Merge branch 'kk/revwalk-slop-too-many-commit-within-a-second' into maint-1.8.1
[git/jnareb-git.git] / sha1_file.c
blob40b23297b2e1e60a3719e9c67256303e39456604
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"
25 #ifndef O_NOATIME
26 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
27 #define O_NOATIME 01000000
28 #else
29 #define O_NOATIME 0
30 #endif
31 #endif
33 #define SZ_FMT PRIuMAX
34 static inline uintmax_t sz_fmt(size_t s) { return s; }
36 const unsigned char null_sha1[20];
39 * This is meant to hold a *small* number of objects that you would
40 * want read_sha1_file() to be able to return, but yet you do not want
41 * to write them into the object store (e.g. a browse-only
42 * application).
44 static struct cached_object {
45 unsigned char sha1[20];
46 enum object_type type;
47 void *buf;
48 unsigned long size;
49 } *cached_objects;
50 static int cached_object_nr, cached_object_alloc;
52 static struct cached_object empty_tree = {
53 EMPTY_TREE_SHA1_BIN_LITERAL,
54 OBJ_TREE,
55 "",
59 static struct packed_git *last_found_pack;
61 static struct cached_object *find_cached_object(const unsigned char *sha1)
63 int i;
64 struct cached_object *co = cached_objects;
66 for (i = 0; i < cached_object_nr; i++, co++) {
67 if (!hashcmp(co->sha1, sha1))
68 return co;
70 if (!hashcmp(sha1, empty_tree.sha1))
71 return &empty_tree;
72 return NULL;
75 int mkdir_in_gitdir(const char *path)
77 if (mkdir(path, 0777)) {
78 int saved_errno = errno;
79 struct stat st;
80 struct strbuf sb = STRBUF_INIT;
82 if (errno != EEXIST)
83 return -1;
85 * Are we looking at a path in a symlinked worktree
86 * whose original repository does not yet have it?
87 * e.g. .git/rr-cache pointing at its original
88 * repository in which the user hasn't performed any
89 * conflict resolution yet?
91 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
92 strbuf_readlink(&sb, path, st.st_size) ||
93 !is_absolute_path(sb.buf) ||
94 mkdir(sb.buf, 0777)) {
95 strbuf_release(&sb);
96 errno = saved_errno;
97 return -1;
99 strbuf_release(&sb);
101 return adjust_shared_perm(path);
104 int safe_create_leading_directories(char *path)
106 char *pos = path + offset_1st_component(path);
107 struct stat st;
109 while (pos) {
110 pos = strchr(pos, '/');
111 if (!pos)
112 break;
113 while (*++pos == '/')
115 if (!*pos)
116 break;
117 *--pos = '\0';
118 if (!stat(path, &st)) {
119 /* path exists */
120 if (!S_ISDIR(st.st_mode)) {
121 *pos = '/';
122 return -3;
125 else if (mkdir(path, 0777)) {
126 *pos = '/';
127 return -1;
129 else if (adjust_shared_perm(path)) {
130 *pos = '/';
131 return -2;
133 *pos++ = '/';
135 return 0;
138 int safe_create_leading_directories_const(const char *path)
140 /* path points to cache entries, so xstrdup before messing with it */
141 char *buf = xstrdup(path);
142 int result = safe_create_leading_directories(buf);
143 free(buf);
144 return result;
147 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
149 int i;
150 for (i = 0; i < 20; i++) {
151 static char hex[] = "0123456789abcdef";
152 unsigned int val = sha1[i];
153 char *pos = pathbuf + i*2 + (i > 0);
154 *pos++ = hex[val >> 4];
155 *pos = hex[val & 0xf];
160 * NOTE! This returns a statically allocated buffer, so you have to be
161 * careful about using it. Do an "xstrdup()" if you need to save the
162 * filename.
164 * Also note that this returns the location for creating. Reading
165 * SHA1 file can happen from any alternate directory listed in the
166 * DB_ENVIRONMENT environment variable if it is not found in
167 * the primary object database.
169 char *sha1_file_name(const unsigned char *sha1)
171 static char buf[PATH_MAX];
172 const char *objdir;
173 int len;
175 objdir = get_object_directory();
176 len = strlen(objdir);
178 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
179 if (len + 43 > PATH_MAX)
180 die("insanely long object directory %s", objdir);
181 memcpy(buf, objdir, len);
182 buf[len] = '/';
183 buf[len+3] = '/';
184 buf[len+42] = '\0';
185 fill_sha1_path(buf + len + 1, sha1);
186 return buf;
189 static char *sha1_get_pack_name(const unsigned char *sha1,
190 char **name, char **base, const char *which)
192 static const char hex[] = "0123456789abcdef";
193 char *buf;
194 int i;
196 if (!*base) {
197 const char *sha1_file_directory = get_object_directory();
198 int len = strlen(sha1_file_directory);
199 *base = xmalloc(len + 60);
200 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
201 sha1_file_directory, which);
202 *name = *base + len + 11;
205 buf = *name;
207 for (i = 0; i < 20; i++) {
208 unsigned int val = *sha1++;
209 *buf++ = hex[val >> 4];
210 *buf++ = hex[val & 0xf];
213 return *base;
216 char *sha1_pack_name(const unsigned char *sha1)
218 static char *name, *base;
220 return sha1_get_pack_name(sha1, &name, &base, "pack");
223 char *sha1_pack_index_name(const unsigned char *sha1)
225 static char *name, *base;
227 return sha1_get_pack_name(sha1, &name, &base, "idx");
230 struct alternate_object_database *alt_odb_list;
231 static struct alternate_object_database **alt_odb_tail;
233 static int git_open_noatime(const char *name);
236 * Prepare alternate object database registry.
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
250 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
252 const char *objdir = get_object_directory();
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
258 if (!is_absolute_path(entry) && relative_base) {
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
262 strbuf_addstr(&pathbuf, entry);
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
266 pfxlen = strlen(pathbuf.buf);
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284 /* Detect cases where alternate disappeared */
285 if (!is_directory(ent->base)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (!memcmp(ent->base, alt->base, pfxlen)) {
298 free(ent);
299 return -1;
302 if (!strcmp(ent->base, objdir)) {
303 free(ent);
304 return -1;
307 /* add the alternate entry */
308 *alt_odb_tail = ent;
309 alt_odb_tail = &(ent->next);
310 ent->next = NULL;
312 /* recursively add alternates */
313 read_info_alternates(ent->base, depth + 1);
315 ent->base[pfxlen] = '/';
317 return 0;
320 static void link_alt_odb_entries(const char *alt, int len, int sep,
321 const char *relative_base, int depth)
323 struct string_list entries = STRING_LIST_INIT_NODUP;
324 char *alt_copy;
325 int i;
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
333 alt_copy = xmemdupz(alt, len);
334 string_list_split_in_place(&entries, alt_copy, sep, -1);
335 for (i = 0; i < entries.nr; i++) {
336 const char *entry = entries.items[i].string;
337 if (entry[0] == '\0' || entry[0] == '#')
338 continue;
339 if (!is_absolute_path(entry) && depth) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, entry);
342 } else {
343 link_alt_odb_entry(entry, relative_base, depth);
346 string_list_clear(&entries, 0);
347 free(alt_copy);
350 void read_info_alternates(const char * relative_base, int depth)
352 char *map;
353 size_t mapsz;
354 struct stat st;
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
360 int fd;
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = git_open_noatime(path);
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
368 return;
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
372 close(fd);
374 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
376 munmap(map, mapsz);
379 void add_to_alternates_file(const char *reference)
381 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
382 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
383 char *alt = mkpath("%s\n", reference);
384 write_or_die(fd, alt, strlen(alt));
385 if (commit_lock_file(lock))
386 die("could not close alternates file");
387 if (alt_odb_tail)
388 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
391 void foreach_alt_odb(alt_odb_fn fn, void *cb)
393 struct alternate_object_database *ent;
395 prepare_alt_odb();
396 for (ent = alt_odb_list; ent; ent = ent->next)
397 if (fn(ent, cb))
398 return;
401 void prepare_alt_odb(void)
403 const char *alt;
405 if (alt_odb_tail)
406 return;
408 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409 if (!alt) alt = "";
411 alt_odb_tail = &alt_odb_list;
412 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
414 read_info_alternates(get_object_directory(), 0);
417 static int has_loose_object_local(const unsigned char *sha1)
419 char *name = sha1_file_name(sha1);
420 return !access(name, F_OK);
423 int has_loose_object_nonlocal(const unsigned char *sha1)
425 struct alternate_object_database *alt;
426 prepare_alt_odb();
427 for (alt = alt_odb_list; alt; alt = alt->next) {
428 fill_sha1_path(alt->name, sha1);
429 if (!access(alt->base, F_OK))
430 return 1;
432 return 0;
435 static int has_loose_object(const unsigned char *sha1)
437 return has_loose_object_local(sha1) ||
438 has_loose_object_nonlocal(sha1);
441 static unsigned int pack_used_ctr;
442 static unsigned int pack_mmap_calls;
443 static unsigned int peak_pack_open_windows;
444 static unsigned int pack_open_windows;
445 static unsigned int pack_open_fds;
446 static unsigned int pack_max_fds;
447 static size_t peak_pack_mapped;
448 static size_t pack_mapped;
449 struct packed_git *packed_git;
451 void pack_report(void)
