;
1894 ent->lru.next = &delta_base_cache_lru;
1895 ent->lru.prev = delta_base_cache_lru.prev;
1896 delta_base_cache_lru.prev->next = &ent->lru;
1897 delta_base_cache_lru.prev = &ent->lru;
1900 static void *read_object(const unsigned char *sha1, enum object_type *type,
1901 unsigned long *size);
1903 static void *unpack_delta_entry(struct packed_git *p,
1904 struct pack_window **w_curs,
1905 off_t curpos,
1906 unsigned long delta_size,
1907 off_t obj_offset,
1908 enum object_type *type,
1909 unsigned long *sizep)
1911 void *delta_data, *result, *base;
1912 unsigned long base_size;
1913 off_t base_offset;
1915 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1916 if (!base_offset) {
1917 error("failed to validate delta base reference "
1918 "at offset %"PRIuMAX" from %s",
1919 (uintmax_t)curpos, p->pack_name);
1920 return NULL;
1922 unuse_pack(w_curs);
1923 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1924 if (!base) {
1926 * We're probably in deep shit, but let's try to fetch
1927 * the required base anyway from another pack or loose.
1928 * This is costly but should happen only in the presence
1929 * of a corrupted pack, and is better than failing outright.
1931 struct revindex_entry *revidx;
1932 const unsigned char *base_sha1;
1933 revidx = find_pack_revindex(p, base_offset);
1934 if (!revidx)
1935 return NULL;
1936 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1937 error("failed to read delta base object %s"
1938 " at offset %"PRIuMAX" from %s",
1939 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1940 p->pack_name);
1941 mark_bad_packed_object(p, base_sha1);
1942 base = read_object(base_sha1, type, &base_size);
1943 if (!base)
1944 return NULL;
1947 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1948 if (!delta_data) {
1949 error("failed to unpack compressed delta "
1950 "at offset %"PRIuMAX" from %s",
1951 (uintmax_t)curpos, p->pack_name);
1952 free(base);
1953 return NULL;
1955 result = patch_delta(base, base_size,
1956 delta_data, delta_size,
1957 sizep);
1958 if (!result)
1959 die("failed to apply delta");
1960 free(delta_data);
1961 add_delta_base_cache(p, base_offset, base, base_size, *type);
1962 return result;
1965 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1967 static FILE *log_file;
1969 if (!log_file) {
1970 log_file = fopen(log_pack_access, "w");
1971 if (!log_file) {
1972 error("cannot open pack access log '%s' for writing: %s",
1973 log_pack_access, strerror(errno));
1974 log_pack_access = NULL;
1975 return;
1978 fprintf(log_file, "%s %"PRIuMAX"\n",
1979 p->pack_name, (uintmax_t)obj_offset);
1980 fflush(log_file);
1983 int do_check_packed_object_crc;
1985 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1986 enum object_type *type, unsigned long *sizep)
1988 struct pack_window *w_curs = NULL;
1989 off_t curpos = obj_offset;
1990 void *data;
1992 if (log_pack_access)
1993 write_pack_access_log(p, obj_offset);
1995 if (do_check_packed_object_crc && p->index_version > 1) {
1996 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1997 unsigned long len = revidx[1].offset - obj_offset;
1998 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1999 const unsigned char *sha1 =
2000 nth_packed_object_sha1(p, revidx->nr);
2001 error("bad packed object CRC for %s",
2002 sha1_to_hex(sha1));
2003 mark_bad_packed_object(p, sha1);
2004 unuse_pack(&w_curs);
2005 return NULL;
2009 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
2010 switch (*type) {
2011 case OBJ_OFS_DELTA:
2012 case OBJ_REF_DELTA:
2013 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
2014 obj_offset, type, sizep);
2015 break;
2016 case OBJ_COMMIT:
2017 case OBJ_TREE:
2018 case OBJ_BLOB:
2019 case OBJ_TAG:
2020 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
2021 break;
2022 default:
2023 data = NULL;
