clone: run check_everything_connected
[git/gitweb.git] / sha1_file.c
blob0b99f336e62256606f3fffc3eaf0ee7727d428c5
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 *pos = '/';
128 return -1;
130 else if (adjust_shared_perm(path)) {
131 *pos = '/';
132 return -2;
134 *pos++ = '/';
136 return 0;
139 int safe_create_leading_directories_const(const char *path)
141 /* path points to cache entries, so xstrdup before messing with it */
142 char *buf = xstrdup(path);
143 int result = safe_create_leading_directories(buf);
144 free(buf);
145 return result;
148 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
150 int i;
151 for (i = 0; i < 20; i++) {
152 static char hex[] = "0123456789abcdef";
153 unsigned int val = sha1[i];
154 char *pos = pathbuf + i*2 + (i > 0);
155 *pos++ = hex[val >> 4];
156 *pos = hex[val & 0xf];
161 * NOTE! This returns a statically allocated buffer, so you have to be
162 * careful about using it. Do an "xstrdup()" if you need to save the
163 * filename.
165 * Also note that this returns the location for creating. Reading
166 * SHA1 file can happen from any alternate directory listed in the
167 * DB_ENVIRONMENT environment variable if it is not found in
168 * the primary object database.
170 char *sha1_file_name(const unsigned char *sha1)
172 static char buf[PATH_MAX];
173 const char *objdir;
174 int len;
176 objdir = get_object_directory();
177 len = strlen(objdir);
179 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
180 if (len + 43 > PATH_MAX)
181 die("insanely long object directory %s", objdir);
182 memcpy(buf, objdir, len);
183 buf[len] = '/';
184 buf[len+3] = '/';
185 buf[len+42] = '\0';
186 fill_sha1_path(buf + len + 1, sha1);
187 return buf;
190 static char *sha1_get_pack_name(const unsigned char *sha1,
191 char **name, char **base, const char *which)
193 static const char hex[] = "0123456789abcdef";
194 char *buf;
195 int i;
197 if (!*base) {
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 *base = xmalloc(len + 60);
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
203 *name = *base + len + 11;
206 buf = *name;
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
214 return *base;
217 char *sha1_pack_name(const unsigned char *sha1)
219 static char *name, *base;
221 return sha1_get_pack_name(sha1, &name, &base, "pack");
224 char *sha1_pack_index_name(const unsigned char *sha1)
226 static char *name, *base;
228 return sha1_get_pack_name(sha1, &name, &base, "idx");
231 struct alternate_object_database *alt_odb_list;
232 static struct alternate_object_database **alt_odb_tail;
234 static int git_open_noatime(const char *name);
237 * Prepare alternate object database registry.
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
251 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
253 const char *objdir = get_object_directory();
254 struct alternate_object_database *ent;
255 struct alternate_object_database *alt;
256 int pfxlen, entlen;
257 struct strbuf pathbuf = STRBUF_INIT;
259 if (!is_absolute_path(entry) && relative_base) {
260 strbuf_addstr(&pathbuf, real_path(relative_base));
261 strbuf_addch(&pathbuf, '/');
263 strbuf_addstr(&pathbuf, entry);
265 normalize_path_copy(pathbuf.buf, pathbuf.buf);
267 pfxlen = strlen(pathbuf.buf);
270 * The trailing slash after the directory name is given by
271 * this function at the end. Remove duplicates.
273 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
274 pfxlen -= 1;
276 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent = xmalloc(sizeof(*ent) + entlen);
278 memcpy(ent->base, pathbuf.buf, pfxlen);
279 strbuf_release(&pathbuf);
281 ent->name = ent->base + pfxlen + 1;
282 ent->base[pfxlen + 3] = '/';
283 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285 /* Detect cases where alternate disappeared */
286 if (!is_directory(ent->base)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
289 ent->base);
290 free(ent);
291 return -1;
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt = alt_odb_list; alt; alt = alt->next) {
298 if (!memcmp(ent->base, alt->base, pfxlen)) {
299 free(ent);
300 return -1;
303 if (!strcmp(ent->base, objdir)) {
304 free(ent);
305 return -1;
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
316 ent->base[pfxlen] = '/';
318 return 0;
321 static void link_alt_odb_entries(const char *alt, int len, int sep,
322 const char *relative_base, int depth)
324 struct string_list entries = STRING_LIST_INIT_NODUP;
325 char *alt_copy;
326 int i;
328 if (depth > 5) {
329 error("%s: ignoring alternate object stores, nesting too deep.",
330 relative_base);
331 return;
334 alt_copy = xmemdupz(alt, len);
335 string_list_split_in_place(&entries, alt_copy, sep, -1);
336 for (i = 0; i < entries.nr; i++) {
337 const char *entry = entries.items[i].string;
338 if (entry[0] == '\0' || entry[0] == '#')
339 continue;
340 if (!is_absolute_path(entry) && depth) {
341 error("%s: ignoring relative alternate object store %s",
342 relative_base, entry);
343 } else {
344 link_alt_odb_entry(entry, relative_base, depth);
347 string_list_clear(&entries, 0);
348 free(alt_copy);
351 void read_info_alternates(const char * relative_base, int depth)
353 char *map;
354 size_t mapsz;
355 struct stat st;
356 const char alt_file_name[] = "info/alternates";
357 /* Given that relative_base is no longer than PATH_MAX,
358 ensure that "path" has enough space to append "/", the
359 file name, "info/alternates", and a trailing NUL. */
360 char path[PATH_MAX + 1 + sizeof alt_file_name];
361 int fd;
363 sprintf(path, "%s/%s", relative_base, alt_file_name);
364 fd = git_open_noatime(path);
365 if (fd < 0)
366 return;
367 if (fstat(fd, &st) || (st.st_size == 0)) {
368 close(fd);
369 return;
371 mapsz = xsize_t(st.st_size);
372 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
373 close(fd);
375 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
377 munmap(map, mapsz);
380 void add_to_alternates_file(const char *reference)
382 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
383 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
384 char *alt = mkpath("%s\n", reference);
385 write_or_die(fd, alt, strlen(alt));
386 if (commit_lock_file(lock))
387 die("could not close alternates file");
388 if (alt_odb_tail)
389 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
392 void foreach_alt_odb(alt_odb_fn fn, void *cb)
394 struct alternate_object_database *ent;
396 prepare_alt_odb();
397 for (ent = alt_odb_list; ent; ent = ent->next)
398 if (fn(ent, cb))
399 return;
402 void prepare_alt_odb(void)
404 const char *alt;
406 if (alt_odb_tail)
407 return;
409 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
410 if (!alt) alt = "";
412 alt_odb_tail = &alt_odb_list;
413 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
415 read_info_alternates(get_object_directory(), 0);
418 static int has_loose_object_local(const unsigned char *sha1)
420 char *name = sha1_file_name(sha1);
421 return !access(name, F_OK);
424 int has_loose_object_nonlocal(const unsigned char *sha1)
426 struct alternate_object_database *alt;
427 prepare_alt_odb();
428 for (alt = alt_odb_list; alt; alt = alt->next) {
429 fill_sha1_path(alt->name, sha1);
430 if (!access(alt->base, F_OK))
431 return 1;
433 return 0;
436 static int has_loose_object(const unsigned char *sha1)
438 return has_loose_object_local(sha1) ||
439 has_loose_object_nonlocal(sha1);
442 static unsigned int pack_used_ctr;
443 static unsigned int pack_mmap_calls;
444 static unsigned int peak_pack_open_windows;
445 static unsigned int pack_open_windows;
446 static unsigned int pack_open_fds;
447 static unsigned int pack_max_fds;
448 static size_t peak_pack_mapped;
449 static size_t pack_mapped;
450 struct packed_git *packed_git;
452 void pack_report(void)
454 fprintf(stderr,
455 "pack_report: getpagesize() = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
457 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
458 sz_fmt(getpagesize()),
459 sz_fmt(packed_git_window_size),
460 sz_fmt(packed_git_limit));
461 fprintf(stderr,
462 "pack_report: pack_used_ctr = %10u\n"
463 "pack_report: pack_mmap_calls = %10u\n"
464 "pack_report: pack_open_windows = %10u / %10u\n"
465 "pack_report: pack_mapped = "
466 "%10" SZ_FMT " / %10" SZ_FMT "\n",
467 pack_used_ctr,
468 pack_mmap_calls,
469 pack_open_windows, peak_pack_open_windows,
470 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
473 static int check_packed_git_idx(const char *path, struct packed_git *p)
475 void *idx_map;
476 struct pack_idx_header *hdr;
477 size_t idx_size;
478 uint32_t version, nr, i, *index;
479 int fd = git_open_noatime(path);
480 struct stat st;
482 if (fd < 0)
483 return -1;
484 if (fstat(fd, &st)) {
485 close(fd);
486 return -1;
488 idx_size = xsize_t(st.st_size);
489 if (idx_size < 4 * 256 + 20 + 20) {
490 close(fd);
491 return error("index file %s is too small", path);
493 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
494 close(fd);
496 hdr = idx_map;
497 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
498 version = ntohl(hdr->idx_version);
499 if (version < 2 || version > 2) {
500 munmap(idx_map, idx_size);
501 return error("index file %s is version %"PRIu32
502 " and is not supported by this binary"
503 " (try upgrading GIT to a newer version)",
504 path, version);
506 } else
507 version = 1;
509 nr = 0;
510 index = idx_map;
511 if (version > 1)
512 index += 2; /* skip index header */
513 for (i = 0; i < 256; i++) {
514 uint32_t n = ntohl(index[i]);
515 if (n < nr) {
516 munmap(idx_map, idx_size);
517 return error("non-monotonic index %s", path);
519 nr = n;
522 if (version == 1) {
524 * Total size:
525 * - 256 index entries 4 bytes each
526 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
527 * - 20-byte SHA1 of the packfile
528 * - 20-byte SHA1 file checksum
530 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
531 munmap(idx_map, idx_size);
532 return error("wrong index v1 file size in %s", path);
534 } else if (version == 2) {
536 * Minimum size:
537 * - 8 bytes of header
538 * - 256 index entries 4 bytes each
539 * - 20-byte sha1 entry * nr
540 * - 4-byte crc entry * nr
541 * - 4-byte offset entry * nr
542 * - 20-byte SHA1 of the packfile
543 * - 20-byte SHA1 file checksum
544 * And after the 4-byte offset table might be a
545 * variable sized table containing 8-byte entries
546 * for offsets larger than 2^31.