453 fprintf(stderr,
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
460 fprintf(stderr,
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
466 pack_used_ctr,
467 pack_mmap_calls,
468 pack_open_windows, peak_pack_open_windows,
469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
472 static int check_packed_git_idx(const char *path, struct packed_git *p)
474 void *idx_map;
475 struct pack_idx_header *hdr;
476 size_t idx_size;
477 uint32_t version, nr, i, *index;
478 int fd = git_open_noatime(path);
479 struct stat st;
481 if (fd < 0)
482 return -1;
483 if (fstat(fd, &st)) {
484 close(fd);
485 return -1;
487 idx_size = xsize_t(st.st_size);
488 if (idx_size < 4 * 256 + 20 + 20) {
489 close(fd);
490 return error("index file %s is too small", path);
492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
493 close(fd);
495 hdr = idx_map;
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
503 path, version);
505 } else
506 version = 1;
508 nr = 0;
509 index = idx_map;
510 if (version > 1)
511 index += 2; /* skip index header */
512 for (i = 0; i < 256; i++) {
513 uint32_t n = ntohl(index[i]);
514 if (n < nr) {
515 munmap(idx_map, idx_size);
516 return error("non-monotonic index %s", path);
518 nr = n;
521 if (version == 1) {
523 * Total size:
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
531 return error("wrong index v1 file size in %s", path);
533 } else if (version == 2) {
535 * Minimum size:
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size = min_size;
549 if (nr)
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
552 munmap(idx_map, idx_size);
553 return error("wrong index v2 file size in %s", path);
555 if (idx_size != min_size &&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
567 p->index_version = version;
568 p->index_data = idx_map;
569 p->index_size = idx_size;
570 p->num_objects = nr;
571 return 0;
574 int open_pack_index(struct packed_git *p)
576 char *idx_name;
577 int ret;
579 if (p->index_data)
580 return 0;
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
585 free(idx_name);
586 return ret;
589 static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
594 struct pack_window *w, *w_l;
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
597 if (!w->inuse_cnt) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599 *lru_p = p;
600 *lru_w = w;
601 *lru_l = w_l;
604 w_l = w;
608 static int unuse_one_window(struct packed_git *current, int keep_fd)
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
613 if (current)
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
617 if (lru_p) {
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
620 if (lru_l)
621 lru_l->next = lru_w->next;
622 else {
623 lru_p->windows = lru_w->next;
624 if (!lru_p->windows && lru_p->pack_fd != -1
625 && lru_p->pack_fd != keep_fd) {
626 close(lru_p->pack_fd);
627 pack_open_fds--;
628 lru_p->pack_fd = -1;
631 free(lru_w);
632 pack_open_windows--;
633 return 1;
635 return 0;
638 void release_pack_memory(size_t need, int fd)
640 size_t cur = pack_mapped;
641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
642 ; /* nothing */
645 void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
650 if (!length)
651 return NULL;
652 release_pack_memory(length, fd);
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
657 return ret;
660 void close_pack_windows(struct packed_git *p)
662 while (p->windows) {
663 struct pack_window *w = p->windows;
665 if (w->inuse_cnt)
666 die("pack '%s' still has open windows to it",
667 p->pack_name);
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
670 pack_open_windows--;
671 p->windows = w->next;
672 free(w);
676 void unuse_pack(struct pack_window **w_cursor)
678 struct pack_window *w = *w_cursor;
679 if (w) {
680 w->inuse_cnt--;
681 *w_cursor = NULL;
685 void close_pack_index(struct packed_git *p)
687 if (p->index_data) {
688 munmap((void *)p->index_data, p->index_size);
689 p->index_data = NULL;
694 * This is used by git-repack in case a newly created pack happens to
695 * contain the same set of objects as an existing one. In that case
696 * the resulting file might be different even if its name would be the
697 * same. It is best to close any reference to the old pack before it is
698 * replaced on disk. Of course no index pointers nor windows for given pack
699 * must subsist at this point. If ever objects from this pack are requested
700 * again, the new version of the pack will be reinitialized through
701 * reprepare_packed_git().
703 void free_pack_by_name(const char *pack_name)
705 struct packed_git *p, **pp = &packed_git;
707 while (*pp) {
708 p = *pp;
709 if (strcmp(pack_name, p->pack_name) == 0) {
710 clear_delta_base_cache();
711 close_pack_windows(p);
712 if (p->pack_fd != -1) {
713 close(p->pack_fd);
714 pack_open_fds--;
716 close_pack_index(p);
717 free(p->bad_object_sha1);
718 *pp = p->next;
719 if (last_found_pack == p)
720 last_found_pack = NULL;
721 free(p);
722 return;
724 pp = &p->next;
728 static unsigned int get_max_fd_limit(void)
730 #ifdef RLIMIT_NOFILE
731 struct rlimit lim;
733 if (getrlimit(RLIMIT_NOFILE, &lim))
734 die_errno("cannot get RLIMIT_NOFILE");
736 return lim.rlim_cur;
737 #elif defined(_SC_OPEN_MAX)
738 return sysconf(_SC_OPEN_MAX);
739 #elif defined(OPEN_MAX)
740 return OPEN_MAX;
741 #else
742 return 1; /* see the caller ;-) */
743 #endif
747 * Do not call this directly as this leaks p->pack_fd on error return;
748 * call open_packed_git() instead.
750 static int open_packed_git_1(struct packed_git *p)
752 struct stat st;
753 struct pack_header hdr;
754 unsigned char sha1[20];
755 unsigned char *idx_sha1;
756 long fd_flag;
758 if (!p->index_data && open_pack_index(p))
759 return error("packfile %s index unavailable", p->pack_name);
761 if (!pack_max_fds) {
762 unsigned int max_fds = get_max_fd_limit();
764 /* Save 3 for stdin/stdout/stderr, 22 for work */
765 if (25 < max_fds)
766 pack_max_fds = max_fds - 25;
767 else
768 pack_max_fds = 1;
771 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
772 ; /* nothing */
774 p->pack_fd = git_open_noatime(p->pack_name);
775 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
776 return -1;
777 pack_open_fds++;
779 /* If we created the struct before we had the pack we lack size. */
780 if (!p->pack_size) {
781 if (!S_ISREG(st.st_mode))
782 return error("packfile %s not a regular file", p->pack_name);
783 p->pack_size = st.st_size;
784 } else if (p->pack_size != st.st_size)
785 return error("packfile %s size changed", p->pack_name);
787 /* We leave these file descriptors open with sliding mmap;
788 * there is no point keeping them open across exec(), though.
790 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
791 if (fd_flag < 0)
792 return error("cannot determine file descriptor flags");
793 fd_flag |= FD_CLOEXEC;
794 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
795 return error("cannot set FD_CLOEXEC");
797 /* Verify we recognize this pack file format. */
798 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
799 return error("file %s is far too short to be a packfile", p->pack_name);
800 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
801 return error("file %s is not a GIT packfile", p->pack_name);
802 if (!pack_version_ok(hdr.hdr_version))
803 return error("packfile %s is version %"PRIu32" and not"
804 " supported (try upgrading GIT to a newer version)",
805 p->pack_name, ntohl(hdr.hdr_version));
807 /* Verify the pack matches its index. */
808 if (p->num_objects != ntohl(hdr.hdr_entries))
809 return error("packfile %s claims to have %"PRIu32" objects"
810 " while index indicates %"PRIu32" objects",
811 p->pack_name, ntohl(hdr.hdr_entries),
812 p->num_objects);
813 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
814 return error("end of packfile %s is unavailable", p->pack_name);
815 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
816 return error("packfile %s signature is unavailable", p->pack_name);
817 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
818 if (hashcmp(sha1, idx_sha1))
819 return error("packfile %s does not match index", p->pack_name);
820 return 0;
823 static int open_packed_git(struct packed_git *p)
825 if (!open_packed_git_1(p))
826 return 0;
827 if (p->pack_fd != -1) {
828 close(p->pack_fd);
829 pack_open_fds--;
830 p->pack_fd = -1;
832 return -1;
835 static int in_window(struct pack_window *win, off_t offset)
837 /* We must promise at least 20 bytes (one hash) after the
838 * offset is available from this window, otherwise the offset
839 * is not actually in this window and a different window (which
840 * has that one hash excess) must be used. This is to support
841 * the object header and delta base parsing routines below.
843 off_t win_off = win->offset;
844 return win_off <= offset
845 && (offset + 20) <= (win_off + win->len);
848 unsigned char *use_pack(struct packed_git *p,
849 struct pack_window **w_cursor,
850 off_t offset,
851 unsigned long *left)
853 struct pack_window *win = *w_cursor;
855 /* Since packfiles end in a hash of their content and it's
856 * pointless to ask for an offset into the middle of that
857 * hash, and the in_window function above wouldn't match
858 * don't allow an offset too close to the end of the file.