2024 error("unknown object type %i at offset %"PRIuMAX" in %s",
2025 *type, (uintmax_t)obj_offset, p->pack_name);
2027 unuse_pack(&w_curs);
2028 return data;
2031 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2032 uint32_t n)
2034 const unsigned char *index = p->index_data;
2035 if (!index) {
2036 if (open_pack_index(p))
2037 return NULL;
2038 index = p->index_data;
2040 if (n >= p->num_objects)
2041 return NULL;
2042 index += 4 * 256;
2043 if (p->index_version == 1) {
2044 return index + 24 * n + 4;
2045 } else {
2046 index += 8;
2047 return index + 20 * n;
2051 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2053 const unsigned char *index = p->index_data;
2054 index += 4 * 256;
2055 if (p->index_version == 1) {
2056 return ntohl(*((uint32_t *)(index + 24 * n)));
2057 } else {
2058 uint32_t off;
2059 index += 8 + p->num_objects * (20 + 4);
2060 off = ntohl(*((uint32_t *)(index + 4 * n)));
2061 if (!(off & 0x80000000))
2062 return off;
2063 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2064 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2065 ntohl(*((uint32_t *)(index + 4)));
2069 off_t find_pack_entry_one(const unsigned char *sha1,
2070 struct packed_git *p)
2072 const uint32_t *level1_ofs = p->index_data;
2073 const unsigned char *index = p->index_data;
2074 unsigned hi, lo, stride;
2075 static int use_lookup = -1;
2076 static int debug_lookup = -1;
2078 if (debug_lookup < 0)
2079 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2081 if (!index) {
2082 if (open_pack_index(p))
2083 return 0;
2084 level1_ofs = p->index_data;
2085 index = p->index_data;
2087 if (p->index_version > 1) {
2088 level1_ofs += 2;
2089 index += 8;
2091 index += 4 * 256;
2092 hi = ntohl(level1_ofs[*sha1]);
2093 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2094 if (p->index_version > 1) {
2095 stride = 20;
2096 } else {
2097 stride = 24;
2098 index += 4;
2101 if (debug_lookup)
2102 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2103 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2105 if (use_lookup < 0)
2106 use_lookup = !!getenv("GIT_USE_LOOKUP");
2107 if (use_lookup) {
2108 int pos = sha1_entry_pos(index, stride, 0,
2109 lo, hi, p->num_objects, sha1);
2110 if (pos < 0)
2111 return 0;
2112 return nth_packed_object_offset(p, pos);
2115 do {
2116 unsigned mi = (lo + hi) / 2;
2117 int cmp = hashcmp(index + mi * stride, sha1);
2119 if (debug_lookup)
2120 printf("lo %u hi %u rg %u mi %u\n",
2121 lo, hi, hi - lo, mi);
2122 if (!cmp)
2123 return nth_packed_object_offset(p, mi);
2124 if (cmp > 0)
2125 hi = mi;
2126 else
2127 lo = mi+1;
2128 } while (lo < hi);
2129 return 0;
2132 int is_pack_valid(struct packed_git *p)
2134 /* An already open pack is known to be valid. */
2135 if (p->pack_fd != -1)
2136 return 1;
2138 /* If the pack has one window completely covering the
2139 * file size, the pack is known to be valid even if
2140 * the descriptor is not currently open.
2142 if (p->windows) {
2143 struct pack_window *w = p->windows;
2145 if (!w->offset && w->len == p->pack_size)
2146 return 1;
2149 /* Force the pack to open to prove its valid. */
2150 return !open_packed_git(p);
2153 static int fill_pack_entry(const unsigned char *sha1,
2154 struct pack_entry *e,
2155 struct packed_git *p)
2157 off_t offset;
2159 if (p->num_bad_objects) {
2160 unsigned i;
2161 for (i = 0; i < p->num_bad_objects; i++)
2162 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2163 return 0;
2166 offset = find_pack_entry_one(sha1, p);
2167 if (!offset)
2168 return 0;
2171 * We are about to tell the caller where they can locate the
2172 * requested object. We better make sure the packfile is
2173 * still here and can be accessed before supplying that
2174 * answer, as it may have been deleted since the index was
2175 * loaded!
