gitweb: Fix fixed string (non-regexp) project search
[git/jnareb-git.git] / sha1_file.c
blob32268d11d0c929624729b01f143845b13d7a569c
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 "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "run-command.h"
15 #include "tag.h"
16 #include "tree.h"
17 #include "tree-walk.h"
18 #include "refs.h"
19 #include "pack-revindex.h"
20 #include "sha1-lookup.h"
22 #ifndef O_NOATIME
23 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
24 #define O_NOATIME 01000000
25 #else
26 #define O_NOATIME 0
27 #endif
28 #endif
30 #define SZ_FMT PRIuMAX
31 static inline uintmax_t sz_fmt(size_t s) { return s; }
33 const unsigned char null_sha1[20];
36 * This is meant to hold a *small* number of objects that you would
37 * want read_sha1_file() to be able to return, but yet you do not want
38 * to write them into the object store (e.g. a browse-only
39 * application).
41 static struct cached_object {
42 unsigned char sha1[20];
43 enum object_type type;
44 void *buf;
45 unsigned long size;
46 } *cached_objects;
47 static int cached_object_nr, cached_object_alloc;
49 static struct cached_object empty_tree = {
50 EMPTY_TREE_SHA1_BIN_LITERAL,
51 OBJ_TREE,
52 "",
56 static struct cached_object *find_cached_object(const unsigned char *sha1)
58 int i;
59 struct cached_object *co = cached_objects;
61 for (i = 0; i < cached_object_nr; i++, co++) {
62 if (!hashcmp(co->sha1, sha1))
63 return co;
65 if (!hashcmp(sha1, empty_tree.sha1))
66 return &empty_tree;
67 return NULL;
70 int mkdir_in_gitdir(const char *path)
72 if (mkdir(path, 0777)) {
73 int saved_errno = errno;
74 struct stat st;
75 struct strbuf sb = STRBUF_INIT;
77 if (errno != EEXIST)
78 return -1;
80 * Are we looking at a path in a symlinked worktree
81 * whose original repository does not yet have it?
82 * e.g. .git/rr-cache pointing at its original
83 * repository in which the user hasn't performed any
84 * conflict resolution yet?
86 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
87 strbuf_readlink(&sb, path, st.st_size) ||
88 !is_absolute_path(sb.buf) ||
89 mkdir(sb.buf, 0777)) {
90 strbuf_release(&sb);
91 errno = saved_errno;
92 return -1;
94 strbuf_release(&sb);
96 return adjust_shared_perm(path);
99 int safe_create_leading_directories(char *path)
101 char *pos = path + offset_1st_component(path);
102 struct stat st;
104 while (pos) {
105 pos = strchr(pos, '/');
106 if (!pos)
107 break;
108 while (*++pos == '/')
110 if (!*pos)
111 break;
112 *--pos = '\0';
113 if (!stat(path, &st)) {
114 /* path exists */
115 if (!S_ISDIR(st.st_mode)) {
116 *pos = '/';
117 return -3;
120 else if (mkdir(path, 0777)) {
121 *pos = '/';
122 return -1;
124 else if (adjust_shared_perm(path)) {
125 *pos = '/';
126 return -2;
128 *pos++ = '/';
130 return 0;
133 int safe_create_leading_directories_const(const char *path)
135 /* path points to cache entries, so xstrdup before messing with it */
136 char *buf = xstrdup(path);
137 int result = safe_create_leading_directories(buf);
138 free(buf);
139 return result;
142 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
144 int i;
145 for (i = 0; i < 20; i++) {
146 static char hex[] = "0123456789abcdef";
147 unsigned int val = sha1[i];
148 char *pos = pathbuf + i*2 + (i > 0);
149 *pos++ = hex[val >> 4];
150 *pos = hex[val & 0xf];
155 * NOTE! This returns a statically allocated buffer, so you have to be
156 * careful about using it. Do an "xstrdup()" if you need to save the
157 * filename.
159 * Also note that this returns the location for creating. Reading
160 * SHA1 file can happen from any alternate directory listed in the
161 * DB_ENVIRONMENT environment variable if it is not found in
162 * the primary object database.
164 char *sha1_file_name(const unsigned char *sha1)
166 static char buf[PATH_MAX];
167 const char *objdir;
168 int len;
170 objdir = get_object_directory();
171 len = strlen(objdir);
173 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
174 if (len + 43 > PATH_MAX)
175 die("insanely long object directory %s", objdir);
176 memcpy(buf, objdir, len);
177 buf[len] = '/';
178 buf[len+3] = '/';
179 buf[len+42] = '\0';
180 fill_sha1_path(buf + len + 1, sha1);
181 return buf;
184 static char *sha1_get_pack_name(const unsigned char *sha1,
185 char **name, char **base, const char *which)
187 static const char hex[] = "0123456789abcdef";
188 char *buf;
189 int i;
191 if (!*base) {
192 const char *sha1_file_directory = get_object_directory();
193 int len = strlen(sha1_file_directory);
194 *base = xmalloc(len + 60);
195 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
196 sha1_file_directory, which);
197 *name = *base + len + 11;
200 buf = *name;
202 for (i = 0; i < 20; i++) {
203 unsigned int val = *sha1++;
204 *buf++ = hex[val >> 4];
205 *buf++ = hex[val & 0xf];
208 return *base;
211 char *sha1_pack_name(const unsigned char *sha1)
213 static char *name, *base;
215 return sha1_get_pack_name(sha1, &name, &base, "pack");
218 char *sha1_pack_index_name(const unsigned char *sha1)
220 static char *name, *base;
222 return sha1_get_pack_name(sha1, &name, &base, "idx");
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
229 static int git_open_noatime(const char *name);
232 * Prepare alternate object database registry.
234 * The variable alt_odb_list points at the list of struct
235 * alternate_object_database. The elements on this list come from
236 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
238 * whose contents is similar to that environment variable but can be
239 * LF separated. Its base points at a statically allocated buffer that
240 * contains "/the/directory/corresponding/to/.git/objects/...", while
241 * its name points just after the slash at the end of ".git/objects/"
242 * in the example above, and has enough space to hold 40-byte hex
243 * SHA1, an extra slash for the first level indirection, and the
244 * terminating NUL.