860 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
861 die("packfile %s cannot be accessed", p->pack_name);
862 if (offset > (p->pack_size - 20))
863 die("offset beyond end of packfile (truncated pack?)");
865 if (!win || !in_window(win, offset)) {
866 if (win)
867 win->inuse_cnt--;
868 for (win = p->windows; win; win = win->next) {
869 if (in_window(win, offset))
870 break;
872 if (!win) {
873 size_t window_align = packed_git_window_size / 2;
874 off_t len;
876 if (p->pack_fd == -1 && open_packed_git(p))
877 die("packfile %s cannot be accessed", p->pack_name);
879 win = xcalloc(1, sizeof(*win));
880 win->offset = (offset / window_align) * window_align;
881 len = p->pack_size - win->offset;
882 if (len > packed_git_window_size)
883 len = packed_git_window_size;
884 win->len = (size_t)len;
885 pack_mapped += win->len;
886 while (packed_git_limit < pack_mapped
887 && unuse_one_window(p, p->pack_fd))
888 ; /* nothing */
889 win->base = xmmap(NULL, win->len,
890 PROT_READ, MAP_PRIVATE,
891 p->pack_fd, win->offset);
892 if (win->base == MAP_FAILED)
893 die("packfile %s cannot be mapped: %s",
894 p->pack_name,
895 strerror(errno));
896 if (!win->offset && win->len == p->pack_size
897 && !p->do_not_close) {
898 close(p->pack_fd);
899 pack_open_fds--;
900 p->pack_fd = -1;
902 pack_mmap_calls++;
903 pack_open_windows++;
904 if (pack_mapped > peak_pack_mapped)
905 peak_pack_mapped = pack_mapped;
906 if (pack_open_windows > peak_pack_open_windows)
907 peak_pack_open_windows = pack_open_windows;
908 win->next = p->windows;
909 p->windows = win;
912 if (win != *w_cursor) {
913 win->last_used = pack_used_ctr++;
914 win->inuse_cnt++;
915 *w_cursor = win;
917 offset -= win->offset;
918 if (left)
919 *left = win->len - xsize_t(offset);
920 return win->base + offset;
923 static struct packed_git *alloc_packed_git(int extra)
925 struct packed_git *p = xmalloc(sizeof(*p) + extra);
926 memset(p, 0, sizeof(*p));
927 p->pack_fd = -1;
928 return p;
931 static void try_to_free_pack_memory(size_t size)
933 release_pack_memory(size, -1);
936 struct packed_git *add_packed_git(const char *path, int path_len, int local)
938 static int have_set_try_to_free_routine;
939 struct stat st;
940 struct packed_git *p = alloc_packed_git(path_len + 2);
942 if (!have_set_try_to_free_routine) {
943 have_set_try_to_free_routine = 1;
944 set_try_to_free_routine(try_to_free_pack_memory);
948 * Make sure a corresponding .pack file exists and that
949 * the index looks sane.
951 path_len -= strlen(".idx");
952 if (path_len < 1) {
953 free(p);
954 return NULL;
956 memcpy(p->pack_name, path, path_len);
958 strcpy(p->pack_name + path_len, ".keep");
959 if (!access(p->pack_name, F_OK))
960 p->pack_keep = 1;
962 strcpy(p->pack_name + path_len, ".pack");
963 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
964 free(p);
965 return NULL;
968 /* ok, it looks sane as far as we can check without
969 * actually mapping the pack file.
971 p->pack_size = st.st_size;
972 p->pack_local = local;
973 p->mtime = st.st_mtime;
974 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
975 hashclr(p->sha1);
976 return p;
979 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
981 const char *path = sha1_pack_name(sha1);
982 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
984 strcpy(p->pack_name, path);
985 hashcpy(p->sha1, sha1);
986 if (check_packed_git_idx(idx_path, p)) {
987 free(p);
988 return NULL;
991 return p;
994 void install_packed_git(struct packed_git *pack)
996 if (pack->pack_fd != -1)
997 pack_open_fds++;
999 pack->next = packed_git;
1000 packed_git = pack;
1003 static void prepare_packed_git_one(char *objdir, int local)
1005 /* Ensure that this buffer is large enough so that we can
1006 append "/pack/" without clobbering the stack even if
1007 strlen(objdir) were PATH_MAX. */
1008 char path[PATH_MAX + 1 + 4 + 1 + 1];
1009 int len;
1010 DIR *dir;
1011 struct dirent *de;
1013 sprintf(path, "%s/pack", objdir);
1014 len = strlen(path);
1015 dir = opendir(path);
1016 if (!dir) {
1017 if (errno != ENOENT)
1018 error("unable to open object pack directory: %s: %s",
1019 path, strerror(errno));
1020 return;
1022 path[len++] = '/';
1023 while ((de = readdir(dir)) != NULL) {
1024 int namelen = strlen(de->d_name);
1025 struct packed_git *p;
1027 if (!has_extension(de->d_name, ".idx"))
1028 continue;
1030 if (len + namelen + 1 > sizeof(path))
1031 continue;
1033 /* Don't reopen a pack we already have. */
1034 strcpy(path + len, de->d_name);
1035 for (p = packed_git; p; p = p->next) {
1036 if (!memcmp(path, p->pack_name, len + namelen - 4))
1037 break;
1039 if (p)
1040 continue;
1041 /* See if it really is a valid .idx file with corresponding
1042 * .pack file that we can map.
1044 p = add_packed_git(path, len + namelen, local);
1045 if (!p)
1046 continue;
1047 install_packed_git(p);
1049 closedir(dir);
1052 static int sort_pack(const void *a_, const void *b_)
1054 struct packed_git *a = *((struct packed_git **)a_);
1055 struct packed_git *b = *((struct packed_git **)b_);
1056 int st;
1059 * Local packs tend to contain objects specific to our
1060 * variant of the project than remote ones. In addition,
1061 * remote ones could be on a network mounted filesystem.
1062 * Favor local ones for these reasons.
1064 st = a->pack_local - b->pack_local;
1065 if (st)
1066 return -st;
1069 * Younger packs tend to contain more recent objects,
1070 * and more recent objects tend to get accessed more
1071 * often.
1073 if (a->mtime < b->mtime)
1074 return 1;
1075 else if (a->mtime == b->mtime)
1076 return 0;
1077 return -1;
1080 static void rearrange_packed_git(void)
1082 struct packed_git **ary, *p;
1083 int i, n;
1085 for (n = 0, p = packed_git; p; p = p->next)
1086 n++;
1087 if (n < 2)
1088 return;
1090 /* prepare an array of packed_git for easier sorting */
1091 ary = xcalloc(n, sizeof(struct packed_git *));
1092 for (n = 0, p = packed_git; p; p = p->next)
1093 ary[n++] = p;
1095 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1097 /* link them back again */
1098 for (i = 0; i < n - 1; i++)
1099 ary[i]->next = ary[i + 1];
1100 ary[n - 1]->next = NULL;
1101 packed_git = ary[0];
1103 free(ary);
1106 static int prepare_packed_git_run_once = 0;
1107 void prepare_packed_git(void)
1109 struct alternate_object_database *alt;
1111 if (prepare_packed_git_run_once)
1112 return;
1113 prepare_packed_git_one(get_object_directory(), 1);
1114 prepare_alt_odb();
1115 for (alt = alt_odb_list; alt; alt = alt->next) {
1116 alt->name[-1] = 0;
1117 prepare_packed_git_one(alt->base, 0);
1118 alt->name[-1] = '/';
1120 rearrange_packed_git();
1121 prepare_packed_git_run_once = 1;
1124 void reprepare_packed_git(void)
1126 discard_revindex();
1127 prepare_packed_git_run_once = 0;
1128 prepare_packed_git();
1131 static void mark_bad_packed_object(struct packed_git *p,
1132 const unsigned char *sha1)
1134 unsigned i;
1135 for (i = 0; i < p->num_bad_objects; i++)
1136 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1137 return;
1138 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1139 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1140 p->num_bad_objects++;
1143 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1145 struct packed_git *p;
1146 unsigned i;
1148 for (p = packed_git; p; p = p->next)
1149 for (i = 0; i < p->num_bad_objects; i++)
1150 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1151 return p;
1152 return NULL;
1156 * With an in-core object data in "map", rehash it to make sure the
1157 * object name actually matches "sha1" to detect object corruption.
1158 * With "map" == NULL, try reading the object named with "sha1" using
1159 * the streaming interface and rehash it to do the same.