548 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
549 unsigned long max_size = min_size;
550 if (nr)
551 max_size += (nr - 1)*8;
552 if (idx_size < min_size || idx_size > max_size) {
553 munmap(idx_map, idx_size);
554 return error("wrong index v2 file size in %s", path);
556 if (idx_size != min_size &&
558 * make sure we can deal with large pack offsets.
559 * 31-bit signed offset won't be enough, neither
560 * 32-bit unsigned one will be.
562 (sizeof(off_t) <= 4)) {
563 munmap(idx_map, idx_size);
564 return error("pack too large for current definition of off_t in %s", path);
568 p->index_version = version;
569 p->index_data = idx_map;
570 p->index_size = idx_size;
571 p->num_objects = nr;
572 return 0;
575 int open_pack_index(struct packed_git *p)
577 char *idx_name;
578 int ret;
580 if (p->index_data)
581 return 0;
583 idx_name = xstrdup(p->pack_name);
584 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
585 ret = check_packed_git_idx(idx_name, p);
586 free(idx_name);
587 return ret;
590 static void scan_windows(struct packed_git *p,
591 struct packed_git **lru_p,
592 struct pack_window **lru_w,
593 struct pack_window **lru_l)
595 struct pack_window *w, *w_l;
597 for (w_l = NULL, w = p->windows; w; w = w->next) {
598 if (!w->inuse_cnt) {
599 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
600 *lru_p = p;
601 *lru_w = w;
602 *lru_l = w_l;
605 w_l = w;
609 static int unuse_one_window(struct packed_git *current, int keep_fd)
611 struct packed_git *p, *lru_p = NULL;
612 struct pack_window *lru_w = NULL, *lru_l = NULL;
614 if (current)
615 scan_windows(current, &lru_p, &lru_w, &lru_l);
616 for (p = packed_git; p; p = p->next)
617 scan_windows(p, &lru_p, &lru_w, &lru_l);
618 if (lru_p) {
619 munmap(lru_w->base, lru_w->len);
620 pack_mapped -= lru_w->len;
621 if (lru_l)
622 lru_l->next = lru_w->next;
623 else {
624 lru_p->windows = lru_w->next;
625 if (!lru_p->windows && lru_p->pack_fd != -1
626 && lru_p->pack_fd != keep_fd) {
627 close(lru_p->pack_fd);
628 pack_open_fds--;
629 lru_p->pack_fd = -1;
632 free(lru_w);
633 pack_open_windows--;
634 return 1;
636 return 0;
639 void release_pack_memory(size_t need, int fd)
641 size_t cur = pack_mapped;
642 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
643 ; /* nothing */
646 void *xmmap(void *start, size_t length,
647 int prot, int flags, int fd, off_t offset)
649 void *ret = mmap(start, length, prot, flags, fd, offset);
650 if (ret == MAP_FAILED) {
651 if (!length)
652 return NULL;
653 release_pack_memory(length, fd);
654 ret = mmap(start, length, prot, flags, fd, offset);
655 if (ret == MAP_FAILED)
656 die_errno("Out of memory? mmap failed");
658 return ret;
661 void close_pack_windows(struct packed_git *p)
663 while (p->windows) {
664 struct pack_window *w = p->windows;
666 if (w->inuse_cnt)
667 die("pack '%s' still has open windows to it",
668 p->pack_name);
669 munmap(w->base, w->len);
670 pack_mapped -= w->len;
671 pack_open_windows--;
672 p->windows = w->next;
673 free(w);
677 void unuse_pack(struct pack_window **w_cursor)
679 struct pack_window *w = *w_cursor;
680 if (w) {
681 w->inuse_cnt--;
682 *w_cursor = NULL;
686 void close_pack_index(struct packed_git *p)
688 if (p->index_data) {
689 munmap((void *)p->index_data, p->index_size);
690 p->index_data = NULL;
695 * This is used by git-repack in case a newly created pack happens to
696 * contain the same set of objects as an existing one. In that case
697 * the resulting file might be different even if its name would be the
698 * same. It is best to close any reference to the old pack before it is
699 * replaced on disk. Of course no index pointers nor windows for given pack
700 * must subsist at this point. If ever objects from this pack are requested
701 * again, the new version of the pack will be reinitialized through
702 * reprepare_packed_git().
704 void free_pack_by_name(const char *pack_name)
706 struct packed_git *p, **pp = &packed_git;
708 while (*pp) {
709 p = *pp;
710 if (strcmp(pack_name, p->pack_name) == 0) {
711 clear_delta_base_cache();
712 close_pack_windows(p);
713 if (p->pack_fd != -1) {
714 close(p->pack_fd);
715 pack_open_fds--;
717 close_pack_index(p);
718 free(p->bad_object_sha1);
719 *pp = p->next;
720 if (last_found_pack == p)
721 last_found_pack = NULL;
722 free(p);
723 return;
725 pp = &p->next;
729 static unsigned int get_max_fd_limit(void)
731 #ifdef RLIMIT_NOFILE
732 struct rlimit lim;
734 if (getrlimit(RLIMIT_NOFILE, &lim))
735 die_errno("cannot get RLIMIT_NOFILE");
737 return lim.rlim_cur;
738 #elif defined(_SC_OPEN_MAX)
739 return sysconf(_SC_OPEN_MAX);
740 #elif defined(OPEN_MAX)
741 return OPEN_MAX;
742 #else
743 return 1; /* see the caller ;-) */
744 #endif
748 * Do not call this directly as this leaks p->pack_fd on error return;
749 * call open_packed_git() instead.
751 static int open_packed_git_1(struct packed_git *p)
753 struct stat st;
754 struct pack_header hdr;
755 unsigned char sha1[20];
756 unsigned char *idx_sha1;
757 long fd_flag;
759 if (!p->index_data && open_pack_index(p))
760 return error("packfile %s index unavailable", p->pack_name);
762 if (!pack_max_fds) {
763 unsigned int max_fds = get_max_fd_limit();
765 /* Save 3 for stdin/stdout/stderr, 22 for work */
766 if (25 < max_fds)
767 pack_max_fds = max_fds - 25;
768 else
769 pack_max_fds = 1;
772 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
773 ; /* nothing */
775 p->pack_fd = git_open_noatime(p->pack_name);
776 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
777 return -1;
778 pack_open_fds++;
780 /* If we created the struct before we had the pack we lack size. */
781 if (!p->pack_size) {
782 if (!S_ISREG(st.st_mode))
783 return error("packfile %s not a regular file", p->pack_name);
784 p->pack_size = st.st_size;
785 } else if (p->pack_size != st.st_size)
786 return error("packfile %s size changed", p->pack_name);
788 /* We leave these file descriptors open with sliding mmap;
789 * there is no point keeping them open across exec(), though.
791 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
792 if (fd_flag < 0)
793 return error("cannot determine file descriptor flags");
794 fd_flag |= FD_CLOEXEC;
795 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
796 return error("cannot set FD_CLOEXEC");
798 /* Verify we recognize this pack file format. */
799 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
800 return error("file %s is far too short to be a packfile", p->pack_name);
801 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
802 return error("file %s is not a GIT packfile", p->pack_name);
803 if (!pack_version_ok(hdr.hdr_version))
804 return error("packfile %s is version %"PRIu32" and not"
805 " supported (try upgrading GIT to a newer version)",
806 p->pack_name, ntohl(hdr.hdr_version));
808 /* Verify the pack matches its index. */
809 if (p->num_objects != ntohl(hdr.hdr_entries))
810 return error("packfile %s claims to have %"PRIu32" objects"
811 " while index indicates %"PRIu32" objects",
812 p->pack_name, ntohl(hdr.hdr_entries),
813 p->num_objects);
814 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
815 return error("end of packfile %s is unavailable", p->pack_name);
816 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
817 return error("packfile %s signature is unavailable", p->pack_name);
818 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
819 if (hashcmp(sha1, idx_sha1))
820 return error("packfile %s does not match index", p->pack_name);
821 return 0;
824 static int open_packed_git(struct packed_git *p)
826 if (!open_packed_git_1(p))
827 return 0;
828 if (p->pack_fd != -1) {
829 close(p->pack_fd);
830 pack_open_fds--;
831 p->pack_fd = -1;
833 return -1;
836 static int in_window(struct pack_window *win, off_t offset)
838 /* We must promise at least 20 bytes (one hash) after the
839 * offset is available from this window, otherwise the offset
840 * is not actually in this window and a different window (which
841 * has that one hash excess) must be used. This is to support
842 * the object header and delta base parsing routines below.