2177 if (!is_pack_valid(p)) {
2178 warning("packfile %s cannot be accessed", p->pack_name);
2179 return 0;
2181 e->offset = offset;
2182 e->p = p;
2183 hashcpy(e->sha1, sha1);
2184 return 1;
2187 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2189 struct packed_git *p;
2191 prepare_packed_git();
2192 if (!packed_git)
2193 return 0;
2195 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2196 return 1;
2198 for (p = packed_git; p; p = p->next) {
2199 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2200 continue;
2202 last_found_pack = p;
2203 return 1;
2205 return 0;
2208 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2209 struct packed_git *packs)
2211 struct packed_git *p;
2213 for (p = packs; p; p = p->next) {
2214 if (find_pack_entry_one(sha1, p))
2215 return p;
2217 return NULL;
2221 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2223 int status;
2224 unsigned long mapsize, size;
2225 void *map;
2226 git_zstream stream;
2227 char hdr[32];
2229 map = map_sha1_file(sha1, &mapsize);
2230 if (!map)
2231 return error("unable to find %s", sha1_to_hex(sha1));
2232 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2233 status = error("unable to unpack %s header",
2234 sha1_to_hex(sha1));
2235 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2236 status = error("unable to parse %s header", sha1_to_hex(sha1));
2237 else if (sizep)
2238 *sizep = size;
2239 git_inflate_end(&stream);
2240 munmap(map, mapsize);
2241 return status;
2244 /* returns enum object_type or negative */
2245 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2247 struct cached_object *co;
2248 struct pack_entry e;
2249 int status, rtype;
2251 co = find_cached_object(sha1);
2252 if (co) {
2253 if (oi->sizep)
2254 *(oi->sizep) = co->size;
2255 oi->whence = OI_CACHED;
2256 return co->type;
2259 if (!find_pack_entry(sha1, &e)) {
2260 /* Most likely it's a loose object. */
2261 status = sha1_loose_object_info(sha1, oi->sizep);
2262 if (status >= 0) {
2263 oi->whence = OI_LOOSE;
2264 return status;
2267 /* Not a loose object; someone else may have just packed it. */
2268 reprepare_packed_git();
2269 if (!find_pack_entry(sha1, &e))
2270 return status;
2273 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2274 if (status < 0) {
2275 mark_bad_packed_object(e.p, sha1);
2276 status = sha1_object_info_extended(sha1, oi);
2277 } else if (in_delta_base_cache(e.p, e.offset)) {
2278 oi->whence = OI_DBCACHED;
2279 } else {
2280 oi->whence = OI_PACKED;
2281 oi->u.packed.offset = e.offset;
2282 oi->u.packed.pack = e.p;
2283 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2284 rtype == OBJ_OFS_DELTA);
2287 return status;
2290 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2292 struct object_info oi;
2294 oi.sizep = sizep;
2295 return sha1_object_info_extended(sha1, &oi);
2298 static void *read_packed_sha1(const unsigned char *sha1,
2299 enum object_type *type, unsigned long *size)
2301 struct pack_entry e;
2302 void *data;
2304 if (!find_pack_entry(sha1, &e))
2305 return NULL;
2306 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2307 if (!data) {
2309 * We're probably in deep shit, but let's try to fetch
2310 * the required object anyway from another pack or loose.
2311 * This should happen only in the presence of a corrupted
2312 * pack, and is better than failing outright.