1161 int check_sha1_signature(const unsigned char *sha1, void *map,
1162 unsigned long size, const char *type)
1164 unsigned char real_sha1[20];
1165 enum object_type obj_type;
1166 struct git_istream *st;
1167 git_SHA_CTX c;
1168 char hdr[32];
1169 int hdrlen;
1171 if (map) {
1172 hash_sha1_file(map, size, type, real_sha1);
1173 return hashcmp(sha1, real_sha1) ? -1 : 0;
1176 st = open_istream(sha1, &obj_type, &size, NULL);
1177 if (!st)
1178 return -1;
1180 /* Generate the header */
1181 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1183 /* Sha1.. */
1184 git_SHA1_Init(&c);
1185 git_SHA1_Update(&c, hdr, hdrlen);
1186 for (;;) {
1187 char buf[1024 * 16];
1188 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1190 if (!readlen)
1191 break;
1192 git_SHA1_Update(&c, buf, readlen);
1194 git_SHA1_Final(real_sha1, &c);
1195 close_istream(st);
1196 return hashcmp(sha1, real_sha1) ? -1 : 0;
1199 static int git_open_noatime(const char *name)
1201 static int sha1_file_open_flag = O_NOATIME;
1203 for (;;) {
1204 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1205 if (fd >= 0)
1206 return fd;
1208 /* Might the failure be due to O_NOATIME? */
1209 if (errno != ENOENT && sha1_file_open_flag) {
1210 sha1_file_open_flag = 0;
1211 continue;
1214 return -1;
1218 static int open_sha1_file(const unsigned char *sha1)
1220 int fd;
1221 char *name = sha1_file_name(sha1);
1222 struct alternate_object_database *alt;
1224 fd = git_open_noatime(name);
1225 if (fd >= 0)
1226 return fd;
1228 prepare_alt_odb();
1229 errno = ENOENT;
1230 for (alt = alt_odb_list; alt; alt = alt->next) {
1231 name = alt->name;
1232 fill_sha1_path(name, sha1);
1233 fd = git_open_noatime(alt->base);
1234 if (fd >= 0)
1235 return fd;
1237 return -1;
1240 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1242 void *map;
1243 int fd;
1245 fd = open_sha1_file(sha1);
1246 map = NULL;
1247 if (fd >= 0) {
1248 struct stat st;
1250 if (!fstat(fd, &st)) {
1251 *size = xsize_t(st.st_size);
1252 if (!*size) {
1253 /* mmap() is forbidden on empty files */
1254 error("object file %s is empty", sha1_file_name(sha1));
1255 return NULL;
1257 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1259 close(fd);
1261 return map;
1265 * There used to be a second loose object header format which
1266 * was meant to mimic the in-pack format, allowing for direct
1267 * copy of the object data. This format turned up not to be
1268 * really worth it and we no longer write loose objects in that
1269 * format.
1271 static int experimental_loose_object(unsigned char *map)
1273 unsigned int word;
1276 * We must determine if the buffer contains the standard
1277 * zlib-deflated stream or the experimental format based
1278 * on the in-pack object format. Compare the header byte
1279 * for each format:
1281 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1282 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1284 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1285 * in standard loose-object format, UNLESS it is a Git-pack
1286 * format object *exactly* 8 bytes in size when inflated.
1288 * However, RFC1950 also specifies that the 1st 16-bit word
1289 * must be divisible by 31 - this checksum tells us our buffer
1290 * is in the standard format, giving a false positive only if
1291 * the 1st word of the Git-pack format object happens to be
1292 * divisible by 31, ie:
1293 * ((byte0 * 256) + byte1) % 31 = 0
1294 * => 0ttt10000www1000 % 31 = 0
1296 * As it happens, this case can only arise for www=3 & ttt=1
1297 * - ie, a Commit object, which would have to be 8 bytes in
1298 * size. As no Commit can be that small, we find that the
1299 * combination of these two criteria (bitmask & checksum)
1300 * can always correctly determine the buffer format.
1302 word = (map[0] << 8) + map[1];
1303 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1304 return 0;
1305 else
1306 return 1;
1309 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1310 unsigned long len, enum object_type *type, unsigned long *sizep)
1312 unsigned shift;
1313 unsigned long size, c;
1314 unsigned long used = 0;
1316 c = buf[used++];
1317 *type = (c >> 4) & 7;
1318 size = c & 15;
1319 shift = 4;
1320 while (c & 0x80) {
1321 if (len <= used || bitsizeof(long) <= shift) {
1322 error("bad object header");
1323 size = used = 0;
1324 break;
1326 c = buf[used++];
1327 size += (c & 0x7f) << shift;
1328 shift += 7;
1330 *sizep = size;
1331 return used;
1334 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1336 unsigned long size, used;
1337 static const char valid_loose_object_type[8] = {
1338 0, /* OBJ_EXT */
1339 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1340 0, /* "delta" and others are invalid in a loose object */
1342 enum object_type type;
1344 /* Get the data stream */
1345 memset(stream, 0, sizeof(*stream));
1346 stream->next_in = map;
1347 stream->avail_in = mapsize;
1348 stream->next_out = buffer;
1349 stream->avail_out = bufsiz;
1351 if (experimental_loose_object(map)) {
1353 * The old experimental format we no longer produce;
1354 * we can still read it.
1356 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1357 if (!used || !valid_loose_object_type[type])
1358 return -1;
1359 map += used;
1360 mapsize -= used;
1362 /* Set up the stream for the rest.. */
1363 stream->next_in = map;
1364 stream->avail_in = mapsize;
1365 git_inflate_init(stream);
1367 /* And generate the fake traditional header */
1368 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1369 typename(type), size);
1370 return 0;
1372 git_inflate_init(stream);
1373 return git_inflate(stream, 0);
1376 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1378 int bytes = strlen(buffer) + 1;
1379 unsigned char *buf = xmallocz(size);
1380 unsigned long n;
1381 int status = Z_OK;
1383 n = stream->total_out - bytes;
1384 if (n > size)
1385 n = size;
1386 memcpy(buf, (char *) buffer + bytes, n);
1387 bytes = n;
1388 if (bytes <= size) {
1390 * The above condition must be (bytes <= size), not
1391 * (bytes < size). In other words, even though we
1392 * expect no more output and set avail_out to zero,
1393 * the input zlib stream may have bytes that express
1394 * "this concludes the stream", and we *do* want to
1395 * eat that input.
1397 * Otherwise we would not be able to test that we
1398 * consumed all the input to reach the expected size;
1399 * we also want to check that zlib tells us that all
1400 * went well with status == Z_STREAM_END at the end.
1402 stream->next_out = buf + bytes;
1403 stream->avail_out = size - bytes;
1404 while (status == Z_OK)
1405 status = git_inflate(stream, Z_FINISH);
1407 if (status == Z_STREAM_END && !stream->avail_in) {
1408 git_inflate_end(stream);
1409 return buf;
1412 if (status < 0)
1413 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1414 else if (stream->avail_in)
1415 error("garbage at end of loose object '%s'",
1416 sha1_to_hex(sha1));
1417 free(buf);
1418 return NULL;
1422 * We used to just use "sscanf()", but that's actually way
1423 * too permissive for what we want to check. So do an anal
1424 * object header parse by hand.
1426 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1428 char type[10];
1429 int i;
1430 unsigned long size;
1433 * The type can be at most ten bytes (including the
1434 * terminating '\0' that we add), and is followed by
1435 * a space.
1437 i = 0;
1438 for (;;) {
1439 char c = *hdr++;
1440 if (c == ' ')
1441 break;
1442 type[i++] = c;
1443 if (i >= sizeof(type))
1444 return -1;
1446 type[i] = 0;
1449 * The length must follow immediately, and be in canonical
1450 * decimal format (ie "010" is not valid).
1452 size = *hdr++ - '0';
1453 if (size > 9)
1454 return -1;
1455 if (size) {
1456 for (;;) {
1457 unsigned long c = *hdr - '0';
1458 if (c > 9)
1459 break;
1460 hdr++;
1461 size = size * 10 + c;
1464 *sizep = size;
1467 * The length must be followed by a zero byte
1469 return *hdr ? -1 : type_from_string(type);
1472 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1474 int ret;
1475 git_zstream stream;
1476 char hdr[8192];
1478 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1479 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1480 return NULL;
1482 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1485 unsigned long get_size_from_delta(struct packed_git *p,
1486 struct pack_window **w_curs,
1487 off_t curpos)
1489 const unsigned char *data;
1490 unsigned char delta_head[20], *in;
1491 git_zstream stream;
1492 int st;
1494 memset(&stream, 0, sizeof(stream));
1495 stream.next_out = delta_head;
1496 stream.avail_out = sizeof(delta_head);
1498 git_inflate_init(&stream);
1499 do {
1500 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1501 stream.next_in = in;
1502 st = git_inflate(&stream, Z_FINISH);
1503 curpos += stream.next_in - in;
1504 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1505 stream.total_out < sizeof(delta_head));
1506 git_inflate_end(&stream);
1507 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1508 error("delta data unpack-initial failed");
1509 return 0;
1512 /* Examine the initial part of the delta to figure out
1513 * the result size.
1515 data = delta_head;
1517 /* ignore base size */
1518 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1520 /* Read the result size */
1521 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1524 static off_t get_delta_base(struct packed_git *p,
1525 struct pack_window **w_curs,
1526 off_t *curpos,
1527 enum object_type type,
1528 off_t delta_obj_offset)
1530 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1531 off_t base_offset;
1533 /* use_pack() assured us we have [base_info, base_info + 20)
1534 * as a range that we can look at without walking off the
1535 * end of the mapped window. Its actually the hash size
1536 * that is assured. An OFS_DELTA longer than the hash size
1537 * is stupid, as then a REF_DELTA would be smaller to store.