844 off_t win_off = win->offset;
845 return win_off <= offset
846 && (offset + 20) <= (win_off + win->len);
849 unsigned char *use_pack(struct packed_git *p,
850 struct pack_window **w_cursor,
851 off_t offset,
852 unsigned long *left)
854 struct pack_window *win = *w_cursor;
856 /* Since packfiles end in a hash of their content and it's
857 * pointless to ask for an offset into the middle of that
858 * hash, and the in_window function above wouldn't match
859 * don't allow an offset too close to the end of the file.
861 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
862 die("packfile %s cannot be accessed", p->pack_name);
863 if (offset > (p->pack_size - 20))
864 die("offset beyond end of packfile (truncated pack?)");
866 if (!win || !in_window(win, offset)) {
867 if (win)
868 win->inuse_cnt--;
869 for (win = p->windows; win; win = win->next) {
870 if (in_window(win, offset))
871 break;
873 if (!win) {
874 size_t window_align = packed_git_window_size / 2;
875 off_t len;
877 if (p->pack_fd == -1 && open_packed_git(p))
878 die("packfile %s cannot be accessed", p->pack_name);
880 win = xcalloc(1, sizeof(*win));
881 win->offset = (offset / window_align) * window_align;
882 len = p->pack_size - win->offset;
883 if (len > packed_git_window_size)
884 len = packed_git_window_size;
885 win->len = (size_t)len;
886 pack_mapped += win->len;
887 while (packed_git_limit < pack_mapped
888 && unuse_one_window(p, p->pack_fd))
889 ; /* nothing */
890 win->base = xmmap(NULL, win->len,
891 PROT_READ, MAP_PRIVATE,
892 p->pack_fd, win->offset);
893 if (win->base == MAP_FAILED)
894 die("packfile %s cannot be mapped: %s",
895 p->pack_name,
896 strerror(errno));
897 if (!win->offset && win->len == p->pack_size
898 && !p->do_not_close) {
899 close(p->pack_fd);
900 pack_open_fds--;
901 p->pack_fd = -1;
903 pack_mmap_calls++;
904 pack_open_windows++;
905 if (pack_mapped > peak_pack_mapped)
906 peak_pack_mapped = pack_mapped;
907 if (pack_open_windows > peak_pack_open_windows)
908 peak_pack_open_windows = pack_open_windows;
909 win->next = p->windows;
910 p->windows = win;
913 if (win != *w_cursor) {
914 win->last_used = pack_used_ctr++;
915 win->inuse_cnt++;
916 *w_cursor = win;
918 offset -= win->offset;
919 if (left)
920 *left = win->len - xsize_t(offset);
921 return win->base + offset;
924 static struct packed_git *alloc_packed_git(int extra)
926 struct packed_git *p = xmalloc(sizeof(*p) + extra);
927 memset(p, 0, sizeof(*p));
928 p->pack_fd = -1;
929 return p;
932 static void try_to_free_pack_memory(size_t size)
934 release_pack_memory(size, -1);
937 struct packed_git *add_packed_git(const char *path, int path_len, int local)
939 static int have_set_try_to_free_routine;
940 struct stat st;
941 struct packed_git *p = alloc_packed_git(path_len + 2);
943 if (!have_set_try_to_free_routine) {
944 have_set_try_to_free_routine = 1;
945 set_try_to_free_routine(try_to_free_pack_memory);
949 * Make sure a corresponding .pack file exists and that
950 * the index looks sane.
952 path_len -= strlen(".idx");
953 if (path_len < 1) {
954 free(p);
955 return NULL;
957 memcpy(p->pack_name, path, path_len);
959 strcpy(p->pack_name + path_len, ".keep");
960 if (!access(p->pack_name, F_OK))
961 p->pack_keep = 1;
963 strcpy(p->pack_name + path_len, ".pack");
964 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
965 free(p);
966 return NULL;
969 /* ok, it looks sane as far as we can check without
970 * actually mapping the pack file.
972 p->pack_size = st.st_size;
973 p->pack_local = local;
974 p->mtime = st.st_mtime;
975 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
976 hashclr(p->sha1);
977 return p;
980 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
982 const char *path = sha1_pack_name(sha1);
983 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
985 strcpy(p->pack_name, path);
986 hashcpy(p->sha1, sha1);
987 if (check_packed_git_idx(idx_path, p)) {
988 free(p);
989 return NULL;
992 return p;
995 void install_packed_git(struct packed_git *pack)
997 if (pack->pack_fd != -1)
998 pack_open_fds++;
1000 pack->next = packed_git;
1001 packed_git = pack;
1004 void (*report_garbage)(const char *desc, const char *path);
1006 static void report_helper(const struct string_list *list,
1007 int seen_bits, int first, int last)
1009 const char *msg;
1010 switch (seen_bits) {
1011 case 0:
1012 msg = "no corresponding .idx nor .pack";
1013 break;
1014 case 1:
1015 msg = "no corresponding .idx";
1016 break;
1017 case 2:
1018 msg = "no corresponding .pack";
1019 break;
1020 default:
1021 return;
1023 for (; first < last; first++)
1024 report_garbage(msg, list->items[first].string);
1027 static void report_pack_garbage(struct string_list *list)
1029 int i, baselen = -1, first = 0, seen_bits = 0;
1031 if (!report_garbage)
1032 return;
1034 sort_string_list(list);
1036 for (i = 0; i < list->nr; i++) {
1037 const char *path = list->items[i].string;
1038 if (baselen != -1 &&
1039 strncmp(path, list->items[first].string, baselen)) {
1040 report_helper(list, seen_bits, first, i);
1041 baselen = -1;
1042 seen_bits = 0;
1044 if (baselen == -1) {
1045 const char *dot = strrchr(path, '.');
1046 if (!dot) {
1047 report_garbage("garbage found", path);
1048 continue;
1050 baselen = dot - path + 1;
1051 first = i;
1053 if (!strcmp(path + baselen, "pack"))
1054 seen_bits |= 1;
1055 else if (!strcmp(path + baselen, "idx"))
1056 seen_bits |= 2;
1058 report_helper(list, seen_bits, first, list->nr);
1061 static void prepare_packed_git_one(char *objdir, int local)
1063 /* Ensure that this buffer is large enough so that we can
1064 append "/pack/" without clobbering the stack even if
1065 strlen(objdir) were PATH_MAX. */
1066 char path[PATH_MAX + 1 + 4 + 1 + 1];
1067 int len;
1068 DIR *dir;
1069 struct dirent *de;
1070 struct string_list garbage = STRING_LIST_INIT_DUP;
1072 sprintf(path, "%s/pack", objdir);
1073 len = strlen(path);
1074 dir = opendir(path);
1075 if (!dir) {
1076 if (errno != ENOENT)
1077 error("unable to open object pack directory: %s: %s",
1078 path, strerror(errno));
1079 return;
1081 path[len++] = '/';
1082 while ((de = readdir(dir)) != NULL) {
1083 int namelen = strlen(de->d_name);
1084 struct packed_git *p;
1086 if (len + namelen + 1 > sizeof(path)) {
1087 if (report_garbage) {
1088 struct strbuf sb = STRBUF_INIT;
1089 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1090 report_garbage("path too long", sb.buf);
1091 strbuf_release(&sb);
1093 continue;
1096 if (is_dot_or_dotdot(de->d_name))
1097 continue;
1099 strcpy(path + len, de->d_name);
1101 if (has_extension(de->d_name, ".idx")) {
1102 /* Don't reopen a pack we already have. */
1103 for (p = packed_git; p; p = p->next) {
1104 if (!memcmp(path, p->pack_name, len + namelen - 4))
1105 break;
1107 if (p == NULL &&
1109 * See if it really is a valid .idx file with
1110 * corresponding .pack file that we can map.
1112 (p = add_packed_git(path, len + namelen, local)) != NULL)
1113 install_packed_git(p);
1116 if (!report_garbage)
1117 continue;
1119 if (has_extension(de->d_name, ".idx") ||
1120 has_extension(de->d_name, ".pack") ||
1121 has_extension(de->d_name, ".keep"))
1122 string_list_append(&garbage, path);
1123 else
1124 report_garbage("garbage found", path);
1126 closedir(dir);
1127 report_pack_garbage(&garbage);
1128 string_list_clear(&garbage, 0);
1131 static int sort_pack(const void *a_, const void *b_)
1133 struct packed_git *a = *((struct packed_git **)a_);
1134 struct packed_git *b = *((struct packed_git **)b_);
1135 int st;
1138 * Local packs tend to contain objects specific to our
1139 * variant of the project than remote ones. In addition,
1140 * remote ones could be on a network mounted filesystem.
1141 * Favor local ones for these reasons.
1143 st = a->pack_local - b->pack_local;
1144 if (st)
1145 return -st;
1148 * Younger packs tend to contain more recent objects,
1149 * and more recent objects tend to get accessed more
1150 * often.