246 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
254 int base_len = -1;
256 if (!is_absolute_path(entry) && relative_base) {
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (!is_absolute_path(entry) && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
270 else
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (!is_directory(ent->base)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
310 return 0;
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
324 last = alt;
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
333 while (cp < ep && *cp != sep)
334 cp++;
335 if (last != cp) {
336 if (!is_absolute_path(last) && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
345 cp++;
346 last = cp;
350 static 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, 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, 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, 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 free(p);
720 return;
722 pp = &p->next;
727 * Do not call this directly as this leaks p->pack_fd on error return;
728 * call open_packed_git() instead.
730 static int open_packed_git_1(struct packed_git *p)
732 struct stat st;
733 struct pack_header hdr;
734 unsigned char sha1[20];
735 unsigned char *idx_sha1;
736 long fd_flag;
738 if (!p->index_data && open_pack_index(p))
739 return error("packfile %s index unavailable", p->pack_name);
741 if (!pack_max_fds) {
742 struct rlimit lim;
743 unsigned int max_fds;
745 if (getrlimit(RLIMIT_NOFILE, &lim))
746 die_errno("cannot get RLIMIT_NOFILE");
748 max_fds = lim.rlim_cur;
750 /* Save 3 for stdin/stdout/stderr, 22 for work */
751 if (25 < max_fds)
752 pack_max_fds = max_fds - 25;
753 else
754 pack_max_fds = 1;
757 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
758 ; /* nothing */
760 p->pack_fd = git_open_noatime(p->pack_name);
761 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
762 return -1;
763 pack_open_fds++;
765 /* If we created the struct before we had the pack we lack size. */
766 if (!p->pack_size) {
767 if (!S_ISREG(st.st_mode))
768 return error("packfile %s not a regular file", p->pack_name);
769 p->pack_size = st.st_size;
770 } else if (p->pack_size != st.st_size)
771 return error("packfile %s size changed", p->pack_name);
773 /* We leave these file descriptors open with sliding mmap;
774 * there is no point keeping them open across exec(), though.
776 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
777 if (fd_flag < 0)
778 return error("cannot determine file descriptor flags");
779 fd_flag |= FD_CLOEXEC;
780 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
781 return error("cannot set FD_CLOEXEC");
783 /* Verify we recognize this pack file format. */
784 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
785 return error("file %s is far too short to be a packfile", p->pack_name);
786 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
787 return error("file %s is not a GIT packfile", p->pack_name);
788 if (!pack_version_ok(hdr.hdr_version))
789 return error("packfile %s is version %"PRIu32" and not"
790 " supported (try upgrading GIT to a newer version)",
791 p->pack_name, ntohl(hdr.hdr_version));
793 /* Verify the pack matches its index. */
794 if (p->num_objects != ntohl(hdr.hdr_entries))
795 return error("packfile %s claims to have %"PRIu32" objects"
796 " while index indicates %"PRIu32" objects",
797 p->pack_name, ntohl(hdr.hdr_entries),
798 p->num_objects);
799 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
800 return error("end of packfile %s is unavailable", p->pack_name);
801 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
802 return error("packfile %s signature is unavailable", p->pack_name);
803 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
804 if (hashcmp(sha1, idx_sha1))
805 return error("packfile %s does not match index", p->pack_name);
806 return 0;
809 static int open_packed_git(struct packed_git *p)
811 if (!open_packed_git_1(p))
812 return 0;
813 if (p->pack_fd != -1) {
814 close(p->pack_fd);
815 pack_open_fds--;
816 p->pack_fd = -1;
818 return -1;
821 static int in_window(struct pack_window *win, off_t offset)
823 /* We must promise at least 20 bytes (one hash) after the
824 * offset is available from this window, otherwise the offset
825 * is not actually in this window and a different window (which
826 * has that one hash excess) must be used. This is to support
827 * the object header and delta base parsing routines below.
829 off_t win_off = win->offset;
830 return win_off <= offset
831 && (offset + 20) <= (win_off + win->len);
834 unsigned char *use_pack(struct packed_git *p,
835 struct pack_window **w_cursor,
836 off_t offset,
837 unsigned long *left)
839 struct pack_window *win = *w_cursor;
841 /* Since packfiles end in a hash of their content and it's
842 * pointless to ask for an offset into the middle of that
843 * hash, and the in_window function above wouldn't match
844 * don't allow an offset too close to the end of the file.
846 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
847 die("packfile %s cannot be accessed", p->pack_name);
848 if (offset > (p->pack_size - 20))
849 die("offset beyond end of packfile (truncated pack?)");
851 if (!win || !in_window(win, offset)) {
852 if (win)
853 win->inuse_cnt--;
854 for (win = p->windows; win; win = win->next) {
855 if (in_window(win, offset))
856 break;
858 if (!win) {
859 size_t window_align = packed_git_window_size / 2;
860 off_t len;
862 if (p->pack_fd == -1 && open_packed_git(p))
863 die("packfile %s cannot be accessed", p->pack_name);
865 win = xcalloc(1, sizeof(*win));
866 win->offset = (offset / window_align) * window_align;
867 len = p->pack_size - win->offset;
868 if (len > packed_git_window_size)
869 len = packed_git_window_size;
870 win->len = (size_t)len;
871 pack_mapped += win->len;
872 while (packed_git_limit < pack_mapped
873 && unuse_one_window(p, p->pack_fd))
874 ; /* nothing */
875 win->base = xmmap(NULL, win->len,
876 PROT_READ, MAP_PRIVATE,
877 p->pack_fd, win->offset);
878 if (win->base == MAP_FAILED)
879 die("packfile %s cannot be mapped: %s",
880 p->pack_name,
881 strerror(errno));
882 if (!win->offset && win->len == p->pack_size
883 && !p->do_not_close) {
884 close(p->pack_fd);
885 pack_open_fds--;
886 p->pack_fd = -1;
888 pack_mmap_calls++;
889 pack_open_windows++;
890 if (pack_mapped > peak_pack_mapped)
891 peak_pack_mapped = pack_mapped;
892 if (pack_open_windows > peak_pack_open_windows)
893 peak_pack_open_windows = pack_open_windows;
894 win->next = p->windows;
895 p->windows = win;
898 if (win != *w_cursor) {
899 win->last_used = pack_used_ctr++;
900 win->inuse_cnt++;
901 *w_cursor = win;
903 offset -= win->offset;
904 if (left)
905 *left = win->len - xsize_t(offset);
906 return win->base + offset;
909 static struct packed_git *alloc_packed_git(int extra)
911 struct packed_git *p = xmalloc(sizeof(*p) + extra);
912 memset(p, 0, sizeof(*p));
913 p->pack_fd = -1;
914 return p;
917 static void try_to_free_pack_memory(size_t size)
919 release_pack_memory(size, -1);
922 struct packed_git *add_packed_git(const char *path, int path_len, int local)
924 static int have_set_try_to_free_routine;
925 struct stat st;
926 struct packed_git *p = alloc_packed_git(path_len + 2);
928 if (!have_set_try_to_free_routine) {
929 have_set_try_to_free_routine = 1;
930 set_try_to_free_routine(try_to_free_pack_memory);
934 * Make sure a corresponding .pack file exists and that
935 * the index looks sane.