1539 if (type == OBJ_OFS_DELTA) {
1540 unsigned used = 0;
1541 unsigned char c = base_info[used++];
1542 base_offset = c & 127;
1543 while (c & 128) {
1544 base_offset += 1;
1545 if (!base_offset || MSB(base_offset, 7))
1546 return 0; /* overflow */
1547 c = base_info[used++];
1548 base_offset = (base_offset << 7) + (c & 127);
1550 base_offset = delta_obj_offset - base_offset;
1551 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1552 return 0; /* out of bound */
1553 *curpos += used;
1554 } else if (type == OBJ_REF_DELTA) {
1555 /* The base entry _must_ be in the same pack */
1556 base_offset = find_pack_entry_one(base_info, p);
1557 *curpos += 20;
1558 } else
1559 die("I am totally screwed");
1560 return base_offset;
1563 /* forward declaration for a mutually recursive function */
1564 static int packed_object_info(struct packed_git *p, off_t offset,
1565 unsigned long *sizep, int *rtype);
1567 static int packed_delta_info(struct packed_git *p,
1568 struct pack_window **w_curs,
1569 off_t curpos,
1570 enum object_type type,
1571 off_t obj_offset,
1572 unsigned long *sizep)
1574 off_t base_offset;
1576 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1577 if (!base_offset)
1578 return OBJ_BAD;
1579 type = packed_object_info(p, base_offset, NULL, NULL);
1580 if (type <= OBJ_NONE) {
1581 struct revindex_entry *revidx;
1582 const unsigned char *base_sha1;
1583 revidx = find_pack_revindex(p, base_offset);
1584 if (!revidx)
1585 return OBJ_BAD;
1586 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1587 mark_bad_packed_object(p, base_sha1);
1588 type = sha1_object_info(base_sha1, NULL);
1589 if (type <= OBJ_NONE)
1590 return OBJ_BAD;
1593 /* We choose to only get the type of the base object and
1594 * ignore potentially corrupt pack file that expects the delta
1595 * based on a base with a wrong size. This saves tons of
1596 * inflate() calls.
1598 if (sizep) {
1599 *sizep = get_size_from_delta(p, w_curs, curpos);
1600 if (*sizep == 0)
1601 type = OBJ_BAD;
1604 return type;
1607 int unpack_object_header(struct packed_git *p,
1608 struct pack_window **w_curs,
1609 off_t *curpos,
1610 unsigned long *sizep)
1612 unsigned char *base;
1613 unsigned long left;
1614 unsigned long used;
1615 enum object_type type;
1617 /* use_pack() assures us we have [base, base + 20) available
1618 * as a range that we can look at. (Its actually the hash
1619 * size that is assured.) With our object header encoding
1620 * the maximum deflated object size is 2^137, which is just
1621 * insane, so we know won't exceed what we have been given.
1623 base = use_pack(p, w_curs, *curpos, &left);
1624 used = unpack_object_header_buffer(base, left, &type, sizep);
1625 if (!used) {
1626 type = OBJ_BAD;
1627 } else
1628 *curpos += used;
1630 return type;
1633 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1634 unsigned long *sizep, int *rtype)
1636 struct pack_window *w_curs = NULL;
1637 unsigned long size;
1638 off_t curpos = obj_offset;
1639 enum object_type type;
1641 type = unpack_object_header(p, &w_curs, &curpos, &size);
1642 if (rtype)
1643 *rtype = type; /* representation type */
1645 switch (type) {
1646 case OBJ_OFS_DELTA:
1647 case OBJ_REF_DELTA:
1648 type = packed_delta_info(p, &w_curs, curpos,
1649 type, obj_offset, sizep);
1650 break;
1651 case OBJ_COMMIT:
1652 case OBJ_TREE:
1653 case OBJ_BLOB:
1654 case OBJ_TAG:
1655 if (sizep)
1656 *sizep = size;
1657 break;
1658 default:
1659 error("unknown object type %i at offset %"PRIuMAX" in %s",
1660 type, (uintmax_t)obj_offset, p->pack_name);
1661 type = OBJ_BAD;
1663 unuse_pack(&w_curs);
1664 return type;
1667 static void *unpack_compressed_entry(struct packed_git *p,
1668 struct pack_window **w_curs,
1669 off_t curpos,
1670 unsigned long size)
1672 int st;
1673 git_zstream stream;
1674 unsigned char *buffer, *in;
1676 buffer = xmallocz(size);
1677 memset(&stream, 0, sizeof(stream));
1678 stream.next_out = buffer;
1679 stream.avail_out = size + 1;
1681 git_inflate_init(&stream);
1682 do {
1683 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1684 stream.next_in = in;
1685 st = git_inflate(&stream, Z_FINISH);
1686 if (!stream.avail_out)
1687 break; /* the payload is larger than it should be */
1688 curpos += stream.next_in - in;
1689 } while (st == Z_OK || st == Z_BUF_ERROR);
1690 git_inflate_end(&stream);
1691 if ((st != Z_STREAM_END) || stream.total_out != size) {
1692 free(buffer);
1693 return NULL;
1696 return buffer;
1699 #define MAX_DELTA_CACHE (256)
1701 static size_t delta_base_cached;
1703 static struct delta_base_cache_lru_list {
1704 struct delta_base_cache_lru_list *prev;
1705 struct delta_base_cache_lru_list *next;
1706 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1708 static struct delta_base_cache_entry {
1709 struct delta_base_cache_lru_list lru;
1710 void *data;
1711 struct packed_git *p;
1712 off_t base_offset;
1713 unsigned long size;
1714 enum object_type type;
1715 } delta_base_cache[MAX_DELTA_CACHE];
1717 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1719 unsigned long hash;
1721 hash = (unsigned long)p + (unsigned long)base_offset;
1722 hash += (hash >> 8) + (hash >> 16);
1723 return hash % MAX_DELTA_CACHE;
1726 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1728 unsigned long hash = pack_entry_hash(p, base_offset);
1729 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1730 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1733 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1734 unsigned long *base_size, enum object_type *type, int keep_cache)
1736 void *ret;
1737 unsigned long hash = pack_entry_hash(p, base_offset);
1738 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1740 ret = ent->data;
1741 if (!ret || ent->p != p || ent->base_offset != base_offset)
1742 return unpack_entry(p, base_offset, type, base_size);
1744 if (!keep_cache) {
1745 ent->data = NULL;
1746 ent->lru.next->prev = ent->lru.prev;
1747 ent->lru.prev->next = ent->lru.next;
1748 delta_base_cached -= ent->size;
1749 } else {
1750 ret = xmemdupz(ent->data, ent->size);
1752 *type = ent->type;
1753 *base_size = ent->size;
1754 return ret;
1757 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1759 if (ent->data) {
1760 free(ent->data);
1761 ent->data = NULL;
1762 ent->lru.next->prev = ent->lru.prev;
1763 ent->lru.prev->next = ent->lru.next;
1764 delta_base_cached -= ent->size;
1768 void clear_delta_base_cache(void)
1770 unsigned long p;
1771 for (p = 0; p < MAX_DELTA_CACHE; p++)
1772 release_delta_base_cache(&delta_base_cache[p]);
1775 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1776 void *base, unsigned long base_size, enum object_type type)
1778 unsigned long hash = pack_entry_hash(p, base_offset);
1779 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1780 struct delta_base_cache_lru_list *lru;
1782 release_delta_base_cache(ent);
1783 delta_base_cached += base_size;
1785 for (lru = delta_base_cache_lru.next;
1786 delta_base_cached > delta_base_cache_limit
1787 && lru != &delta_base_cache_lru;
1788 lru = lru->next) {
1789 struct delta_base_cache_entry *f = (void *)lru;
1790 if (f->type == OBJ_BLOB)
1791 release_delta_base_cache(f);
1793 for (lru = delta_base_cache_lru.next;
1794 delta_base_cached > delta_base_cache_limit
1795 && lru != &delta_base_cache_lru;
1796 lru = lru->next) {
1797 struct delta_base_cache_entry *f = (void *)lru;
1798 release_delta_base_cache(f);
1801 ent->p = p;
1802 ent->base_offset = base_offset;
1803 ent->type = type;
1804 ent->data = base;
1805 ent->size = base_size;
1806 ent->lru.next = &delta_base_cache_lru;
1807 ent->lru.prev = delta_base_cache_lru.prev;
1808 delta_base_cache_lru.prev->next = &ent->lru;
1809 delta_base_cache_lru.prev = &ent->lru;
1812 static void *read_object(const unsigned char *sha1, enum object_type *type,
1813 unsigned long *size);
1815 static void *unpack_delta_entry(struct packed_git *p,
1816 struct pack_window **w_curs,
1817 off_t curpos,
1818 unsigned long delta_size,
1819 off_t obj_offset,
1820 enum object_type *type,
1821 unsigned long *sizep)
1823 void *delta_data, *result, *base;
1824 unsigned long base_size;
1825 off_t base_offset;
1827 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1828 if (!base_offset) {
1829 error("failed to validate delta base reference "
1830 "at offset %"PRIuMAX" from %s",
1831 (uintmax_t)curpos, p->pack_name);
1832 return NULL;
1834 unuse_pack(w_curs);
1835 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1836 if (!base) {
1838 * We're probably in deep shit, but let's try to fetch
1839 * the required base anyway from another pack or loose.