1152 if (a->mtime < b->mtime)
1153 return 1;
1154 else if (a->mtime == b->mtime)
1155 return 0;
1156 return -1;
1159 static void rearrange_packed_git(void)
1161 struct packed_git **ary, *p;
1162 int i, n;
1164 for (n = 0, p = packed_git; p; p = p->next)
1165 n++;
1166 if (n < 2)
1167 return;
1169 /* prepare an array of packed_git for easier sorting */
1170 ary = xcalloc(n, sizeof(struct packed_git *));
1171 for (n = 0, p = packed_git; p; p = p->next)
1172 ary[n++] = p;
1174 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1176 /* link them back again */
1177 for (i = 0; i < n - 1; i++)
1178 ary[i]->next = ary[i + 1];
1179 ary[n - 1]->next = NULL;
1180 packed_git = ary[0];
1182 free(ary);
1185 static int prepare_packed_git_run_once = 0;
1186 void prepare_packed_git(void)
1188 struct alternate_object_database *alt;
1190 if (prepare_packed_git_run_once)
1191 return;
1192 prepare_packed_git_one(get_object_directory(), 1);
1193 prepare_alt_odb();
1194 for (alt = alt_odb_list; alt; alt = alt->next) {
1195 alt->name[-1] = 0;
1196 prepare_packed_git_one(alt->base, 0);
1197 alt->name[-1] = '/';
1199 rearrange_packed_git();
1200 prepare_packed_git_run_once = 1;
1203 void reprepare_packed_git(void)
1205 discard_revindex();
1206 prepare_packed_git_run_once = 0;
1207 prepare_packed_git();
1210 static void mark_bad_packed_object(struct packed_git *p,
1211 const unsigned char *sha1)
1213 unsigned i;
1214 for (i = 0; i < p->num_bad_objects; i++)
1215 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1216 return;
1217 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1218 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1219 p->num_bad_objects++;
1222 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1224 struct packed_git *p;
1225 unsigned i;
1227 for (p = packed_git; p; p = p->next)
1228 for (i = 0; i < p->num_bad_objects; i++)
1229 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1230 return p;
1231 return NULL;
1235 * With an in-core object data in "map", rehash it to make sure the
1236 * object name actually matches "sha1" to detect object corruption.
1237 * With "map" == NULL, try reading the object named with "sha1" using
1238 * the streaming interface and rehash it to do the same.
1240 int check_sha1_signature(const unsigned char *sha1, void *map,
1241 unsigned long size, const char *type)
1243 unsigned char real_sha1[20];
1244 enum object_type obj_type;
1245 struct git_istream *st;
1246 git_SHA_CTX c;
1247 char hdr[32];
1248 int hdrlen;
1250 if (map) {
1251 hash_sha1_file(map, size, type, real_sha1);
1252 return hashcmp(sha1, real_sha1) ? -1 : 0;
1255 st = open_istream(sha1, &obj_type, &size, NULL);
1256 if (!st)
1257 return -1;
1259 /* Generate the header */
1260 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1262 /* Sha1.. */
1263 git_SHA1_Init(&c);
1264 git_SHA1_Update(&c, hdr, hdrlen);
1265 for (;;) {
1266 char buf[1024 * 16];
1267 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1269 if (readlen < 0) {
1270 close_istream(st);
1271 return -1;
1273 if (!readlen)
1274 break;
1275 git_SHA1_Update(&c, buf, readlen);
1277 git_SHA1_Final(real_sha1, &c);
1278 close_istream(st);
1279 return hashcmp(sha1, real_sha1) ? -1 : 0;
1282 static int git_open_noatime(const char *name)
1284 static int sha1_file_open_flag = O_NOATIME;
1286 for (;;) {
1287 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1288 if (fd >= 0)
1289 return fd;
1291 /* Might the failure be due to O_NOATIME? */
1292 if (errno != ENOENT && sha1_file_open_flag) {
1293 sha1_file_open_flag = 0;
1294 continue;
1297 return -1;
1301 static int open_sha1_file(const unsigned char *sha1)
1303 int fd;
1304 char *name = sha1_file_name(sha1);
1305 struct alternate_object_database *alt;
1307 fd = git_open_noatime(name);
1308 if (fd >= 0)
1309 return fd;
1311 prepare_alt_odb();
1312 errno = ENOENT;
1313 for (alt = alt_odb_list; alt; alt = alt->next) {
1314 name = alt->name;
1315 fill_sha1_path(name, sha1);
1316 fd = git_open_noatime(alt->base);
1317 if (fd >= 0)
1318 return fd;
1320 return -1;
1323 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1325 void *map;
1326 int fd;
1328 fd = open_sha1_file(sha1);
1329 map = NULL;
1330 if (fd >= 0) {
1331 struct stat st;
1333 if (!fstat(fd, &st)) {
1334 *size = xsize_t(st.st_size);
1335 if (!*size) {
1336 /* mmap() is forbidden on empty files */
1337 error("object file %s is empty", sha1_file_name(sha1));
1338 return NULL;
1340 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1342 close(fd);
1344 return map;
1348 * There used to be a second loose object header format which
1349 * was meant to mimic the in-pack format, allowing for direct
1350 * copy of the object data. This format turned up not to be
1351 * really worth it and we no longer write loose objects in that
1352 * format.
1354 static int experimental_loose_object(unsigned char *map)
1356 unsigned int word;
1359 * We must determine if the buffer contains the standard
1360 * zlib-deflated stream or the experimental format based
1361 * on the in-pack object format. Compare the header byte
1362 * for each format:
1364 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1365 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1367 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1368 * in standard loose-object format, UNLESS it is a Git-pack
1369 * format object *exactly* 8 bytes in size when inflated.
1371 * However, RFC1950 also specifies that the 1st 16-bit word
1372 * must be divisible by 31 - this checksum tells us our buffer
1373 * is in the standard format, giving a false positive only if
1374 * the 1st word of the Git-pack format object happens to be
1375 * divisible by 31, ie:
1376 * ((byte0 * 256) + byte1) % 31 = 0
1377 * => 0ttt10000www1000 % 31 = 0
1379 * As it happens, this case can only arise for www=3 & ttt=1
1380 * - ie, a Commit object, which would have to be 8 bytes in
1381 * size. As no Commit can be that small, we find that the
1382 * combination of these two criteria (bitmask & checksum)
1383 * can always correctly determine the buffer format.
1385 word = (map[0] << 8) + map[1];
1386 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1387 return 0;
1388 else
1389 return 1;
1392 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1393 unsigned long len, enum object_type *type, unsigned long *sizep)
1395 unsigned shift;
1396 unsigned long size, c;
1397 unsigned long used = 0;
1399 c = buf[used++];
1400 *type = (c >> 4) & 7;
1401 size = c & 15;
1402 shift = 4;
1403 while (c & 0x80) {
1404 if (len <= used || bitsizeof(long) <= shift) {
1405 error("bad object header");
1406 size = used = 0;
1407 break;
1409 c = buf[used++];
1410 size += (c & 0x7f) << shift;
1411 shift += 7;
1413 *sizep = size;
1414 return used;
1417 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1419 unsigned long size, used;
1420 static const char valid_loose_object_type[8] = {
1421 0, /* OBJ_EXT */
1422 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1423 0, /* "delta" and others are invalid in a loose object */
1425 enum object_type type;
1427 /* Get the data stream */
1428 memset(stream, 0, sizeof(*stream));
1429 stream->next_in = map;
1430 stream->avail_in = mapsize;
1431 stream->next_out = buffer;
1432 stream->avail_out = bufsiz;
1434 if (experimental_loose_object(map)) {
1436 * The old experimental format we no longer produce;
1437 * we can still read it.
1439 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1440 if (!used || !valid_loose_object_type[type])
1441 return -1;
1442 map += used;
1443 mapsize -= used;
1445 /* Set up the stream for the rest.. */
1446 stream->next_in = map;
1447 stream->avail_in = mapsize;
1448 git_inflate_init(stream);
1450 /* And generate the fake traditional header */
1451 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1452 typename(type), size);
1453 return 0;
1455 git_inflate_init(stream);
1456 return git_inflate(stream, 0);
1459 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1461 int bytes = strlen(buffer) + 1;
1462 unsigned char *buf = xmallocz(size);
1463 unsigned long n;
1464 int status = Z_OK;
1466 n = stream->total_out - bytes;
1467 if (n > size)
1468 n = size;
1469 memcpy(buf, (char *) buffer + bytes, n);
1470 bytes = n;
1471 if (bytes <= size) {
1473 * The above condition must be (bytes <= size), not
1474 * (bytes < size). In other words, even though we
1475 * expect no more output and set avail_out to zero,
1476 * the input zlib stream may have bytes that express
1477 * "this concludes the stream", and we *do* want to
1478 * eat that input.
1480 * Otherwise we would not be able to test that we
1481 * consumed all the input to reach the expected size;
1482 * we also want to check that zlib tells us that all
1483 * went well with status == Z_STREAM_END at the end.
1485 stream->next_out = buf + bytes;
1486 stream->avail_out = size - bytes;
1487 while (status == Z_OK)
1488 status = git_inflate(stream, Z_FINISH);
1490 if (status == Z_STREAM_END && !stream->avail_in) {
1491 git_inflate_end(stream);
1492 return buf;
1495 if (status < 0)
1496 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1497 else if (stream->avail_in)
1498 error("garbage at end of loose object '%s'",
1499 sha1_to_hex(sha1));
1500 free(buf);
1501 return NULL;
1505 * We used to just use "sscanf()", but that's actually way
1506 * too permissive for what we want to check. So do an anal
1507 * object header parse by hand.
1509 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1511 char type[10];
1512 int i;
1513 unsigned long size;
1516 * The type can be at most ten bytes (including the
1517 * terminating '\0' that we add), and is followed by
1518 * a space.
1520 i = 0;
1521 for (;;) {
1522 char c = *hdr++;
1523 if (c == ' ')
1524 break;
1525 type[i++] = c;
1526 if (i >= sizeof(type))
1527 return -1;
1529 type[i] = 0;
1532 * The length must follow immediately, and be in canonical
1533 * decimal format (ie "010" is not valid).