937 path_len -= strlen(".idx");
938 if (path_len < 1) {
939 free(p);
940 return NULL;
942 memcpy(p->pack_name, path, path_len);
944 strcpy(p->pack_name + path_len, ".keep");
945 if (!access(p->pack_name, F_OK))
946 p->pack_keep = 1;
948 strcpy(p->pack_name + path_len, ".pack");
949 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
950 free(p);
951 return NULL;
954 /* ok, it looks sane as far as we can check without
955 * actually mapping the pack file.
957 p->pack_size = st.st_size;
958 p->pack_local = local;
959 p->mtime = st.st_mtime;
960 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
961 hashclr(p->sha1);
962 return p;
965 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
967 const char *path = sha1_pack_name(sha1);
968 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
970 strcpy(p->pack_name, path);
971 hashcpy(p->sha1, sha1);
972 if (check_packed_git_idx(idx_path, p)) {
973 free(p);
974 return NULL;
977 return p;
980 void install_packed_git(struct packed_git *pack)
982 if (pack->pack_fd != -1)
983 pack_open_fds++;
985 pack->next = packed_git;
986 packed_git = pack;
989 static void prepare_packed_git_one(char *objdir, int local)
991 /* Ensure that this buffer is large enough so that we can
992 append "/pack/" without clobbering the stack even if
993 strlen(objdir) were PATH_MAX. */
994 char path[PATH_MAX + 1 + 4 + 1 + 1];
995 int len;
996 DIR *dir;
997 struct dirent *de;
999 sprintf(path, "%s/pack", objdir);
1000 len = strlen(path);
1001 dir = opendir(path);
1002 if (!dir) {
1003 if (errno != ENOENT)
1004 error("unable to open object pack directory: %s: %s",
1005 path, strerror(errno));
1006 return;
1008 path[len++] = '/';
1009 while ((de = readdir(dir)) != NULL) {
1010 int namelen = strlen(de->d_name);
1011 struct packed_git *p;
1013 if (!has_extension(de->d_name, ".idx"))
1014 continue;
1016 if (len + namelen + 1 > sizeof(path))
1017 continue;
1019 /* Don't reopen a pack we already have. */
1020 strcpy(path + len, de->d_name);
1021 for (p = packed_git; p; p = p->next) {
1022 if (!memcmp(path, p->pack_name, len + namelen - 4))
1023 break;
1025 if (p)
1026 continue;
1027 /* See if it really is a valid .idx file with corresponding
1028 * .pack file that we can map.
1030 p = add_packed_git(path, len + namelen, local);
1031 if (!p)
1032 continue;
1033 install_packed_git(p);
1035 closedir(dir);
1038 static int sort_pack(const void *a_, const void *b_)
1040 struct packed_git *a = *((struct packed_git **)a_);
1041 struct packed_git *b = *((struct packed_git **)b_);
1042 int st;
1045 * Local packs tend to contain objects specific to our
1046 * variant of the project than remote ones. In addition,
1047 * remote ones could be on a network mounted filesystem.
1048 * Favor local ones for these reasons.
1050 st = a->pack_local - b->pack_local;
1051 if (st)
1052 return -st;
1055 * Younger packs tend to contain more recent objects,
1056 * and more recent objects tend to get accessed more
1057 * often.