2314 error("failed to read object %s at offset %"PRIuMAX" from %s",
2315 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2316 mark_bad_packed_object(e.p, sha1);
2317 data = read_object(sha1, type, size);
2319 return data;
2322 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2323 unsigned char *sha1)
2325 struct cached_object *co;
2327 hash_sha1_file(buf, len, typename(type), sha1);
2328 if (has_sha1_file(sha1) || find_cached_object(sha1))
2329 return 0;
2330 if (cached_object_alloc <= cached_object_nr) {
2331 cached_object_alloc = alloc_nr(cached_object_alloc);
2332 cached_objects = xrealloc(cached_objects,
2333 sizeof(*cached_objects) *
2334 cached_object_alloc);
2336 co = &cached_objects[cached_object_nr++];
2337 co->size = len;
2338 co->type = type;
2339 co->buf = xmalloc(len);
2340 memcpy(co->buf, buf, len);
2341 hashcpy(co->sha1, sha1);
2342 return 0;
2345 static void *read_object(const unsigned char *sha1, enum object_type *type,
2346 unsigned long *size)
2348 unsigned long mapsize;
2349 void *map, *buf;
2350 struct cached_object *co;
2352 co = find_cached_object(sha1);
2353 if (co) {
2354 *type = co->type;
2355 *size = co->size;
2356 return xmemdupz(co->buf, co->size);
2359 buf = read_packed_sha1(sha1, type, size);
2360 if (buf)
2361 return buf;
2362 map = map_sha1_file(sha1, &mapsize);
2363 if (map) {
2364 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2365 munmap(map, mapsize);
2366 return buf;
2368 reprepare_packed_git();
2369 return read_packed_sha1(sha1, type, size);
2373 * This function dies on corrupt objects; the callers who want to
2374 * deal with them should arrange to call read_object() and give error
2375 * messages themselves.
2377 void *read_sha1_file_extended(const unsigned char *sha1,
2378 enum object_type *type,
2379 unsigned long *size,
2380 unsigned flag)
2382 void *data;
2383 char *path;
2384 const struct packed_git *p;
2385 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2386 ? lookup_replace_object(sha1) : sha1;
2388 errno = 0;
2389 data = read_object(repl, type, size);
2390 if (data)
2391 return data;
2393 if (errno && errno != ENOENT)
2394 die_errno("failed to read object %s", sha1_to_hex(sha1));
2396 /* die if we replaced an object with one that does not exist */
2397 if (repl != sha1)
2398 die("replacement %s not found for %s",
2399 sha1_to_hex(repl), sha1_to_hex(sha1));
2401 if (has_loose_object(repl)) {
2402 path = sha1_file_name(sha1);
2403 die("loose object %s (stored in %s) is corrupt",
2404 sha1_to_hex(repl), path);
2407 if ((p = has_packed_and_bad(repl)) != NULL)
2408 die("packed object %s (stored in %s) is corrupt",
2409 sha1_to_hex(repl), p->pack_name);
2411 return NULL;
2414 void *read_object_with_reference(const unsigned char *sha1,
2415 const char *required_type_name,
2416 unsigned long *size,
2417 unsigned char *actual_sha1_return)
2419 enum object_type type, required_type;
2420 void *buffer;
2421 unsigned long isize;
2422 unsigned char actual_sha1[20];
2424 required_type = type_from_string(required_type_name);
2425 hashcpy(actual_sha1, sha1);
2426 while (1) {
2427 int ref_length = -1;
2428 const char *ref_type = NULL;
2430 buffer = read_sha1_file(actual_sha1, &type, &isize);
2431 if (!buffer)
2432 return NULL;
2433 if (type == required_type) {
2434 *size = isize;
2435 if (actual_sha1_return)
2436 hashcpy(actual_sha1_return, actual_sha1);
2437 return buffer;
2439 /* Handle references */
2440 else if (type == OBJ_COMMIT)
2441 ref_type = "tree ";
2442 else if (type == OBJ_TAG)
2443 ref_type = "object ";
2444 else {
2445 free(buffer);
2446 return NULL;
2448 ref_length = strlen(ref_type);
2450 if (ref_length + 40 > isize ||
2451 memcmp(buffer, ref_type, ref_length) ||
2452 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2453 free(buffer);
2454 return NULL;
2456 free(buffer);
2457 /* Now we have the ID of the referred-to object in
2458 * actual_sha1. Check again. */
2462 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2463 const char *type, unsigned char *sha1,
2464 char *hdr, int *hdrlen)
2466 git_SHA_CTX c;
2468 /* Generate the header */
2469 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2471 /* Sha1.. */
2472 git_SHA1_Init(&c);
2473 git_SHA1_Update(&c, hdr, *hdrlen);
2474 git_SHA1_Update(&c, buf, len);
2475 git_SHA1_Final(sha1, &c);
2479 * Move the just written object into its final resting place.
2480 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2481 * "moving" is only a part of what it does, when no patch between
2482 * master to pu changes the call sites of this function.