1535 size = *hdr++ - '0';
1536 if (size > 9)
1537 return -1;
1538 if (size) {
1539 for (;;) {
1540 unsigned long c = *hdr - '0';
1541 if (c > 9)
1542 break;
1543 hdr++;
1544 size = size * 10 + c;
1547 *sizep = size;
1550 * The length must be followed by a zero byte
1552 return *hdr ? -1 : type_from_string(type);
1555 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1557 int ret;
1558 git_zstream stream;
1559 char hdr[8192];
1561 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1562 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1563 return NULL;
1565 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1568 unsigned long get_size_from_delta(struct packed_git *p,
1569 struct pack_window **w_curs,
1570 off_t curpos)
1572 const unsigned char *data;
1573 unsigned char delta_head[20], *in;
1574 git_zstream stream;
1575 int st;
1577 memset(&stream, 0, sizeof(stream));
1578 stream.next_out = delta_head;
1579 stream.avail_out = sizeof(delta_head);
1581 git_inflate_init(&stream);
1582 do {
1583 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1584 stream.next_in = in;
1585 st = git_inflate(&stream, Z_FINISH);
1586 curpos += stream.next_in - in;
1587 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1588 stream.total_out < sizeof(delta_head));
1589 git_inflate_end(&stream);
1590 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1591 error("delta data unpack-initial failed");
1592 return 0;
1595 /* Examine the initial part of the delta to figure out
1596 * the result size.
1598 data = delta_head;
1600 /* ignore base size */
1601 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1603 /* Read the result size */
1604 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1607 static off_t get_delta_base(struct packed_git *p,
1608 struct pack_window **w_curs,
1609 off_t *curpos,
1610 enum object_type type,
1611 off_t delta_obj_offset)
1613 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1614 off_t base_offset;
1616 /* use_pack() assured us we have [base_info, base_info + 20)
1617 * as a range that we can look at without walking off the
1618 * end of the mapped window. Its actually the hash size
1619 * that is assured. An OFS_DELTA longer than the hash size
1620 * is stupid, as then a REF_DELTA would be smaller to store.
1622 if (type == OBJ_OFS_DELTA) {
1623 unsigned used = 0;
1624 unsigned char c = base_info[used++];
1625 base_offset = c & 127;
1626 while (c & 128) {
1627 base_offset += 1;
1628 if (!base_offset || MSB(base_offset, 7))
1629 return 0; /* overflow */
1630 c = base_info[used++];
1631 base_offset = (base_offset << 7) + (c & 127);
1633 base_offset = delta_obj_offset - base_offset;
1634 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1635 return 0; /* out of bound */
1636 *curpos += used;
1637 } else if (type == OBJ_REF_DELTA) {
1638 /* The base entry _must_ be in the same pack */
1639 base_offset = find_pack_entry_one(base_info, p);
1640 *curpos += 20;
1641 } else
1642 die("I am totally screwed");
1643 return base_offset;
1646 /* forward declaration for a mutually recursive function */
1647 static int packed_object_info(struct packed_git *p, off_t offset,
1648 unsigned long *sizep, int *rtype);
1650 static int packed_delta_info(struct packed_git *p,
1651 struct pack_window **w_curs,
1652 off_t curpos,
1653 enum object_type type,
1654 off_t obj_offset,
1655 unsigned long *sizep)
1657 off_t base_offset;
1659 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1660 if (!base_offset)
1661 return OBJ_BAD;
1662 type = packed_object_info(p, base_offset, NULL, NULL);
1663 if (type <= OBJ_NONE) {
1664 struct revindex_entry *revidx;
1665 const unsigned char *base_sha1;
1666 revidx = find_pack_revindex(p, base_offset);
1667 if (!revidx)
1668 return OBJ_BAD;
1669 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1670 mark_bad_packed_object(p, base_sha1);
1671 type = sha1_object_info(base_sha1, NULL);
1672 if (type <= OBJ_NONE)
1673 return OBJ_BAD;
1676 /* We choose to only get the type of the base object and
1677 * ignore potentially corrupt pack file that expects the delta
1678 * based on a base with a wrong size. This saves tons of
1679 * inflate() calls.
1681 if (sizep) {
1682 *sizep = get_size_from_delta(p, w_curs, curpos);
1683 if (*sizep == 0)
1684 type = OBJ_BAD;
1687 return type;
1690 int unpack_object_header(struct packed_git *p,
1691 struct pack_window **w_curs,
1692 off_t *curpos,
1693 unsigned long *sizep)
1695 unsigned char *base;
1696 unsigned long left;
1697 unsigned long used;
1698 enum object_type type;
1700 /* use_pack() assures us we have [base, base + 20) available
1701 * as a range that we can look at. (Its actually the hash
1702 * size that is assured.) With our object header encoding
1703 * the maximum deflated object size is 2^137, which is just
1704 * insane, so we know won't exceed what we have been given.
1706 base = use_pack(p, w_curs, *curpos, &left);
1707 used = unpack_object_header_buffer(base, left, &type, sizep);
1708 if (!used) {
1709 type = OBJ_BAD;
1710 } else
1711 *curpos += used;
1713 return type;
1716 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1717 unsigned long *sizep, int *rtype)
1719 struct pack_window *w_curs = NULL;
1720 unsigned long size;
1721 off_t curpos = obj_offset;
1722 enum object_type type;
1724 type = unpack_object_header(p, &w_curs, &curpos, &size);
1725 if (rtype)
1726 *rtype = type; /* representation type */
1728 switch (type) {
1729 case OBJ_OFS_DELTA:
1730 case OBJ_REF_DELTA:
1731 type = packed_delta_info(p, &w_curs, curpos,
1732 type, obj_offset, sizep);
1733 break;
1734 case OBJ_COMMIT:
1735 case OBJ_TREE:
1736 case OBJ_BLOB:
1737 case OBJ_TAG:
1738 if (sizep)
1739 *sizep = size;
1740 break;
1741 default:
1742 error("unknown object type %i at offset %"PRIuMAX" in %s",
1743 type, (uintmax_t)obj_offset, p->pack_name);
1744 type = OBJ_BAD;
1746 unuse_pack(&w_curs);
1747 return type;
1750 static void *unpack_compressed_entry(struct packed_git *p,
1751 struct pack_window **w_curs,
1752 off_t curpos,
1753 unsigned long size)
1755 int st;
1756 git_zstream stream;
1757 unsigned char *buffer, *in;
1759 buffer = xmallocz(size);
1760 memset(&stream, 0, sizeof(stream));
1761 stream.next_out = buffer;
1762 stream.avail_out = size + 1;
1764 git_inflate_init(&stream);
1765 do {
1766 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1767 stream.next_in = in;
1768 st = git_inflate(&stream, Z_FINISH);
1769 if (!stream.avail_out)
1770 break; /* the payload is larger than it should be */
1771 curpos += stream.next_in - in;
1772 } while (st == Z_OK || st == Z_BUF_ERROR);
1773 git_inflate_end(&stream);
1774 if ((st != Z_STREAM_END) || stream.total_out != size) {
1775 free(buffer);
1776 return NULL;
1779 return buffer;
1782 #define MAX_DELTA_CACHE (256)
1784 static size_t delta_base_cached;
1786 static struct delta_base_cache_lru_list {
1787 struct delta_base_cache_lru_list *prev;
1788 struct delta_base_cache_lru_list *next;
1789 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1791 static struct delta_base_cache_entry {
1792 struct delta_base_cache_lru_list lru;
1793 void *data;
1794 struct packed_git *p;
1795 off_t base_offset;
1796 unsigned long size;
1797 enum object_type type;
1798 } delta_base_cache[MAX_DELTA_CACHE];
1800 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1802 unsigned long hash;
1804 hash = (unsigned long)p + (unsigned long)base_offset;
1805 hash += (hash >> 8) + (hash >> 16);
1806 return hash % MAX_DELTA_CACHE;
1809 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1811 unsigned long hash = pack_entry_hash(p, base_offset);
1812 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1813 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1816 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1817 unsigned long *base_size, enum object_type *type, int keep_cache)
1819 void *ret;
1820 unsigned long hash = pack_entry_hash(p, base_offset);
1821 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1823 ret = ent->data;
1824 if (!ret || ent->p != p || ent->base_offset != base_offset)
1825 return unpack_entry(p, base_offset, type, base_size);
1827 if (!keep_cache) {
1828 ent->data = NULL;
1829 ent->lru.next->prev = ent->lru.prev;
1830 ent->lru.prev->next = ent->lru.next;
1831 delta_base_cached -= ent->size;
1832 } else {
1833 ret = xmemdupz(ent->data, ent->size);
1835 *type = ent->type;
1836 *base_size = ent->size;
1837 return ret;
1840 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1842 if (ent->data) {
1843 free(ent->data);
1844 ent->data = NULL;
1845 ent->lru.next->prev = ent->lru.prev;
1846 ent->lru.prev->next = ent->lru.next;
1847 delta_base_cached -= ent->size;
1851 void clear_delta_base_cache(void)
1853 unsigned long p;
1854 for (p = 0; p < MAX_DELTA_CACHE; p++)
1855 release_delta_base_cache(&delta_base_cache[p]);
1858 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1859 void *base, unsigned long base_size, enum object_type type)
1861 unsigned long hash = pack_entry_hash(p, base_offset);
1862 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1863 struct delta_base_cache_lru_list *lru;
1865 release_delta_base_cache(ent);
1866 delta_base_cached += base_size;
1868 for (lru = delta_base_cache_lru.next;
1869 delta_base_cached > delta_base_cache_limit
1870 && lru != &delta_base_cache_lru;
1871 lru = lru->next) {
1872 struct delta_base_cache_entry *f = (void *)lru;
1873 if (f->type == OBJ_BLOB)
1874 release_delta_base_cache(f);
1876 for (lru = delta_base_cache_lru.next;
1877 delta_base_cached > delta_base_cache_limit
1878 && lru != &delta_base_cache_lru;
1879 lru = lru->next) {
1880 struct delta_base_cache_entry *f = (void *)lru;
1881 release_delta_base_cache(f);
1884 ent->p = p;
1885 ent->base_offset = base_offset;
1886 ent->type = type;
1887 ent->data = base;
1888 ent->size = base_size;
1889 ent->lru.next = &delta_base_cache_lru;
1890 ent->lru.prev = delta_base_cache_lru.prev;
1891 delta_base_cache_lru.prev->next = &ent->lru;
1892 delta_base_cache_lru.prev = &ent->lru;
1895 static void *read_object(const unsigned char *sha1, enum object_type *type,
1896 unsigned long *size);
1898 static void *unpack_delta_entry(struct packed_git *p,
1899 struct pack_window **w_curs,
1900 off_t curpos,
1901 unsigned long delta_size,
1902 off_t obj_offset,
1903 enum object_type *type,
1904 unsigned long *sizep)
1906 void *delta_data, *result, *base;
1907 unsigned long base_size;
1908 off_t base_offset;
1910 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1911 if (!base_offset) {
1912 error("failed to validate delta base reference "
1913 "at offset %"PRIuMAX" from %s",
1914 (uintmax_t)curpos, p->pack_name);
1915 return NULL;
1917 unuse_pack(w_curs);
1918 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1919 if (!base) {
1921 * We're probably in deep shit, but let's try to fetch
1922 * the required base anyway from another pack or loose.