1059 if (a->mtime < b->mtime)
1060 return 1;
1061 else if (a->mtime == b->mtime)
1062 return 0;
1063 return -1;
1066 static void rearrange_packed_git(void)
1068 struct packed_git **ary, *p;
1069 int i, n;
1071 for (n = 0, p = packed_git; p; p = p->next)
1072 n++;
1073 if (n < 2)
1074 return;
1076 /* prepare an array of packed_git for easier sorting */
1077 ary = xcalloc(n, sizeof(struct packed_git *));
1078 for (n = 0, p = packed_git; p; p = p->next)
1079 ary[n++] = p;
1081 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1083 /* link them back again */
1084 for (i = 0; i < n - 1; i++)
1085 ary[i]->next = ary[i + 1];
1086 ary[n - 1]->next = NULL;
1087 packed_git = ary[0];
1089 free(ary);
1092 static int prepare_packed_git_run_once = 0;
1093 void prepare_packed_git(void)
1095 struct alternate_object_database *alt;
1097 if (prepare_packed_git_run_once)
1098 return;
1099 prepare_packed_git_one(get_object_directory(), 1);
1100 prepare_alt_odb();
1101 for (alt = alt_odb_list; alt; alt = alt->next) {
1102 alt->name[-1] = 0;
1103 prepare_packed_git_one(alt->base, 0);
1104 alt->name[-1] = '/';
1106 rearrange_packed_git();
1107 prepare_packed_git_run_once = 1;
1110 void reprepare_packed_git(void)
1112 discard_revindex();
1113 prepare_packed_git_run_once = 0;
1114 prepare_packed_git();
1117 static void mark_bad_packed_object(struct packed_git *p,
1118 const unsigned char *sha1)
1120 unsigned i;
1121 for (i = 0; i < p->num_bad_objects; i++)
1122 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1123 return;
1124 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1125 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1126 p->num_bad_objects++;
1129 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1131 struct packed_git *p;
1132 unsigned i;
1134 for (p = packed_git; p; p = p->next)
1135 for (i = 0; i < p->num_bad_objects; i++)
1136 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1137 return p;
1138 return NULL;
1141 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1143 unsigned char real_sha1[20];
1144 hash_sha1_file(map, size, type, real_sha1);
1145 return hashcmp(sha1, real_sha1) ? -1 : 0;
1148 static int git_open_noatime(const char *name)
1150 static int sha1_file_open_flag = O_NOATIME;
1152 for (;;) {
1153 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1154 if (fd >= 0)
1155 return fd;
1157 /* Might the failure be due to O_NOATIME? */
1158 if (errno != ENOENT && sha1_file_open_flag) {
1159 sha1_file_open_flag = 0;
1160 continue;
1163 return -1;
1167 static int open_sha1_file(const unsigned char *sha1)
1169 int fd;
1170 char *name = sha1_file_name(sha1);
1171 struct alternate_object_database *alt;
1173 fd = git_open_noatime(name);
1174 if (fd >= 0)
1175 return fd;
1177 prepare_alt_odb();
1178 errno = ENOENT;
1179 for (alt = alt_odb_list; alt; alt = alt->next) {
1180 name = alt->name;
1181 fill_sha1_path(name, sha1);
1182 fd = git_open_noatime(alt->base);
1183 if (fd >= 0)
1184 return fd;
1186 return -1;
1189 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1191 void *map;
1192 int fd;
1194 fd = open_sha1_file(sha1);
1195 map = NULL;
1196 if (fd >= 0) {
1197 struct stat st;
1199 if (!fstat(fd, &st)) {
1200 *size = xsize_t(st.st_size);
1201 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1203 close(fd);
1205 return map;
1209 * There used to be a second loose object header format which
1210 * was meant to mimic the in-pack format, allowing for direct
1211 * copy of the object data. This format turned up not to be
1212 * really worth it and we no longer write loose objects in that
1213 * format.
1215 static int experimental_loose_object(unsigned char *map)
1217 unsigned int word;
1220 * Is it a zlib-compressed buffer? If so, the first byte
1221 * must be 0x78 (15-bit window size, deflated), and the
1222 * first 16-bit word is evenly divisible by 31. If so,
1223 * we are looking at the official format, not the experimental
1224 * one.
1226 word = (map[0] << 8) + map[1];
1227 if (map[0] == 0x78 && !(word % 31))
1228 return 0;
1229 else
1230 return 1;
1233 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1234 unsigned long len, enum object_type *type, unsigned long *sizep)
1236 unsigned shift;
1237 unsigned long size, c;
1238 unsigned long used = 0;
1240 c = buf[used++];
1241 *type = (c >> 4) & 7;
1242 size = c & 15;
1243 shift = 4;
1244 while (c & 0x80) {
1245 if (len <= used || bitsizeof(long) <= shift) {
1246 error("bad object header");
1247 return 0;
1249 c = buf[used++];
1250 size += (c & 0x7f) << shift;
1251 shift += 7;
1253 *sizep = size;
1254 return used;
1257 static int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1259 unsigned long size, used;
1260 static const char valid_loose_object_type[8] = {
1261 0, /* OBJ_EXT */
1262 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1263 0, /* "delta" and others are invalid in a loose object */
1265 enum object_type type;
1267 /* Get the data stream */
1268 memset(stream, 0, sizeof(*stream));
1269 stream->next_in = map;
1270 stream->avail_in = mapsize;
1271 stream->next_out = buffer;
1272 stream->avail_out = bufsiz;
1274 if (experimental_loose_object(map)) {
1276 * The old experimental format we no longer produce;
1277 * we can still read it.
1279 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1280 if (!used || !valid_loose_object_type[type])
1281 return -1;
1282 map += used;
1283 mapsize -= used;
1285 /* Set up the stream for the rest.. */
1286 stream->next_in = map;
1287 stream->avail_in = mapsize;
1288 git_inflate_init(stream);
1290 /* And generate the fake traditional header */
1291 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1292 typename(type), size);
1293 return 0;
1295 git_inflate_init(stream);
1296 return git_inflate(stream, 0);
1299 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1301 int bytes = strlen(buffer) + 1;
1302 unsigned char *buf = xmallocz(size);
1303 unsigned long n;
1304 int status = Z_OK;
1306 n = stream->total_out - bytes;
1307 if (n > size)
1308 n = size;
1309 memcpy(buf, (char *) buffer + bytes, n);
1310 bytes = n;
1311 if (bytes <= size) {
1313 * The above condition must be (bytes <= size), not
1314 * (bytes < size). In other words, even though we
1315 * expect no more output and set avail_out to zero,
1316 * the input zlib stream may have bytes that express
1317 * "this concludes the stream", and we *do* want to
1318 * eat that input.
1320 * Otherwise we would not be able to test that we
1321 * consumed all the input to reach the expected size;
1322 * we also want to check that zlib tells us that all
1323 * went well with status == Z_STREAM_END at the end.
1325 stream->next_out = buf + bytes;
1326 stream->avail_out = size - bytes;
1327 while (status == Z_OK)
1328 status = git_inflate(stream, Z_FINISH);
1330 if (status == Z_STREAM_END && !stream->avail_in) {
1331 git_inflate_end(stream);
1332 return buf;
1335 if (status < 0)
1336 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1337 else if (stream->avail_in)
1338 error("garbage at end of loose object '%s'",
1339 sha1_to_hex(sha1));
1340 free(buf);
1341 return NULL;
1345 * We used to just use "sscanf()", but that's actually way
1346 * too permissive for what we want to check. So do an anal
1347 * object header parse by hand.