1840 * This is costly but should happen only in the presence
1841 * of a corrupted pack, and is better than failing outright.
1843 struct revindex_entry *revidx;
1844 const unsigned char *base_sha1;
1845 revidx = find_pack_revindex(p, base_offset);
1846 if (!revidx)
1847 return NULL;
1848 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1849 error("failed to read delta base object %s"
1850 " at offset %"PRIuMAX" from %s",
1851 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1852 p->pack_name);
1853 mark_bad_packed_object(p, base_sha1);
1854 base = read_object(base_sha1, type, &base_size);
1855 if (!base)
1856 return NULL;
1859 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1860 if (!delta_data) {
1861 error("failed to unpack compressed delta "
1862 "at offset %"PRIuMAX" from %s",
1863 (uintmax_t)curpos, p->pack_name);
1864 free(base);
1865 return NULL;
1867 result = patch_delta(base, base_size,
1868 delta_data, delta_size,
1869 sizep);
1870 if (!result)
1871 die("failed to apply delta");
1872 free(delta_data);
1873 add_delta_base_cache(p, base_offset, base, base_size, *type);
1874 return result;
1877 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1879 static FILE *log_file;
1881 if (!log_file) {
1882 log_file = fopen(log_pack_access, "w");
1883 if (!log_file) {
1884 error("cannot open pack access log '%s' for writing: %s",
1885 log_pack_access, strerror(errno));
1886 log_pack_access = NULL;
1887 return;
1890 fprintf(log_file, "%s %"PRIuMAX"\n",
1891 p->pack_name, (uintmax_t)obj_offset);
1892 fflush(log_file);
1895 int do_check_packed_object_crc;
1897 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1898 enum object_type *type, unsigned long *sizep)
1900 struct pack_window *w_curs = NULL;
1901 off_t curpos = obj_offset;
1902 void *data;
1904 if (log_pack_access)
1905 write_pack_access_log(p, obj_offset);
1907 if (do_check_packed_object_crc && p->index_version > 1) {
1908 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1909 unsigned long len = revidx[1].offset - obj_offset;
1910 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1911 const unsigned char *sha1 =
1912 nth_packed_object_sha1(p, revidx->nr);
1913 error("bad packed object CRC for %s",
1914 sha1_to_hex(sha1));
1915 mark_bad_packed_object(p, sha1);
1916 unuse_pack(&w_curs);
1917 return NULL;
1921 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1922 switch (*type) {
1923 case OBJ_OFS_DELTA:
1924 case OBJ_REF_DELTA:
1925 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1926 obj_offset, type, sizep);
1927 break;
1928 case OBJ_COMMIT:
1929 case OBJ_TREE:
1930 case OBJ_BLOB:
1931 case OBJ_TAG:
1932 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1933 break;
1934 default:
1935 data = NULL;
1936 error("unknown object type %i at offset %"PRIuMAX" in %s",
1937 *type, (uintmax_t)obj_offset, p->pack_name);
1939 unuse_pack(&w_curs);
1940 return data;
1943 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1944 uint32_t n)
1946 const unsigned char *index = p->index_data;
1947 if (!index) {
1948 if (open_pack_index(p))
1949 return NULL;
1950 index = p->index_data;
1952 if (n >= p->num_objects)
1953 return NULL;
1954 index += 4 * 256;
1955 if (p->index_version == 1) {
1956 return index + 24 * n + 4;
1957 } else {
1958 index += 8;
1959 return index + 20 * n;
1963 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1965 const unsigned char *index = p->index_data;
1966 index += 4 * 256;
1967 if (p->index_version == 1) {
1968 return ntohl(*((uint32_t *)(index + 24 * n)));
1969 } else {
1970 uint32_t off;
1971 index += 8 + p->num_objects * (20 + 4);
1972 off = ntohl(*((uint32_t *)(index + 4 * n)));
1973 if (!(off & 0x80000000))
1974 return off;
1975 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1976 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1977 ntohl(*((uint32_t *)(index + 4)));
1981 off_t find_pack_entry_one(const unsigned char *sha1,
1982 struct packed_git *p)
1984 const uint32_t *level1_ofs = p->index_data;
1985 const unsigned char *index = p->index_data;
1986 unsigned hi, lo, stride;
1987 static int use_lookup = -1;
1988 static int debug_lookup = -1;
1990 if (debug_lookup < 0)
1991 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1993 if (!index) {
1994 if (open_pack_index(p))
1995 return 0;
1996 level1_ofs = p->index_data;
1997 index = p->index_data;
1999 if (p->index_version > 1) {
2000 level1_ofs += 2;
2001 index += 8;
2003 index += 4 * 256;
2004 hi = ntohl(level1_ofs[*sha1]);
2005 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2006 if (p->index_version > 1) {
2007 stride = 20;
2008 } else {
2009 stride = 24;
2010 index += 4;
2013 if (debug_lookup)
2014 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2015 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2017 if (use_lookup < 0)
2018 use_lookup = !!getenv("GIT_USE_LOOKUP");
2019 if (use_lookup) {
2020 int pos = sha1_entry_pos(index, stride, 0,
2021 lo, hi, p->num_objects, sha1);
2022 if (pos < 0)
2023 return 0;
2024 return nth_packed_object_offset(p, pos);
2027 do {
2028 unsigned mi = (lo + hi) / 2;
2029 int cmp = hashcmp(index + mi * stride, sha1);
2031 if (debug_lookup)
2032 printf("lo %u hi %u rg %u mi %u\n",
2033 lo, hi, hi - lo, mi);
2034 if (!cmp)
2035 return nth_packed_object_offset(p, mi);
2036 if (cmp > 0)
2037 hi = mi;
2038 else
2039 lo = mi+1;
2040 } while (lo < hi);
2041 return 0;
2044 int is_pack_valid(struct packed_git *p)
2046 /* An already open pack is known to be valid. */
2047 if (p->pack_fd != -1)
2048 return 1;
2050 /* If the pack has one window completely covering the
2051 * file size, the pack is known to be valid even if
2052 * the descriptor is not currently open.
2054 if (p->windows) {
2055 struct pack_window *w = p->windows;
2057 if (!w->offset && w->len == p->pack_size)
2058 return 1;
2061 /* Force the pack to open to prove its valid. */
2062 return !open_packed_git(p);
2065 static int fill_pack_entry(const unsigned char *sha1,
2066 struct pack_entry *e,
2067 struct packed_git *p)
2069 off_t offset;
2071 if (p->num_bad_objects) {
2072 unsigned i;
2073 for (i = 0; i < p->num_bad_objects; i++)
2074 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2075 return 0;
2078 offset = find_pack_entry_one(sha1, p);
2079 if (!offset)
2080 return 0;
2083 * We are about to tell the caller where they can locate the
2084 * requested object. We better make sure the packfile is
2085 * still here and can be accessed before supplying that
2086 * answer, as it may have been deleted since the index was
2087 * loaded!
2089 if (!is_pack_valid(p)) {
2090 warning("packfile %s cannot be accessed", p->pack_name);
2091 return 0;
2093 e->offset = offset;
2094 e->p = p;
2095 hashcpy(e->sha1, sha1);
2096 return 1;
2099 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2101 struct packed_git *p;
2103 prepare_packed_git();
2104 if (!packed_git)
2105 return 0;
2107 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2108 return 1;
2110 for (p = packed_git; p; p = p->next) {
2111 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2112 continue;
2114 last_found_pack = p;
2115 return 1;
2117 return 0;
2120 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2121 struct packed_git *packs)
2123 struct packed_git *p;
2125 for (p = packs; p; p = p->next) {
2126 if (find_pack_entry_one(sha1, p))
2127 return p;
2129 return NULL;
2133 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2135 int status;
2136 unsigned long mapsize, size;
2137 void *map;
2138 git_zstream stream;
2139 char hdr[32];
2141 map = map_sha1_file(sha1, &mapsize);
2142 if (!map)
2143 return error("unable to find %s", sha1_to_hex(sha1));
2144 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2145 status = error("unable to unpack %s header",
2146 sha1_to_hex(sha1));
2147 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2148 status = error("unable to parse %s header", sha1_to_hex(sha1));
2149 else if (sizep)
2150 *sizep = size;
2151 git_inflate_end(&stream);
2152 munmap(map, mapsize);
2153 return status;
2156 /* returns enum object_type or negative */
2157 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2159 struct cached_object *co;
2160 struct pack_entry e;
2161 int status, rtype;
2163 co = find_cached_object(sha1);
2164 if (co) {
2165 if (oi->sizep)
2166 *(oi->sizep) = co->size;
2167 oi->whence = OI_CACHED;
2168 return co->type;
2171 if (!find_pack_entry(sha1, &e)) {
2172 /* Most likely it's a loose object. */
2173 status = sha1_loose_object_info(sha1, oi->sizep);
2174 if (status >= 0) {
2175 oi->whence = OI_LOOSE;
2176 return status;
2179 /* Not a loose object; someone else may have just packed it. */
2180 reprepare_packed_git();
2181 if (!find_pack_entry(sha1, &e))
2182 return status;
2185 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2186 if (status < 0) {
2187 mark_bad_packed_object(e.p, sha1);
2188 status = sha1_object_info_extended(sha1, oi);
2189 } else if (in_delta_base_cache(e.p, e.offset)) {
2190 oi->whence = OI_DBCACHED;
2191 } else {
2192 oi->whence = OI_PACKED;
2193 oi->u.packed.offset = e.offset;
2194 oi->u.packed.pack = e.p;
2195 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2196 rtype == OBJ_OFS_DELTA);
2199 return status;
2202 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2204 struct object_info oi;
2206 oi.sizep = sizep;
2207 return sha1_object_info_extended(sha1, &oi);
2210 static void *read_packed_sha1(const unsigned char *sha1,
2211 enum object_type *type, unsigned long *size)
2213 struct pack_entry e;
2214 void *data;
2216 if (!find_pack_entry(sha1, &e))
2217 return NULL;
2218 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2219 if (!data) {
2221 * We're probably in deep shit, but let's try to fetch
2222 * the required object anyway from another pack or loose.