2484 int move_temp_to_file(const char *tmpfile, const char *filename)
2486 int ret = 0;
2488 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2489 goto try_rename;
2490 else if (link(tmpfile, filename))
2491 ret = errno;
2494 * Coda hack - coda doesn't like cross-directory links,
2495 * so we fall back to a rename, which will mean that it
2496 * won't be able to check collisions, but that's not a
2497 * big deal.
2499 * The same holds for FAT formatted media.
2501 * When this succeeds, we just return. We have nothing
2502 * left to unlink.
2504 if (ret && ret != EEXIST) {
2505 try_rename:
2506 if (!rename(tmpfile, filename))
2507 goto out;
2508 ret = errno;
2510 unlink_or_warn(tmpfile);
2511 if (ret) {
2512 if (ret != EEXIST) {
2513 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2515 /* FIXME!!! Collision check here ? */
2518 out:
2519 if (adjust_shared_perm(filename))
2520 return error("unable to set permission to '%s'", filename);
2521 return 0;
2524 static int write_buffer(int fd, const void *buf, size_t len)
2526 if (write_in_full(fd, buf, len) < 0)
2527 return error("file write error (%s)", strerror(errno));
2528 return 0;
2531 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2532 unsigned char *sha1)
2534 char hdr[32];
2535 int hdrlen;
2536 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2537 return 0;
2540 /* Finalize a file on disk, and close it. */
2541 static void close_sha1_file(int fd)
2543 if (fsync_object_files)
2544 fsync_or_die(fd, "sha1 file");
2545 if (close(fd) != 0)
2546 die_errno("error when closing sha1 file");
2549 /* Size of directory component, including the ending '/' */
2550 static inline int directory_size(const char *filename)
2552 const char *s = strrchr(filename, '/');
2553 if (!s)
2554 return 0;
2555 return s - filename + 1;
2559 * This creates a temporary file in the same directory as the final
2560 * 'filename'
2562 * We want to avoid cross-directory filename renames, because those
2563 * can have problems on various filesystems (FAT, NFS, Coda).
2565 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2567 int fd, dirlen = directory_size(filename);
2569 if (dirlen + 20 > bufsiz) {
2570 errno = ENAMETOOLONG;
2571 return -1;
2573 memcpy(buffer, filename, dirlen);
2574 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2575 fd = git_mkstemp_mode(buffer, 0444);
2576 if (fd < 0 && dirlen && errno == ENOENT) {
2577 /* Make sure the directory exists */
2578 memcpy(buffer, filename, dirlen);
2579 buffer[dirlen-1] = 0;
2580 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2581 return -1;
2583 /* Try again */
2584 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2585 fd = git_mkstemp_mode(buffer, 0444);
2587 return fd;
2590 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2591 const void *buf, unsigned long len, time_t mtime)
2593 int fd, ret;
2594 unsigned char compressed[4096];
2595 git_zstream stream;
2596 git_SHA_CTX c;
2597 unsigned char parano_sha1[20];
2598 char *filename;
2599 static char tmp_file[PATH_MAX];
2601 filename = sha1_file_name(sha1);
2602 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2603 if (fd < 0) {
2604 if (errno == EACCES)
2605 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2606 else
2607 return error("unable to create temporary file: %s", strerror(errno));
2610 /* Set it up */
2611 memset(&stream, 0, sizeof(stream));
2612 git_deflate_init(&stream, zlib_compression_level);
2613 stream.next_out = compressed;
2614 stream.avail_out = sizeof(compressed);
2615 git_SHA1_Init(&c);
2617 /* First header.. */
2618 stream.next_in = (unsigned char *)hdr;
2619 stream.avail_in = hdrlen;
2620 while (git_deflate(&stream, 0) == Z_OK)
2621 ; /* nothing */
2622 git_SHA1_Update(&c, hdr, hdrlen);
2624 /* Then the data itself.. */
2625 stream.next_in = (void *)buf;
2626 stream.avail_in = len;
2627 do {
2628 unsigned char *in0 = stream.next_in;
2629 ret = git_deflate(&stream, Z_FINISH);
2630 git_SHA1_Update(&c, in0, stream.next_in - in0);
2631 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2632 die("unable to write sha1 file");
2633 stream.next_out = compressed;
2634 stream.avail_out = sizeof(compressed);
2635 } while (ret == Z_OK);
2637 if (ret != Z_STREAM_END)
2638 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2639 ret = git_deflate_end_gently(&stream);
2640 if (ret != Z_OK)
2641 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2642 git_SHA1_Final(parano_sha1, &c);
2643 if (hashcmp(sha1, parano_sha1) != 0)
2644 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2646 close_sha1_file(fd);
2648 if (mtime) {
2649 struct utimbuf utb;
2650 utb.actime = mtime;
2651 utb.modtime = mtime;
2652 if (utime(tmp_file, &utb) < 0)
2653 warning("failed utime() on %s: %s",
2654 tmp_file, strerror(errno));
2657 return move_temp_to_file(tmp_file, filename);
2660 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2662 unsigned char sha1[20];
2663 char hdr[32];
2664 int hdrlen;
2666 /* Normally if we have it in the pack then we do not bother writing
2667 * it out into .git/objects/??/?{38} file.