1349 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1351 char type[10];
1352 int i;
1353 unsigned long size;
1356 * The type can be at most ten bytes (including the
1357 * terminating '\0' that we add), and is followed by
1358 * a space.
1360 i = 0;
1361 for (;;) {
1362 char c = *hdr++;
1363 if (c == ' ')
1364 break;
1365 type[i++] = c;
1366 if (i >= sizeof(type))
1367 return -1;
1369 type[i] = 0;
1372 * The length must follow immediately, and be in canonical
1373 * decimal format (ie "010" is not valid).
1375 size = *hdr++ - '0';
1376 if (size > 9)
1377 return -1;
1378 if (size) {
1379 for (;;) {
1380 unsigned long c = *hdr - '0';
1381 if (c > 9)
1382 break;
1383 hdr++;
1384 size = size * 10 + c;
1387 *sizep = size;
1390 * The length must be followed by a zero byte
1392 return *hdr ? -1 : type_from_string(type);
1395 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1397 int ret;
1398 git_zstream stream;
1399 char hdr[8192];
1401 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1402 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1403 return NULL;
1405 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1408 unsigned long get_size_from_delta(struct packed_git *p,
1409 struct pack_window **w_curs,
1410 off_t curpos)
1412 const unsigned char *data;
1413 unsigned char delta_head[20], *in;
1414 git_zstream stream;
1415 int st;
1417 memset(&stream, 0, sizeof(stream));
1418 stream.next_out = delta_head;
1419 stream.avail_out = sizeof(delta_head);
1421 git_inflate_init(&stream);
1422 do {
1423 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1424 stream.next_in = in;
1425 st = git_inflate(&stream, Z_FINISH);
1426 curpos += stream.next_in - in;
1427 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1428 stream.total_out < sizeof(delta_head));
1429 git_inflate_end(&stream);
1430 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1431 error("delta data unpack-initial failed");
1432 return 0;
1435 /* Examine the initial part of the delta to figure out
1436 * the result size.
1438 data = delta_head;
1440 /* ignore base size */
1441 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1443 /* Read the result size */
1444 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1447 static off_t get_delta_base(struct packed_git *p,
1448 struct pack_window **w_curs,
1449 off_t *curpos,
1450 enum object_type type,
1451 off_t delta_obj_offset)
1453 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1454 off_t base_offset;
1456 /* use_pack() assured us we have [base_info, base_info + 20)
1457 * as a range that we can look at without walking off the
1458 * end of the mapped window. Its actually the hash size
1459 * that is assured. An OFS_DELTA longer than the hash size
1460 * is stupid, as then a REF_DELTA would be smaller to store.
1462 if (type == OBJ_OFS_DELTA) {
1463 unsigned used = 0;
1464 unsigned char c = base_info[used++];
1465 base_offset = c & 127;
1466 while (c & 128) {
1467 base_offset += 1;
1468 if (!base_offset || MSB(base_offset, 7))
1469 return 0; /* overflow */
1470 c = base_info[used++];
1471 base_offset = (base_offset << 7) + (c & 127);
1473 base_offset = delta_obj_offset - base_offset;
1474 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1475 return 0; /* out of bound */
1476 *curpos += used;
1477 } else if (type == OBJ_REF_DELTA) {
1478 /* The base entry _must_ be in the same pack */
1479 base_offset = find_pack_entry_one(base_info, p);
1480 *curpos += 20;
1481 } else
1482 die("I am totally screwed");
1483 return base_offset;
1486 /* forward declaration for a mutually recursive function */
1487 static int packed_object_info(struct packed_git *p, off_t offset,
1488 unsigned long *sizep);
1490 static int packed_delta_info(struct packed_git *p,
1491 struct pack_window **w_curs,
1492 off_t curpos,
1493 enum object_type type,
1494 off_t obj_offset,
1495 unsigned long *sizep)
1497 off_t base_offset;
1499 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1500 if (!base_offset)
1501 return OBJ_BAD;
1502 type = packed_object_info(p, base_offset, NULL);
1503 if (type <= OBJ_NONE) {
1504 struct revindex_entry *revidx;
1505 const unsigned char *base_sha1;
1506 revidx = find_pack_revindex(p, base_offset);
1507 if (!revidx)
1508 return OBJ_BAD;
1509 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1510 mark_bad_packed_object(p, base_sha1);
1511 type = sha1_object_info(base_sha1, NULL);
1512 if (type <= OBJ_NONE)
1513 return OBJ_BAD;
1516 /* We choose to only get the type of the base object and
1517 * ignore potentially corrupt pack file that expects the delta
1518 * based on a base with a wrong size. This saves tons of
1519 * inflate() calls.
1521 if (sizep) {
1522 *sizep = get_size_from_delta(p, w_curs, curpos);
1523 if (*sizep == 0)
1524 type = OBJ_BAD;
1527 return type;
1530 static int unpack_object_header(struct packed_git *p,
1531 struct pack_window **w_curs,
1532 off_t *curpos,
1533 unsigned long *sizep)
1535 unsigned char *base;
1536 unsigned long left;
1537 unsigned long used;
1538 enum object_type type;
1540 /* use_pack() assures us we have [base, base + 20) available
1541 * as a range that we can look at. (Its actually the hash
1542 * size that is assured.) With our object header encoding
1543 * the maximum deflated object size is 2^137, which is just
1544 * insane, so we know won't exceed what we have been given.