1923 * This is costly but should happen only in the presence
1924 * of a corrupted pack, and is better than failing outright.
1926 struct revindex_entry *revidx;
1927 const unsigned char *base_sha1;
1928 revidx = find_pack_revindex(p, base_offset);
1929 if (!revidx)
1930 return NULL;
1931 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1932 error("failed to read delta base object %s"
1933 " at offset %"PRIuMAX" from %s",
1934 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1935 p->pack_name);
1936 mark_bad_packed_object(p, base_sha1);
1937 base = read_object(base_sha1, type, &base_size);
1938 if (!base)
1939 return NULL;
1942 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1943 if (!delta_data) {
1944 error("failed to unpack compressed delta "
1945 "at offset %"PRIuMAX" from %s",
1946 (uintmax_t)curpos, p->pack_name);
1947 free(base);
1948 return NULL;
1950 result = patch_delta(base, base_size,
1951 delta_data, delta_size,
1952 sizep);
1953 if (!result)
1954 die("failed to apply delta");
1955 free(delta_data);
1956 add_delta_base_cache(p, base_offset, base, base_size, *type);
1957 return result;
1960 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1962 static FILE *log_file;
1964 if (!log_file) {
1965 log_file = fopen(log_pack_access, "w");
1966 if (!log_file) {
1967 error("cannot open pack access log '%s' for writing: %s",
1968 log_pack_access, strerror(errno));
1969 log_pack_access = NULL;
1970 return;
1973 fprintf(log_file, "%s %"PRIuMAX"\n",
1974 p->pack_name, (uintmax_t)obj_offset);
1975 fflush(log_file);
1978 int do_check_packed_object_crc;
1980 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1981 enum object_type *type, unsigned long *sizep)
1983 struct pack_window *w_curs = NULL;
1984 off_t curpos = obj_offset;
1985 void *data;
1987 if (log_pack_access)
1988 write_pack_access_log(p, obj_offset);
1990 if (do_check_packed_object_crc && p->index_version > 1) {
1991 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1992 unsigned long len = revidx[1].offset - obj_offset;
1993 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1994 const unsigned char *sha1 =
1995 nth_packed_object_sha1(p, revidx->nr);
1996 error("bad packed object CRC for %s",
1997 sha1_to_hex(sha1));
1998 mark_bad_packed_object(p, sha1);
1999 unuse_pack(&w_curs);
2000 return NULL;
2004 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
2005 switch (*type) {
2006 case OBJ_OFS_DELTA:
2007 case OBJ_REF_DELTA:
2008 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
2009 obj_offset, type, sizep);
2010 break;
2011 case OBJ_COMMIT:
2012 case OBJ_TREE:
2013 case OBJ_BLOB:
2014 case OBJ_TAG:
2015 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
2016 break;
2017 default:
2018 data = NULL;
2019 error("unknown object type %i at offset %"PRIuMAX" in %s",
2020 *type, (uintmax_t)obj_offset, p->pack_name);
2022 unuse_pack(&w_curs);
2023 return data;
2026 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2027 uint32_t n)
2029 const unsigned char *index = p->index_data;
2030 if (!index) {
2031 if (open_pack_index(p))
2032 return NULL;
2033 index = p->index_data;
2035 if (n >= p->num_objects)
2036 return NULL;
2037 index += 4 * 256;
2038 if (p->index_version == 1) {
2039 return index + 24 * n + 4;
2040 } else {
2041 index += 8;
2042 return index + 20 * n;
2046 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2048 const unsigned char *index = p->index_data;
2049 index += 4 * 256;
2050 if (p->index_version == 1) {
2051 return ntohl(*((uint32_t *)(index + 24 * n)));
2052 } else {
2053 uint32_t off;
2054 index += 8 + p->num_objects * (20 + 4);
2055 off = ntohl(*((uint32_t *)(index + 4 * n)));
2056 if (!(off & 0x80000000))
2057 return off;
2058 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2059 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2060 ntohl(*((uint32_t *)(index + 4)));
2064 off_t find_pack_entry_one(const unsigned char *sha1,
2065 struct packed_git *p)
2067 const uint32_t *level1_ofs = p->index_data;
2068 const unsigned char *index = p->index_data;
2069 unsigned hi, lo, stride;
2070 static int use_lookup = -1;
2071 static int debug_lookup = -1;
2073 if (debug_lookup < 0)
2074 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2076 if (!index) {
2077 if (open_pack_index(p))
2078 return 0;
2079 level1_ofs = p->index_data;
2080 index = p->index_data;
2082 if (p->index_version > 1) {
2083 level1_ofs += 2;
2084 index += 8;
2086 index += 4 * 256;
2087 hi = ntohl(level1_ofs[*sha1]);
2088 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2089 if (p->index_version > 1) {
2090 stride = 20;
2091 } else {
2092 stride = 24;
2093 index += 4;
2096 if (debug_lookup)
2097 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2098 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2100 if (use_lookup < 0)
2101 use_lookup = !!getenv("GIT_USE_LOOKUP");
2102 if (use_lookup) {
2103 int pos = sha1_entry_pos(index, stride, 0,
2104 lo, hi, p->num_objects, sha1);
2105 if (pos < 0)
2106 return 0;
2107 return nth_packed_object_offset(p, pos);
2110 do {
2111 unsigned mi = (lo + hi) / 2;
2112 int cmp = hashcmp(index + mi * stride, sha1);
2114 if (debug_lookup)
2115 printf("lo %u hi %u rg %u mi %u\n",
2116 lo, hi, hi - lo, mi);
2117 if (!cmp)
2118 return nth_packed_object_offset(p, mi);
2119 if (cmp > 0)
2120 hi = mi;
2121 else
2122 lo = mi+1;
2123 } while (lo < hi);
2124 return 0;
2127 int is_pack_valid(struct packed_git *p)
2129 /* An already open pack is known to be valid. */
2130 if (p->pack_fd != -1)
2131 return 1;
2133 /* If the pack has one window completely covering the
2134 * file size, the pack is known to be valid even if
2135 * the descriptor is not currently open.
2137 if (p->windows) {
2138 struct pack_window *w = p->windows;
2140 if (!w->offset && w->len == p->pack_size)
2141 return 1;
2144 /* Force the pack to open to prove its valid. */
2145 return !open_packed_git(p);
2148 static int fill_pack_entry(const unsigned char *sha1,
2149 struct pack_entry *e,
2150 struct packed_git *p)
2152 off_t offset;
2154 if (p->num_bad_objects) {
2155 unsigned i;
2156 for (i = 0; i < p->num_bad_objects; i++)
2157 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2158 return 0;
2161 offset = find_pack_entry_one(sha1, p);
2162 if (!offset)
2163 return 0;
2166 * We are about to tell the caller where they can locate the
2167 * requested object. We better make sure the packfile is
2168 * still here and can be accessed before supplying that
2169 * answer, as it may have been deleted since the index was
2170 * loaded!