2669 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2670 if (returnsha1)
2671 hashcpy(returnsha1, sha1);
2672 if (has_sha1_file(sha1))
2673 return 0;
2674 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2677 int force_object_loose(const unsigned char *sha1, time_t mtime)
2679 void *buf;
2680 unsigned long len;
2681 enum object_type type;
2682 char hdr[32];
2683 int hdrlen;
2684 int ret;
2686 if (has_loose_object(sha1))
2687 return 0;
2688 buf = read_packed_sha1(sha1, &type, &len);
2689 if (!buf)
2690 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2691 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2692 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2693 free(buf);
2695 return ret;
2698 int has_pack_index(const unsigned char *sha1)
2700 struct stat st;
2701 if (stat(sha1_pack_index_name(sha1), &st))
2702 return 0;
2703 return 1;
2706 int has_sha1_pack(const unsigned char *sha1)
2708 struct pack_entry e;
2709 return find_pack_entry(sha1, &e);
2712 int has_sha1_file(const unsigned char *sha1)
2714 struct pack_entry e;
2716 if (find_pack_entry(sha1, &e))
2717 return 1;
2718 return has_loose_object(sha1);
2721 static void check_tree(const void *buf, size_t size)
2723 struct tree_desc desc;
2724 struct name_entry entry;
2726 init_tree_desc(&desc, buf, size);
2727 while (tree_entry(&desc, &entry))
2728 /* do nothing
2729 * tree_entry() will die() on malformed entries */
2733 static void check_commit(const void *buf, size_t size)
2735 struct commit c;
2736 memset(&c, 0, sizeof(c));
2737 if (parse_commit_buffer(&c, buf, size))
2738 die("corrupt commit");
2741 static void check_tag(const void *buf, size_t size)
2743 struct tag t;
2744 memset(&t, 0, sizeof(t));
2745 if (parse_tag_buffer(&t, buf, size))
2746 die("corrupt tag");
2749 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2750 enum object_type type,
2751 const char *path, unsigned flags)
2753 int ret, re_allocated = 0;
2754 int write_object = flags & HASH_WRITE_OBJECT;
2756 if (!type)
2757 type = OBJ_BLOB;
2760 * Convert blobs to git internal format
2762 if ((type == OBJ_BLOB) && path) {
2763 struct strbuf nbuf = STRBUF_INIT;
2764 if (convert_to_git(path, buf, size, &nbuf,
2765 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2766 buf = strbuf_detach(&nbuf, &size);
2767 re_allocated = 1;
2770 if (flags & HASH_FORMAT_CHECK) {
2771 if (type == OBJ_TREE)
2772 check_tree(buf, size);
2773 if (type == OBJ_COMMIT)
2774 check_commit(buf, size);
2775 if (type == OBJ_TAG)
2776 check_tag(buf, size);
2779 if (write_object)
2780 ret = write_sha1_file(buf, size, typename(type), sha1);
2781 else
2782 ret = hash_sha1_file(buf, size, typename(type), sha1);
2783 if (re_allocated)
2784 free(buf);
2785 return ret;
2788 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2789 const char *path, unsigned flags)
2791 struct strbuf sbuf = STRBUF_INIT;
2792 int ret;
2794 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2795 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2796 else
2797 ret = -1;
2798 strbuf_release(&sbuf);
2799 return ret;
2802 #define SMALL_FILE_SIZE (32*1024)
2804 static int index_core(unsigned char *sha1, int fd, size_t size,
2805 enum object_type type, const char *path,
2806 unsigned flags)
2808 int ret;
2810 if (!size) {
2811 ret = index_mem(sha1, NULL, size, type, path, flags);
2812 } else if (size <= SMALL_FILE_SIZE) {
2813 char *buf = xmalloc(size);
2814 if (size == read_in_full(fd, buf, size))
2815 ret = index_mem(sha1, buf, size, type, path, flags);
2816 else
2817 ret = error("short read %s", strerror(errno));
2818 free(buf);
2819 } else {
2820 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2821 ret = index_mem(sha1, buf, size, type, path, flags);
2822 munmap(buf, size);
2824 return ret;
2828 * This creates one packfile per large blob unless bulk-checkin
2829 * machinery is "plugged".