1546 base = use_pack(p, w_curs, *curpos, &left);
1547 used = unpack_object_header_buffer(base, left, &type, sizep);
1548 if (!used) {
1549 type = OBJ_BAD;
1550 } else
1551 *curpos += used;
1553 return type;
1556 const char *packed_object_info_detail(struct packed_git *p,
1557 off_t obj_offset,
1558 unsigned long *size,
1559 unsigned long *store_size,
1560 unsigned int *delta_chain_length,
1561 unsigned char *base_sha1)
1563 struct pack_window *w_curs = NULL;
1564 off_t curpos;
1565 unsigned long dummy;
1566 unsigned char *next_sha1;
1567 enum object_type type;
1568 struct revindex_entry *revidx;
1570 *delta_chain_length = 0;
1571 curpos = obj_offset;
1572 type = unpack_object_header(p, &w_curs, &curpos, size);
1574 revidx = find_pack_revindex(p, obj_offset);
1575 *store_size = revidx[1].offset - obj_offset;
1577 for (;;) {
1578 switch (type) {
1579 default:
1580 die("pack %s contains unknown object type %d",
1581 p->pack_name, type);
1582 case OBJ_COMMIT:
1583 case OBJ_TREE:
1584 case OBJ_BLOB:
1585 case OBJ_TAG:
1586 unuse_pack(&w_curs);
1587 return typename(type);
1588 case OBJ_OFS_DELTA:
1589 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1590 if (!obj_offset)
1591 die("pack %s contains bad delta base reference of type %s",
1592 p->pack_name, typename(type));
1593 if (*delta_chain_length == 0) {
1594 revidx = find_pack_revindex(p, obj_offset);
1595 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1597 break;
1598 case OBJ_REF_DELTA:
1599 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1600 if (*delta_chain_length == 0)
1601 hashcpy(base_sha1, next_sha1);
1602 obj_offset = find_pack_entry_one(next_sha1, p);
1603 break;
1605 (*delta_chain_length)++;
1606 curpos = obj_offset;
1607 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1611 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1612 unsigned long *sizep)
1614 struct pack_window *w_curs = NULL;
1615 unsigned long size;
1616 off_t curpos = obj_offset;
1617 enum object_type type;
1619 type = unpack_object_header(p, &w_curs, &curpos, &size);
1621 switch (type) {
1622 case OBJ_OFS_DELTA:
1623 case OBJ_REF_DELTA:
1624 type = packed_delta_info(p, &w_curs, curpos,
1625 type, obj_offset, sizep);
1626 break;
1627 case OBJ_COMMIT:
1628 case OBJ_TREE:
1629 case OBJ_BLOB:
1630 case OBJ_TAG:
1631 if (sizep)
1632 *sizep = size;
1633 break;
1634 default:
1635 error("unknown object type %i at offset %"PRIuMAX" in %s",
1636 type, (uintmax_t)obj_offset, p->pack_name);
1637 type = OBJ_BAD;
1639 unuse_pack(&w_curs);
1640 return type;
1643 static void *unpack_compressed_entry(struct packed_git *p,
1644 struct pack_window **w_curs,
1645 off_t curpos,
1646 unsigned long size)
1648 int st;
1649 git_zstream stream;
1650 unsigned char *buffer, *in;
1652 buffer = xmallocz(size);
1653 memset(&stream, 0, sizeof(stream));
1654 stream.next_out = buffer;
1655 stream.avail_out = size + 1;
1657 git_inflate_init(&stream);
1658 do {
1659 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1660 stream.next_in = in;
1661 st = git_inflate(&stream, Z_FINISH);
1662 if (!stream.avail_out)
1663 break; /* the payload is larger than it should be */
1664 curpos += stream.next_in - in;
1665 } while (st == Z_OK || st == Z_BUF_ERROR);
1666 git_inflate_end(&stream);
1667 if ((st != Z_STREAM_END) || stream.total_out != size) {
1668 free(buffer);
1669 return NULL;
1672 return buffer;
1675 #define MAX_DELTA_CACHE (256)
1677 static size_t delta_base_cached;
1679 static struct delta_base_cache_lru_list {
1680 struct delta_base_cache_lru_list *prev;
1681 struct delta_base_cache_lru_list *next;
1682 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1684 static struct delta_base_cache_entry {
1685 struct delta_base_cache_lru_list lru;
1686 void *data;
1687 struct packed_git *p;
1688 off_t base_offset;
1689 unsigned long size;
1690 enum object_type type;
1691 } delta_base_cache[MAX_DELTA_CACHE];
1693 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1695 unsigned long hash;
1697 hash = (unsigned long)p + (unsigned long)base_offset;
1698 hash += (hash >> 8) + (hash >> 16);
1699 return hash % MAX_DELTA_CACHE;
1702 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1703 unsigned long *base_size, enum object_type *type, int keep_cache)
1705 void *ret;
1706 unsigned long hash = pack_entry_hash(p, base_offset);
1707 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1709 ret = ent->data;
1710 if (!ret || ent->p != p || ent->base_offset != base_offset)
1711 return unpack_entry(p, base_offset, type, base_size);
1713 if (!keep_cache) {
1714 ent->data = NULL;
1715 ent->lru.next->prev = ent->lru.prev;
1716 ent->lru.prev->next = ent->lru.next;
1717 delta_base_cached -= ent->size;
1718 } else {
1719 ret = xmemdupz(ent->data, ent->size);
1721 *type = ent->type;
1722 *base_size = ent->size;
1723 return ret;
1726 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1728 if (ent->data) {
1729 free(ent->data);
1730 ent->data = NULL;
1731 ent->lru.next->prev = ent->lru.prev;
1732 ent->lru.prev->next = ent->lru.next;
1733 delta_base_cached -= ent->size;
1737 void clear_delta_base_cache(void)
1739 unsigned long p;
1740 for (p = 0; p < MAX_DELTA_CACHE; p++)
1741 release_delta_base_cache(&delta_base_cache[p]);
1744 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1745 void *base, unsigned long base_size, enum object_type type)
1747 unsigned long hash = pack_entry_hash(p, base_offset);
1748 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1749 struct delta_base_cache_lru_list *lru;
1751 release_delta_base_cache(ent);
1752 delta_base_cached += base_size;
1754 for (lru = delta_base_cache_lru.next;
1755 delta_base_cached > delta_base_cache_limit
1756 && lru != &delta_base_cache_lru;
1757 lru = lru->next) {
1758 struct delta_base_cache_entry *f = (void *)lru;
1759 if (f->type == OBJ_BLOB)
1760 release_delta_base_cache(f);
1762 for (lru = delta_base_cache_lru.next;
1763 delta_base_cached > delta_base_cache_limit
1764 && lru != &delta_base_cache_lru;
1765 lru = lru->next) {
1766 struct delta_base_cache_entry *f = (void *)lru;
1767 release_delta_base_cache(f);
1770 ent->p = p;
1771 ent->base_offset = base_offset;
1772 ent->type = type;
1773 ent->data = base;
1774 ent->size = base_size;
1775 ent->lru.next = &delta_base_cache_lru;
1776 ent->lru.prev = delta_base_cache_lru.prev;
1777 delta_base_cache_lru.prev->next = &ent->lru;
1778 delta_base_cache_lru.prev = &ent->lru;
1781 static void *read_object(const unsigned char *sha1, enum object_type *type,
1782 unsigned long *size);
1784 static void *unpack_delta_entry(struct packed_git *p,
1785 struct pack_window **w_curs,
1786 off_t curpos,
1787 unsigned long delta_size,
1788 off_t obj_offset,
1789 enum object_type *type,
1790 unsigned long *sizep)
1792 void *delta_data, *result, *base;
1793 unsigned long base_size;
1794 off_t base_offset;
1796 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1797 if (!base_offset) {
1798 error("failed to validate delta base reference "
1799 "at offset %"PRIuMAX" from %s",
1800 (uintmax_t)curpos, p->pack_name);
1801 return NULL;
1803 unuse_pack(w_curs);
1804 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1805 if (!base) {
1807 * We're probably in deep shit, but let's try to fetch
1808 * the required base anyway from another pack or loose.