2172 if (!is_pack_valid(p)) {
2173 warning("packfile %s cannot be accessed", p->pack_name);
2174 return 0;
2176 e->offset = offset;
2177 e->p = p;
2178 hashcpy(e->sha1, sha1);
2179 return 1;
2182 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2184 struct packed_git *p;
2186 prepare_packed_git();
2187 if (!packed_git)
2188 return 0;
2190 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2191 return 1;
2193 for (p = packed_git; p; p = p->next) {
2194 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2195 continue;
2197 last_found_pack = p;
2198 return 1;
2200 return 0;
2203 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2204 struct packed_git *packs)
2206 struct packed_git *p;
2208 for (p = packs; p; p = p->next) {
2209 if (find_pack_entry_one(sha1, p))
2210 return p;
2212 return NULL;
2216 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2218 int status;
2219 unsigned long mapsize, size;
2220 void *map;
2221 git_zstream stream;
2222 char hdr[32];
2224 map = map_sha1_file(sha1, &mapsize);
2225 if (!map)
2226 return error("unable to find %s", sha1_to_hex(sha1));
2227 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2228 status = error("unable to unpack %s header",
2229 sha1_to_hex(sha1));
2230 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2231 status = error("unable to parse %s header", sha1_to_hex(sha1));
2232 else if (sizep)
2233 *sizep = size;
2234 git_inflate_end(&stream);
2235 munmap(map, mapsize);
2236 return status;
2239 /* returns enum object_type or negative */
2240 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2242 struct cached_object *co;
2243 struct pack_entry e;
2244 int status, rtype;
2246 co = find_cached_object(sha1);
2247 if (co) {
2248 if (oi->sizep)
2249 *(oi->sizep) = co->size;
2250 oi->whence = OI_CACHED;
2251 return co->type;
2254 if (!find_pack_entry(sha1, &e)) {
2255 /* Most likely it's a loose object. */
2256 status = sha1_loose_object_info(sha1, oi->sizep);
2257 if (status >= 0) {
2258 oi->whence = OI_LOOSE;
2259 return status;
2262 /* Not a loose object; someone else may have just packed it. */
2263 reprepare_packed_git();
2264 if (!find_pack_entry(sha1, &e))
2265 return status;
2268 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2269 if (status < 0) {
2270 mark_bad_packed_object(e.p, sha1);
2271 status = sha1_object_info_extended(sha1, oi);
2272 } else if (in_delta_base_cache(e.p, e.offset)) {
2273 oi->whence = OI_DBCACHED;
2274 } else {
2275 oi->whence = OI_PACKED;
2276 oi->u.packed.offset = e.offset;
2277 oi->u.packed.pack = e.p;
2278 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2279 rtype == OBJ_OFS_DELTA);
2282 return status;
2285 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2287 struct object_info oi;
2289 oi.sizep = sizep;
2290 return sha1_object_info_extended(sha1, &oi);
2293 static void *read_packed_sha1(const unsigned char *sha1,
2294 enum object_type *type, unsigned long *size)
2296 struct pack_entry e;
2297 void *data;
2299 if (!find_pack_entry(sha1, &e))
2300 return NULL;
2301 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2302 if (!data) {
2304 * We're probably in deep shit, but let's try to fetch
2305 * the required object anyway from another pack or loose.
2306 * This should happen only in the presence of a corrupted
2307 * pack, and is better than failing outright.
2309 error("failed to read object %s at offset %"PRIuMAX" from %s",
2310 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2311 mark_bad_packed_object(e.p, sha1);
2312 data = read_object(sha1, type, size);
2314 return data;
2317 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2318 unsigned char *sha1)
2320 struct cached_object *co;
2322 hash_sha1_file(buf, len, typename(type), sha1);
2323 if (has_sha1_file(sha1) || find_cached_object(sha1))
2324 return 0;
2325 if (cached_object_alloc <= cached_object_nr) {
2326 cached_object_alloc = alloc_nr(cached_object_alloc);
2327 cached_objects = xrealloc(cached_objects,
2328 sizeof(*cached_objects) *
2329 cached_object_alloc);
2331 co = &cached_objects[cached_object_nr++];
2332 co->size = len;
2333 co->type = type;
2334 co->buf = xmalloc(len);
2335 memcpy(co->buf, buf, len);
2336 hashcpy(co->sha1, sha1);
2337 return 0;
2340 static void *read_object(const unsigned char *sha1, enum object_type *type,
2341 unsigned long *size)
2343 unsigned long mapsize;
2344 void *map, *buf;
2345 struct cached_object *co;
2347 co = find_cached_object(sha1);
2348 if (co) {
2349 *type = co->type;
2350 *size = co->size;
2351 return xmemdupz(co->buf, co->size);
2354 buf = read_packed_sha1(sha1, type, size);
2355 if (buf)
2356 return buf;
2357 map = map_sha1_file(sha1, &mapsize);
2358 if (map) {
2359 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2360 munmap(map, mapsize);
2361 return buf;
2363 reprepare_packed_git();
2364 return read_packed_sha1(sha1, type, size);
2368 * This function dies on corrupt objects; the callers who want to
2369 * deal with them should arrange to call read_object() and give error
2370 * messages themselves.
2372 void *read_sha1_file_extended(const unsigned char *sha1,
2373 enum object_type *type,
2374 unsigned long *size,
2375 unsigned flag)
2377 void *data;
2378 char *path;
2379 const struct packed_git *p;
2380 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2381 ? lookup_replace_object(sha1) : sha1;
2383 errno = 0;
2384 data = read_object(repl, type, size);
2385 if (data)
2386 return data;
2388 if (errno && errno != ENOENT)
2389 die_errno("failed to read object %s", sha1_to_hex(sha1));
2391 /* die if we replaced an object with one that does not exist */
2392 if (repl != sha1)
2393 die("replacement %s not found for %s",
2394 sha1_to_hex(repl), sha1_to_hex(sha1));
2396 if (has_loose_object(repl)) {
2397 path = sha1_file_name(sha1);
2398 die("loose object %s (stored in %s) is corrupt",
2399 sha1_to_hex(repl), path);
2402 if ((p = has_packed_and_bad(repl)) != NULL)
2403 die("packed object %s (stored in %s) is corrupt",
2404 sha1_to_hex(repl), p->pack_name);
2406 return NULL;
2409 void *read_object_with_reference(const unsigned char *sha1,
2410 const char *required_type_name,
2411 unsigned long *size,
2412 unsigned char *actual_sha1_return)
2414 enum object_type type, required_type;
2415 void *buffer;
2416 unsigned long isize;
2417 unsigned char actual_sha1[20];
2419 required_type = type_from_string(required_type_name);
2420 hashcpy(actual_sha1, sha1);
2421 while (1) {
2422 int ref_length = -1;
2423 const char *ref_type = NULL;
2425 buffer = read_sha1_file(actual_sha1, &type, &isize);
2426 if (!buffer)
2427 return NULL;
2428 if (type == required_type) {
2429 *size = isize;
2430 if (actual_sha1_return)
2431 hashcpy(actual_sha1_return, actual_sha1);
2432 return buffer;
2434 /* Handle references */
2435 else if (type == OBJ_COMMIT)
2436 ref_type = "tree ";
2437 else if (type == OBJ_TAG)
2438 ref_type = "object ";
2439 else {
2440 free(buffer);
2441 return NULL;
2443 ref_length = strlen(ref_type);
2445 if (ref_length + 40 > isize ||
2446 memcmp(buffer, ref_type, ref_length) ||
2447 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2448 free(buffer);
2449 return NULL;
2451 free(buffer);
2452 /* Now we have the ID of the referred-to object in
2453 * actual_sha1. Check again. */
2457 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2458 const char *type, unsigned char *sha1,
2459 char *hdr, int *hdrlen)
2461 git_SHA_CTX c;
2463 /* Generate the header */
2464 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2466 /* Sha1.. */
2467 git_SHA1_Init(&c);
2468 git_SHA1_Update(&c, hdr, *hdrlen);
2469 git_SHA1_Update(&c, buf, len);
2470 git_SHA1_Final(sha1, &c);
2474 * Move the just written object into its final resting place.
2475 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2476 * "moving" is only a part of what it does, when no patch between
2477 * master to pu changes the call sites of this function.
2479 int move_temp_to_file(const char *tmpfile, const char *filename)
2481 int ret = 0;
2483 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2484 goto try_rename;
2485 else if (link(tmpfile, filename))
2486 ret = errno;
2489 * Coda hack - coda doesn't like cross-directory links,
2490 * so we fall back to a rename, which will mean that it
2491 * won't be able to check collisions, but that's not a
2492 * big deal.
2494 * The same holds for FAT formatted media.
2496 * When this succeeds, we just return. We have nothing
2497 * left to unlink.
2499 if (ret && ret != EEXIST) {
2500 try_rename:
2501 if (!rename(tmpfile, filename))
2502 goto out;
2503 ret = errno;
2505 unlink_or_warn(tmpfile);
2506 if (ret) {
2507 if (ret != EEXIST) {
2508 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2510 /* FIXME!!! Collision check here ? */
2513 out:
2514 if (adjust_shared_perm(filename))
2515 return error("unable to set permission to '%s'", filename);
2516 return 0;
2519 static int write_buffer(int fd, const void *buf, size_t len)
2521 if (write_in_full(fd, buf, len) < 0)
2522 return error("file write error (%s)", strerror(errno));
2523 return 0;
2526 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2527 unsigned char *sha1)
2529 char hdr[32];
2530 int hdrlen;
2531 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2532 return 0;
2535 /* Finalize a file on disk, and close it. */
2536 static void close_sha1_file(int fd)
2538 if (fsync_object_files)
2539 fsync_or_die(fd, "sha1 file");
2540 if (close(fd) != 0)
2541 die_errno("error when closing sha1 file");
2544 /* Size of directory component, including the ending '/' */
2545 static inline int directory_size(const char *filename)
2547 const char *s = strrchr(filename, '/');
2548 if (!s)
2549 return 0;
2550 return s - filename + 1;
2554 * This creates a temporary file in the same directory as the final
2555 * 'filename'
2557 * We want to avoid cross-directory filename renames, because those
2558 * can have problems on various filesystems (FAT, NFS, Coda).