2223 * This should happen only in the presence of a corrupted
2224 * pack, and is better than failing outright.
2226 error("failed to read object %s at offset %"PRIuMAX" from %s",
2227 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2228 mark_bad_packed_object(e.p, sha1);
2229 data = read_object(sha1, type, size);
2231 return data;
2234 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2235 unsigned char *sha1)
2237 struct cached_object *co;
2239 hash_sha1_file(buf, len, typename(type), sha1);
2240 if (has_sha1_file(sha1) || find_cached_object(sha1))
2241 return 0;
2242 if (cached_object_alloc <= cached_object_nr) {
2243 cached_object_alloc = alloc_nr(cached_object_alloc);
2244 cached_objects = xrealloc(cached_objects,
2245 sizeof(*cached_objects) *
2246 cached_object_alloc);
2248 co = &cached_objects[cached_object_nr++];
2249 co->size = len;
2250 co->type = type;
2251 co->buf = xmalloc(len);
2252 memcpy(co->buf, buf, len);
2253 hashcpy(co->sha1, sha1);
2254 return 0;
2257 static void *read_object(const unsigned char *sha1, enum object_type *type,
2258 unsigned long *size)
2260 unsigned long mapsize;
2261 void *map, *buf;
2262 struct cached_object *co;
2264 co = find_cached_object(sha1);
2265 if (co) {
2266 *type = co->type;
2267 *size = co->size;
2268 return xmemdupz(co->buf, co->size);
2271 buf = read_packed_sha1(sha1, type, size);
2272 if (buf)
2273 return buf;
2274 map = map_sha1_file(sha1, &mapsize);
2275 if (map) {
2276 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2277 munmap(map, mapsize);
2278 return buf;
2280 reprepare_packed_git();
2281 return read_packed_sha1(sha1, type, size);
2285 * This function dies on corrupt objects; the callers who want to
2286 * deal with them should arrange to call read_object() and give error
2287 * messages themselves.
2289 void *read_sha1_file_extended(const unsigned char *sha1,
2290 enum object_type *type,
2291 unsigned long *size,
2292 unsigned flag)
2294 void *data;
2295 char *path;
2296 const struct packed_git *p;
2297 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2298 ? lookup_replace_object(sha1) : sha1;
2300 errno = 0;
2301 data = read_object(repl, type, size);
2302 if (data)
2303 return data;
2305 if (errno && errno != ENOENT)
2306 die_errno("failed to read object %s", sha1_to_hex(sha1));
2308 /* die if we replaced an object with one that does not exist */
2309 if (repl != sha1)
2310 die("replacement %s not found for %s",
2311 sha1_to_hex(repl), sha1_to_hex(sha1));
2313 if (has_loose_object(repl)) {
2314 path = sha1_file_name(sha1);
2315 die("loose object %s (stored in %s) is corrupt",
2316 sha1_to_hex(repl), path);
2319 if ((p = has_packed_and_bad(repl)) != NULL)
2320 die("packed object %s (stored in %s) is corrupt",
2321 sha1_to_hex(repl), p->pack_name);
2323 return NULL;
2326 void *read_object_with_reference(const unsigned char *sha1,
2327 const char *required_type_name,
2328 unsigned long *size,
2329 unsigned char *actual_sha1_return)
2331 enum object_type type, required_type;
2332 void *buffer;
2333 unsigned long isize;
2334 unsigned char actual_sha1[20];
2336 required_type = type_from_string(required_type_name);
2337 hashcpy(actual_sha1, sha1);
2338 while (1) {
2339 int ref_length = -1;
2340 const char *ref_type = NULL;
2342 buffer = read_sha1_file(actual_sha1, &type, &isize);
2343 if (!buffer)
2344 return NULL;
2345 if (type == required_type) {
2346 *size = isize;
2347 if (actual_sha1_return)
2348 hashcpy(actual_sha1_return, actual_sha1);
2349 return buffer;
2351 /* Handle references */
2352 else if (type == OBJ_COMMIT)
2353 ref_type = "tree ";
2354 else if (type == OBJ_TAG)
2355 ref_type = "object ";
2356 else {
2357 free(buffer);
2358 return NULL;
2360 ref_length = strlen(ref_type);
2362 if (ref_length + 40 > isize ||
2363 memcmp(buffer, ref_type, ref_length) ||
2364 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2365 free(buffer);
2366 return NULL;
2368 free(buffer);
2369 /* Now we have the ID of the referred-to object in
2370 * actual_sha1. Check again. */
2374 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2375 const char *type, unsigned char *sha1,
2376 char *hdr, int *hdrlen)
2378 git_SHA_CTX c;
2380 /* Generate the header */
2381 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2383 /* Sha1.. */
2384 git_SHA1_Init(&c);
2385 git_SHA1_Update(&c, hdr, *hdrlen);
2386 git_SHA1_Update(&c, buf, len);
2387 git_SHA1_Final(sha1, &c);
2391 * Move the just written object into its final resting place.
2392 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2393 * "moving" is only a part of what it does, when no patch between
2394 * master to pu changes the call sites of this function.
2396 int move_temp_to_file(const char *tmpfile, const char *filename)
2398 int ret = 0;
2400 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2401 goto try_rename;
2402 else if (link(tmpfile, filename))
2403 ret = errno;
2406 * Coda hack - coda doesn't like cross-directory links,
2407 * so we fall back to a rename, which will mean that it
2408 * won't be able to check collisions, but that's not a
2409 * big deal.
2411 * The same holds for FAT formatted media.
2413 * When this succeeds, we just return. We have nothing
2414 * left to unlink.
2416 if (ret && ret != EEXIST) {
2417 try_rename:
2418 if (!rename(tmpfile, filename))
2419 goto out;
2420 ret = errno;
2422 unlink_or_warn(tmpfile);
2423 if (ret) {
2424 if (ret != EEXIST) {
2425 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2427 /* FIXME!!! Collision check here ? */
2430 out:
2431 if (adjust_shared_perm(filename))
2432 return error("unable to set permission to '%s'", filename);
2433 return 0;
2436 static int write_buffer(int fd, const void *buf, size_t len)
2438 if (write_in_full(fd, buf, len) < 0)
2439 return error("file write error (%s)", strerror(errno));
2440 return 0;
2443 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2444 unsigned char *sha1)
2446 char hdr[32];
2447 int hdrlen;
2448 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2449 return 0;
2452 /* Finalize a file on disk, and close it. */
2453 static void close_sha1_file(int fd)
2455 if (fsync_object_files)
2456 fsync_or_die(fd, "sha1 file");
2457 if (close(fd) != 0)
2458 die_errno("error when closing sha1 file");
2461 /* Size of directory component, including the ending '/' */
2462 static inline int directory_size(const char *filename)
2464 const char *s = strrchr(filename, '/');
2465 if (!s)
2466 return 0;
2467 return s - filename + 1;
2471 * This creates a temporary file in the same directory as the final
2472 * 'filename'
2474 * We want to avoid cross-directory filename renames, because those
2475 * can have problems on various filesystems (FAT, NFS, Coda).