1809 * This is costly but should happen only in the presence
1810 * of a corrupted pack, and is better than failing outright.
1812 struct revindex_entry *revidx;
1813 const unsigned char *base_sha1;
1814 revidx = find_pack_revindex(p, base_offset);
1815 if (!revidx)
1816 return NULL;
1817 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1818 error("failed to read delta base object %s"
1819 " at offset %"PRIuMAX" from %s",
1820 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1821 p->pack_name);
1822 mark_bad_packed_object(p, base_sha1);
1823 base = read_object(base_sha1, type, &base_size);
1824 if (!base)
1825 return NULL;
1828 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1829 if (!delta_data) {
1830 error("failed to unpack compressed delta "
1831 "at offset %"PRIuMAX" from %s",
1832 (uintmax_t)curpos, p->pack_name);
1833 free(base);
1834 return NULL;
1836 result = patch_delta(base, base_size,
1837 delta_data, delta_size,
1838 sizep);
1839 if (!result)
1840 die("failed to apply delta");
1841 free(delta_data);
1842 add_delta_base_cache(p, base_offset, base, base_size, *type);
1843 return result;
1846 int do_check_packed_object_crc;
1848 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1849 enum object_type *type, unsigned long *sizep)
1851 struct pack_window *w_curs = NULL;
1852 off_t curpos = obj_offset;
1853 void *data;
1855 if (do_check_packed_object_crc && p->index_version > 1) {
1856 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1857 unsigned long len = revidx[1].offset - obj_offset;
1858 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1859 const unsigned char *sha1 =
1860 nth_packed_object_sha1(p, revidx->nr);
1861 error("bad packed object CRC for %s",
1862 sha1_to_hex(sha1));
1863 mark_bad_packed_object(p, sha1);
1864 unuse_pack(&w_curs);
1865 return NULL;
1869 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1870 switch (*type) {
1871 case OBJ_OFS_DELTA:
1872 case OBJ_REF_DELTA:
1873 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1874 obj_offset, type, sizep);
1875 break;
1876 case OBJ_COMMIT:
1877 case OBJ_TREE:
1878 case OBJ_BLOB:
1879 case OBJ_TAG:
1880 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1881 break;
1882 default:
1883 data = NULL;
1884 error("unknown object type %i at offset %"PRIuMAX" in %s",
1885 *type, (uintmax_t)obj_offset, p->pack_name);
1887 unuse_pack(&w_curs);
1888 return data;
1891 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1892 uint32_t n)
1894 const unsigned char *index = p->index_data;
1895 if (!index) {
1896 if (open_pack_index(p))
1897 return NULL;
1898 index = p->index_data;
1900 if (n >= p->num_objects)
1901 return NULL;
1902 index += 4 * 256;
1903 if (p->index_version == 1) {
1904 return index + 24 * n + 4;
1905 } else {
1906 index += 8;
1907 return index + 20 * n;
1911 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1913 const unsigned char *index = p->index_data;
1914 index += 4 * 256;
1915 if (p->index_version == 1) {
1916 return ntohl(*((uint32_t *)(index + 24 * n)));
1917 } else {
1918 uint32_t off;
1919 index += 8 + p->num_objects * (20 + 4);
1920 off = ntohl(*((uint32_t *)(index + 4 * n)));
1921 if (!(off & 0x80000000))
1922 return off;
1923 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1924 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1925 ntohl(*((uint32_t *)(index + 4)));
1929 off_t find_pack_entry_one(const unsigned char *sha1,
1930 struct packed_git *p)
1932 const uint32_t *level1_ofs = p->index_data;
1933 const unsigned char *index = p->index_data;
1934 unsigned hi, lo, stride;
1935 static int use_lookup = -1;
1936 static int debug_lookup = -1;
1938 if (debug_lookup < 0)
1939 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1941 if (!index) {
1942 if (open_pack_index(p))
1943 return 0;
1944 level1_ofs = p->index_data;
1945 index = p->index_data;
1947 if (p->index_version > 1) {
1948 level1_ofs += 2;
1949 index += 8;
1951 index += 4 * 256;
1952 hi = ntohl(level1_ofs[*sha1]);
1953 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1954 if (p->index_version > 1) {
1955 stride = 20;
1956 } else {
1957 stride = 24;
1958 index += 4;
1961 if (debug_lookup)
1962 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1963 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1965 if (use_lookup < 0)
1966 use_lookup = !!getenv("GIT_USE_LOOKUP");
1967 if (use_lookup) {
1968 int pos = sha1_entry_pos(index, stride, 0,
1969 lo, hi, p->num_objects, sha1);
1970 if (pos < 0)
1971 return 0;
1972 return nth_packed_object_offset(p, pos);
1975 do {
1976 unsigned mi = (lo + hi) / 2;
1977 int cmp = hashcmp(index + mi * stride, sha1);
1979 if (debug_lookup)
1980 printf("lo %u hi %u rg %u mi %u\n",
1981 lo, hi, hi - lo, mi);
1982 if (!cmp)
1983 return nth_packed_object_offset(p, mi);
1984 if (cmp > 0)
1985 hi = mi;
1986 else
1987 lo = mi+1;
1988 } while (lo < hi);
1989 return 0;
1992 static int is_pack_valid(struct packed_git *p)
1994 /* An already open pack is known to be valid. */
1995 if (p->pack_fd != -1)
1996 return 1;
1998 /* If the pack has one window completely covering the
1999 * file size, the pack is known to be valid even if
2000 * the descriptor is not currently open.