2560 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2562 int fd, dirlen = directory_size(filename);
2564 if (dirlen + 20 > bufsiz) {
2565 errno = ENAMETOOLONG;
2566 return -1;
2568 memcpy(buffer, filename, dirlen);
2569 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2570 fd = git_mkstemp_mode(buffer, 0444);
2571 if (fd < 0 && dirlen && errno == ENOENT) {
2572 /* Make sure the directory exists */
2573 memcpy(buffer, filename, dirlen);
2574 buffer[dirlen-1] = 0;
2575 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2576 return -1;
2578 /* Try again */
2579 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2580 fd = git_mkstemp_mode(buffer, 0444);
2582 return fd;
2585 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2586 const void *buf, unsigned long len, time_t mtime)
2588 int fd, ret;
2589 unsigned char compressed[4096];
2590 git_zstream stream;
2591 git_SHA_CTX c;
2592 unsigned char parano_sha1[20];
2593 char *filename;
2594 static char tmp_file[PATH_MAX];
2596 filename = sha1_file_name(sha1);
2597 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2598 if (fd < 0) {
2599 if (errno == EACCES)
2600 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2601 else
2602 return error("unable to create temporary file: %s", strerror(errno));
2605 /* Set it up */
2606 memset(&stream, 0, sizeof(stream));
2607 git_deflate_init(&stream, zlib_compression_level);
2608 stream.next_out = compressed;
2609 stream.avail_out = sizeof(compressed);
2610 git_SHA1_Init(&c);
2612 /* First header.. */
2613 stream.next_in = (unsigned char *)hdr;
2614 stream.avail_in = hdrlen;
2615 while (git_deflate(&stream, 0) == Z_OK)
2616 ; /* nothing */
2617 git_SHA1_Update(&c, hdr, hdrlen);
2619 /* Then the data itself.. */
2620 stream.next_in = (void *)buf;
2621 stream.avail_in = len;
2622 do {
2623 unsigned char *in0 = stream.next_in;
2624 ret = git_deflate(&stream, Z_FINISH);
2625 git_SHA1_Update(&c, in0, stream.next_in - in0);
2626 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2627 die("unable to write sha1 file");
2628 stream.next_out = compressed;
2629 stream.avail_out = sizeof(compressed);
2630 } while (ret == Z_OK);
2632 if (ret != Z_STREAM_END)
2633 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2634 ret = git_deflate_end_gently(&stream);
2635 if (ret != Z_OK)
2636 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2637 git_SHA1_Final(parano_sha1, &c);
2638 if (hashcmp(sha1, parano_sha1) != 0)
2639 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2641 close_sha1_file(fd);
2643 if (mtime) {
2644 struct utimbuf utb;
2645 utb.actime = mtime;
2646 utb.modtime = mtime;
2647 if (utime(tmp_file, &utb) < 0)
2648 warning("failed utime() on %s: %s",
2649 tmp_file, strerror(errno));
2652 return move_temp_to_file(tmp_file, filename);
2655 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2657 unsigned char sha1[20];
2658 char hdr[32];
2659 int hdrlen;
2661 /* Normally if we have it in the pack then we do not bother writing
2662 * it out into .git/objects/??/?{38} file.
2664 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2665 if (returnsha1)
2666 hashcpy(returnsha1, sha1);
2667 if (has_sha1_file(sha1))
2668 return 0;
2669 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2672 int force_object_loose(const unsigned char *sha1, time_t mtime)
2674 void *buf;
2675 unsigned long len;
2676 enum object_type type;
2677 char hdr[32];
2678 int hdrlen;
2679 int ret;
2681 if (has_loose_object(sha1))
2682 return 0;
2683 buf = read_packed_sha1(sha1, &type, &len);
2684 if (!buf)
2685 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2686 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2687 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2688 free(buf);
2690 return ret;
2693 int has_pack_index(const unsigned char *sha1)
2695 struct stat st;
2696 if (stat(sha1_pack_index_name(sha1), &st))
2697 return 0;
2698 return 1;
2701 int has_sha1_pack(const unsigned char *sha1)
2703 struct pack_entry e;
2704 return find_pack_entry(sha1, &e);
2707 int has_sha1_file(const unsigned char *sha1)
2709 struct pack_entry e;
2711 if (find_pack_entry(sha1, &e))
2712 return 1;
2713 return has_loose_object(sha1);
2716 static void check_tree(const void *buf, size_t size)
2718 struct tree_desc desc;
2719 struct name_entry entry;
2721 init_tree_desc(&desc, buf, size);
2722 while (tree_entry(&desc, &entry))
2723 /* do nothing
2724 * tree_entry() will die() on malformed entries */
2728 static void check_commit(const void *buf, size_t size)
2730 struct commit c;
2731 memset(&c, 0, sizeof(c));
2732 if (parse_commit_buffer(&c, buf, size))
2733 die("corrupt commit");
2736 static void check_tag(const void *buf, size_t size)
2738 struct tag t;
2739 memset(&t, 0, sizeof(t));
2740 if (parse_tag_buffer(&t, buf, size))
2741 die("corrupt tag");
2744 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2745 enum object_type type,
2746 const char *path, unsigned flags)
2748 int ret, re_allocated = 0;
2749 int write_object = flags & HASH_WRITE_OBJECT;
2751 if (!type)
2752 type = OBJ_BLOB;
2755 * Convert blobs to git internal format
2757 if ((type == OBJ_BLOB) && path) {
2758 struct strbuf nbuf = STRBUF_INIT;
2759 if (convert_to_git(path, buf, size, &nbuf,
2760 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2761 buf = strbuf_detach(&nbuf, &size);
2762 re_allocated = 1;
2765 if (flags & HASH_FORMAT_CHECK) {
2766 if (type == OBJ_TREE)
2767 check_tree(buf, size);
2768 if (type == OBJ_COMMIT)
2769 check_commit(buf, size);
2770 if (type == OBJ_TAG)
2771 check_tag(buf, size);
2774 if (write_object)
2775 ret = write_sha1_file(buf, size, typename(type), sha1);
2776 else
2777 ret = hash_sha1_file(buf, size, typename(type), sha1);
2778 if (re_allocated)
2779 free(buf);
2780 return ret;
2783 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2784 const char *path, unsigned flags)
2786 struct strbuf sbuf = STRBUF_INIT;
2787 int ret;
2789 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2790 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2791 else
2792 ret = -1;
2793 strbuf_release(&sbuf);
2794 return ret;
2797 #define SMALL_FILE_SIZE (32*1024)
2799 static int index_core(unsigned char *sha1, int fd, size_t size,
2800 enum object_type type, const char *path,
2801 unsigned flags)
2803 int ret;
2805 if (!size) {
2806 ret = index_mem(sha1, NULL, size, type, path, flags);
2807 } else if (size <= SMALL_FILE_SIZE) {
2808 char *buf = xmalloc(size);
2809 if (size == read_in_full(fd, buf, size))
2810 ret = index_mem(sha1, buf, size, type, path, flags);
2811 else
2812 ret = error("short read %s", strerror(errno));
2813 free(buf);
2814 } else {
2815 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2816 ret = index_mem(sha1, buf, size, type, path, flags);
2817 munmap(buf, size);
2819 return ret;
2823 * This creates one packfile per large blob unless bulk-checkin
2824 * machinery is "plugged".
2826 * This also bypasses the usual "convert-to-git" dance, and that is on
2827 * purpose. We could write a streaming version of the converting
2828 * functions and insert that before feeding the data to fast-import
2829 * (or equivalent in-core API described above). However, that is
2830 * somewhat complicated, as we do not know the size of the filter
2831 * result, which we need to know beforehand when writing a git object.
2832 * Since the primary motivation for trying to stream from the working
2833 * tree file and to avoid mmaping it in core is to deal with large
2834 * binary blobs, they generally do not want to get any conversion, and
2835 * callers should avoid this code path when filters are requested.
2837 static int index_stream(unsigned char *sha1, int fd, size_t size,
2838 enum object_type type, const char *path,
2839 unsigned flags)
2841 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2844 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2845 enum object_type type, const char *path, unsigned flags)
2847 int ret;
2848 size_t size = xsize_t(st->st_size);
2850 if (!S_ISREG(st->st_mode))
2851 ret = index_pipe(sha1, fd, type, path, flags);
2852 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2853 (path && would_convert_to_git(path, NULL, 0, 0)))
2854 ret = index_core(sha1, fd, size, type, path, flags);
2855 else
2856 ret = index_stream(sha1, fd, size, type, path, flags);
2857 close(fd);
2858 return ret;
2861 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2863 int fd;
2864 struct strbuf sb = STRBUF_INIT;
2866 switch (st->st_mode & S_IFMT) {
2867 case S_IFREG:
2868 fd = open(path, O_RDONLY);
2869 if (fd < 0)
2870 return error("open(\"%s\"): %s", path,
2871 strerror(errno));
2872 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2873 return error("%s: failed to insert into database",
2874 path);
2875 break;
2876 case S_IFLNK:
2877 if (strbuf_readlink(&sb, path, st->st_size)) {
2878 char *errstr = strerror(errno);
2879 return error("readlink(\"%s\"): %s", path,
2880 errstr);
2882 if (!(flags & HASH_WRITE_OBJECT))
2883 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2884 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2885 return error("%s: failed to insert into database",
2886 path);
2887 strbuf_release(&sb);
2888 break;
2889 case S_IFDIR:
2890 return resolve_gitlink_ref(path, "HEAD", sha1);
2891 default:
2892 return error("%s: unsupported file type", path);
2894 return 0;
2897 int read_pack_header(int fd, struct pack_header *header)
2899 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2900 /* "eof before pack header was fully read" */
2901 return PH_ERROR_EOF;
2903 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2904 /* "protocol error (pack signature mismatch detected)" */
2905 return PH_ERROR_PACK_SIGNATURE;
2906 if (!pack_version_ok(header->hdr_version))
2907 /* "protocol error (pack version unsupported)" */
2908 return PH_ERROR_PROTOCOL;
2909 return 0;
2912 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2914 enum object_type type = sha1_object_info(sha1, NULL);
2915 if (type < 0)
2916 die("%s is not a valid object", sha1_to_hex(sha1));
2917 if (type != expect)
2918 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2919 typename(expect));