2477 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2479 int fd, dirlen = directory_size(filename);
2481 if (dirlen + 20 > bufsiz) {
2482 errno = ENAMETOOLONG;
2483 return -1;
2485 memcpy(buffer, filename, dirlen);
2486 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2487 fd = git_mkstemp_mode(buffer, 0444);
2488 if (fd < 0 && dirlen && errno == ENOENT) {
2489 /* Make sure the directory exists */
2490 memcpy(buffer, filename, dirlen);
2491 buffer[dirlen-1] = 0;
2492 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2493 return -1;
2495 /* Try again */
2496 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2497 fd = git_mkstemp_mode(buffer, 0444);
2499 return fd;
2502 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2503 const void *buf, unsigned long len, time_t mtime)
2505 int fd, ret;
2506 unsigned char compressed[4096];
2507 git_zstream stream;
2508 git_SHA_CTX c;
2509 unsigned char parano_sha1[20];
2510 char *filename;
2511 static char tmp_file[PATH_MAX];
2513 filename = sha1_file_name(sha1);
2514 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2515 if (fd < 0) {
2516 if (errno == EACCES)
2517 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2518 else
2519 return error("unable to create temporary file: %s", strerror(errno));
2522 /* Set it up */
2523 memset(&stream, 0, sizeof(stream));
2524 git_deflate_init(&stream, zlib_compression_level);
2525 stream.next_out = compressed;
2526 stream.avail_out = sizeof(compressed);
2527 git_SHA1_Init(&c);
2529 /* First header.. */
2530 stream.next_in = (unsigned char *)hdr;
2531 stream.avail_in = hdrlen;
2532 while (git_deflate(&stream, 0) == Z_OK)
2533 ; /* nothing */
2534 git_SHA1_Update(&c, hdr, hdrlen);
2536 /* Then the data itself.. */
2537 stream.next_in = (void *)buf;
2538 stream.avail_in = len;
2539 do {
2540 unsigned char *in0 = stream.next_in;
2541 ret = git_deflate(&stream, Z_FINISH);
2542 git_SHA1_Update(&c, in0, stream.next_in - in0);
2543 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2544 die("unable to write sha1 file");
2545 stream.next_out = compressed;
2546 stream.avail_out = sizeof(compressed);
2547 } while (ret == Z_OK);
2549 if (ret != Z_STREAM_END)
2550 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2551 ret = git_deflate_end_gently(&stream);
2552 if (ret != Z_OK)
2553 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2554 git_SHA1_Final(parano_sha1, &c);
2555 if (hashcmp(sha1, parano_sha1) != 0)
2556 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2558 close_sha1_file(fd);
2560 if (mtime) {
2561 struct utimbuf utb;
2562 utb.actime = mtime;
2563 utb.modtime = mtime;
2564 if (utime(tmp_file, &utb) < 0)
2565 warning("failed utime() on %s: %s",
2566 tmp_file, strerror(errno));
2569 return move_temp_to_file(tmp_file, filename);
2572 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2574 unsigned char sha1[20];
2575 char hdr[32];
2576 int hdrlen;
2578 /* Normally if we have it in the pack then we do not bother writing
2579 * it out into .git/objects/??/?{38} file.
2581 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2582 if (returnsha1)
2583 hashcpy(returnsha1, sha1);
2584 if (has_sha1_file(sha1))
2585 return 0;
2586 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2589 int force_object_loose(const unsigned char *sha1, time_t mtime)
2591 void *buf;
2592 unsigned long len;
2593 enum object_type type;
2594 char hdr[32];
2595 int hdrlen;
2596 int ret;
2598 if (has_loose_object(sha1))
2599 return 0;
2600 buf = read_packed_sha1(sha1, &type, &len);
2601 if (!buf)
2602 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2603 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2604 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2605 free(buf);
2607 return ret;
2610 int has_pack_index(const unsigned char *sha1)
2612 struct stat st;
2613 if (stat(sha1_pack_index_name(sha1), &st))
2614 return 0;
2615 return 1;
2618 int has_sha1_pack(const unsigned char *sha1)
2620 struct pack_entry e;
2621 return find_pack_entry(sha1, &e);
2624 int has_sha1_file(const unsigned char *sha1)
2626 struct pack_entry e;
2628 if (find_pack_entry(sha1, &e))
2629 return 1;
2630 return has_loose_object(sha1);
2633 static void check_tree(const void *buf, size_t size)
2635 struct tree_desc desc;
2636 struct name_entry entry;
2638 init_tree_desc(&desc, buf, size);
2639 while (tree_entry(&desc, &entry))
2640 /* do nothing
2641 * tree_entry() will die() on malformed entries */
2645 static void check_commit(const void *buf, size_t size)
2647 struct commit c;
2648 memset(&c, 0, sizeof(c));
2649 if (parse_commit_buffer(&c, buf, size))
2650 die("corrupt commit");
2653 static void check_tag(const void *buf, size_t size)
2655 struct tag t;
2656 memset(&t, 0, sizeof(t));
2657 if (parse_tag_buffer(&t, buf, size))
2658 die("corrupt tag");
2661 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2662 enum object_type type,
2663 const char *path, unsigned flags)
2665 int ret, re_allocated = 0;
2666 int write_object = flags & HASH_WRITE_OBJECT;
2668 if (!type)
2669 type = OBJ_BLOB;
2672 * Convert blobs to git internal format
2674 if ((type == OBJ_BLOB) && path) {
2675 struct strbuf nbuf = STRBUF_INIT;
2676 if (convert_to_git(path, buf, size, &nbuf,
2677 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2678 buf = strbuf_detach(&nbuf, &size);
2679 re_allocated = 1;
2682 if (flags & HASH_FORMAT_CHECK) {
2683 if (type == OBJ_TREE)
2684 check_tree(buf, size);
2685 if (type == OBJ_COMMIT)
2686 check_commit(buf, size);
2687 if (type == OBJ_TAG)
2688 check_tag(buf, size);
2691 if (write_object)
2692 ret = write_sha1_file(buf, size, typename(type), sha1);
2693 else
2694 ret = hash_sha1_file(buf, size, typename(type), sha1);
2695 if (re_allocated)
2696 free(buf);
2697 return ret;
2700 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2701 const char *path, unsigned flags)
2703 struct strbuf sbuf = STRBUF_INIT;
2704 int ret;
2706 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2707 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2708 else
2709 ret = -1;
2710 strbuf_release(&sbuf);
2711 return ret;
2714 #define SMALL_FILE_SIZE (32*1024)
2716 static int index_core(unsigned char *sha1, int fd, size_t size,
2717 enum object_type type, const char *path,
2718 unsigned flags)
2720 int ret;
2722 if (!size) {
2723 ret = index_mem(sha1, NULL, size, type, path, flags);
2724 } else if (size <= SMALL_FILE_SIZE) {
2725 char *buf = xmalloc(size);
2726 if (size == read_in_full(fd, buf, size))
2727 ret = index_mem(sha1, buf, size, type, path, flags);
2728 else
2729 ret = error("short read %s", strerror(errno));
2730 free(buf);
2731 } else {
2732 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2733 ret = index_mem(sha1, buf, size, type, path, flags);
2734 munmap(buf, size);
2736 return ret;
2740 * This creates one packfile per large blob unless bulk-checkin
2741 * machinery is "plugged".
2743 * This also bypasses the usual "convert-to-git" dance, and that is on
2744 * purpose. We could write a streaming version of the converting
2745 * functions and insert that before feeding the data to fast-import
2746 * (or equivalent in-core API described above). However, that is
2747 * somewhat complicated, as we do not know the size of the filter
2748 * result, which we need to know beforehand when writing a git object.
2749 * Since the primary motivation for trying to stream from the working
2750 * tree file and to avoid mmaping it in core is to deal with large
2751 * binary blobs, they generally do not want to get any conversion, and
2752 * callers should avoid this code path when filters are requested.
2754 static int index_stream(unsigned char *sha1, int fd, size_t size,
2755 enum object_type type, const char *path,
2756 unsigned flags)
2758 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2761 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2762 enum object_type type, const char *path, unsigned flags)
2764 int ret;
2765 size_t size = xsize_t(st->st_size);
2767 if (!S_ISREG(st->st_mode))
2768 ret = index_pipe(sha1, fd, type, path, flags);
2769 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2770 (path && would_convert_to_git(path, NULL, 0, 0)))
2771 ret = index_core(sha1, fd, size, type, path, flags);
2772 else
2773 ret = index_stream(sha1, fd, size, type, path, flags);
2774 close(fd);
2775 return ret;
2778 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2780 int fd;
2781 struct strbuf sb = STRBUF_INIT;
2783 switch (st->st_mode & S_IFMT) {
2784 case S_IFREG:
2785 fd = open(path, O_RDONLY);
2786 if (fd < 0)
2787 return error("open(\"%s\"): %s", path,
2788 strerror(errno));
2789 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2790 return error("%s: failed to insert into database",
2791 path);
2792 break;
2793 case S_IFLNK:
2794 if (strbuf_readlink(&sb, path, st->st_size)) {
2795 char *errstr = strerror(errno);
2796 return error("readlink(\"%s\"): %s", path,
2797 errstr);
2799 if (!(flags & HASH_WRITE_OBJECT))
2800 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2801 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2802 return error("%s: failed to insert into database",
2803 path);
2804 strbuf_release(&sb);
2805 break;
2806 case S_IFDIR:
2807 return resolve_gitlink_ref(path, "HEAD", sha1);
2808 default:
2809 return error("%s: unsupported file type", path);
2811 return 0;
2814 int read_pack_header(int fd, struct pack_header *header)
2816 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2817 /* "eof before pack header was fully read" */
2818 return PH_ERROR_EOF;
2820 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2821 /* "protocol error (pack signature mismatch detected)" */
2822 return PH_ERROR_PACK_SIGNATURE;
2823 if (!pack_version_ok(header->hdr_version))
2824 /* "protocol error (pack version unsupported)" */
2825 return PH_ERROR_PROTOCOL;
2826 return 0;
2829 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2831 enum object_type type = sha1_object_info(sha1, NULL);
2832 if (type < 0)
2833 die("%s is not a valid object", sha1_to_hex(sha1));
2834 if (type != expect)
2835 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2836 typename(expect));