2831 * This also bypasses the usual "convert-to-git" dance, and that is on
2832 * purpose. We could write a streaming version of the converting
2833 * functions and insert that before feeding the data to fast-import
2834 * (or equivalent in-core API described above). However, that is
2835 * somewhat complicated, as we do not know the size of the filter
2836 * result, which we need to know beforehand when writing a git object.
2837 * Since the primary motivation for trying to stream from the working
2838 * tree file and to avoid mmaping it in core is to deal with large
2839 * binary blobs, they generally do not want to get any conversion, and
2840 * callers should avoid this code path when filters are requested.
2842 static int index_stream(unsigned char *sha1, int fd, size_t size,
2843 enum object_type type, const char *path,
2844 unsigned flags)
2846 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2849 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2850 enum object_type type, const char *path, unsigned flags)
2852 int ret;
2853 size_t size = xsize_t(st->st_size);
2855 if (!S_ISREG(st->st_mode))
2856 ret = index_pipe(sha1, fd, type, path, flags);
2857 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2858 (path && would_convert_to_git(path, NULL, 0, 0)))
2859 ret = index_core(sha1, fd, size, type, path, flags);
2860 else
2861 ret = index_stream(sha1, fd, size, type, path, flags);
2862 close(fd);
2863 return ret;
2866 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2868 int fd;
2869 struct strbuf sb = STRBUF_INIT;
2871 switch (st->st_mode & S_IFMT) {
2872 case S_IFREG:
2873 fd = open(path, O_RDONLY);
2874 if (fd < 0)
2875 return error("open(\"%s\"): %s", path,
2876 strerror(errno));
2877 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2878 return error("%s: failed to insert into database",
2879 path);
2880 break;
2881 case S_IFLNK:
2882 if (strbuf_readlink(&sb, path, st->st_size)) {
2883 char *errstr = strerror(errno);
2884 return error("readlink(\"%s\"): %s", path,
2885 errstr);
2887 if (!(flags & HASH_WRITE_OBJECT))
2888 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2889 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2890 return error("%s: failed to insert into database",
2891 path);
2892 strbuf_release(&sb);
2893 break;
2894 case S_IFDIR:
2895 return resolve_gitlink_ref(path, "HEAD", sha1);
2896 default:
2897 return error("%s: unsupported file type", path);
2899 return 0;
2902 int read_pack_header(int fd, struct pack_header *header)
2904 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2905 /* "eof before pack header was fully read" */
2906 return PH_ERROR_EOF;
2908 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2909 /* "protocol error (pack signature mismatch detected)" */
2910 return PH_ERROR_PACK_SIGNATURE;
2911 if (!pack_version_ok(header->hdr_version))
2912 /* "protocol error (pack version unsupported)" */
2913 return PH_ERROR_PROTOCOL;
2914 return 0;
2917 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2919 enum object_type type = sha1_object_info(sha1, NULL);
2920 if (type < 0)
2921 die("%s is not a valid object", sha1_to_hex(sha1));
2922 if (type != expect)
2923 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2924 typename(expect));