2002 if (p->windows) {
2003 struct pack_window *w = p->windows;
2005 if (!w->offset && w->len == p->pack_size)
2006 return 1;
2009 /* Force the pack to open to prove its valid. */
2010 return !open_packed_git(p);
2013 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2015 static struct packed_git *last_found = (void *)1;
2016 struct packed_git *p;
2017 off_t offset;
2019 prepare_packed_git();
2020 if (!packed_git)
2021 return 0;
2022 p = (last_found == (void *)1) ? packed_git : last_found;
2024 do {
2025 if (p->num_bad_objects) {
2026 unsigned i;
2027 for (i = 0; i < p->num_bad_objects; i++)
2028 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2029 goto next;
2032 offset = find_pack_entry_one(sha1, p);
2033 if (offset) {
2035 * We are about to tell the caller where they can
2036 * locate the requested object. We better make
2037 * sure the packfile is still here and can be
2038 * accessed before supplying that answer, as
2039 * it may have been deleted since the index
2040 * was loaded!
2042 if (!is_pack_valid(p)) {
2043 error("packfile %s cannot be accessed", p->pack_name);
2044 goto next;
2046 e->offset = offset;
2047 e->p = p;
2048 hashcpy(e->sha1, sha1);
2049 last_found = p;
2050 return 1;
2053 next:
2054 if (p == last_found)
2055 p = packed_git;
2056 else
2057 p = p->next;
2058 if (p == last_found)
2059 p = p->next;
2060 } while (p);
2061 return 0;
2064 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2065 struct packed_git *packs)
2067 struct packed_git *p;
2069 for (p = packs; p; p = p->next) {
2070 if (find_pack_entry_one(sha1, p))
2071 return p;
2073 return NULL;
2077 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2079 int status;
2080 unsigned long mapsize, size;
2081 void *map;
2082 git_zstream stream;
2083 char hdr[32];
2085 map = map_sha1_file(sha1, &mapsize);
2086 if (!map)
2087 return error("unable to find %s", sha1_to_hex(sha1));
2088 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2089 status = error("unable to unpack %s header",
2090 sha1_to_hex(sha1));
2091 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2092 status = error("unable to parse %s header", sha1_to_hex(sha1));
2093 else if (sizep)
2094 *sizep = size;
2095 git_inflate_end(&stream);
2096 munmap(map, mapsize);
2097 return status;
2100 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2102 struct cached_object *co;
2103 struct pack_entry e;
2104 int status;
2106 co = find_cached_object(sha1);
2107 if (co) {
2108 if (sizep)
2109 *sizep = co->size;
2110 return co->type;
2113 if (!find_pack_entry(sha1, &e)) {
2114 /* Most likely it's a loose object. */
2115 status = sha1_loose_object_info(sha1, sizep);
2116 if (status >= 0)
2117 return status;
2119 /* Not a loose object; someone else may have just packed it. */
2120 reprepare_packed_git();
2121 if (!find_pack_entry(sha1, &e))
2122 return status;
2125 status = packed_object_info(e.p, e.offset, sizep);
2126 if (status < 0) {
2127 mark_bad_packed_object(e.p, sha1);
2128 status = sha1_object_info(sha1, sizep);
2131 return status;
2134 static void *read_packed_sha1(const unsigned char *sha1,
2135 enum object_type *type, unsigned long *size)
2137 struct pack_entry e;
2138 void *data;
2140 if (!find_pack_entry(sha1, &e))
2141 return NULL;
2142 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2143 if (!data) {
2145 * We're probably in deep shit, but let's try to fetch
2146 * the required object anyway from another pack or loose.
2147 * This should happen only in the presence of a corrupted
2148 * pack, and is better than failing outright.
2150 error("failed to read object %s at offset %"PRIuMAX" from %s",
2151 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2152 mark_bad_packed_object(e.p, sha1);
2153 data = read_object(sha1, type, size);
2155 return data;
2158 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2159 unsigned char *sha1)
2161 struct cached_object *co;
2163 hash_sha1_file(buf, len, typename(type), sha1);
2164 if (has_sha1_file(sha1) || find_cached_object(sha1))
2165 return 0;
2166 if (cached_object_alloc <= cached_object_nr) {
2167 cached_object_alloc = alloc_nr(cached_object_alloc);
2168 cached_objects = xrealloc(cached_objects,
2169 sizeof(*cached_objects) *
2170 cached_object_alloc);
2172 co = &cached_objects[cached_object_nr++];
2173 co->size = len;
2174 co->type = type;
2175 co->buf = xmalloc(len);
2176 memcpy(co->buf, buf, len);
2177 hashcpy(co->sha1, sha1);
2178 return 0;
2181 static void *read_object(const unsigned char *sha1, enum object_type *type,
2182 unsigned long *size)
2184 unsigned long mapsize;
2185 void *map, *buf;
2186 struct cached_object *co;
2188 co = find_cached_object(sha1);
2189 if (co) {
2190 *type = co->type;
2191 *size = co->size;
2192 return xmemdupz(co->buf, co->size);
2195 buf = read_packed_sha1(sha1, type, size);
2196 if (buf)
2197 return buf;
2198 map = map_sha1_file(sha1, &mapsize);
2199 if (map) {
2200 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2201 munmap(map, mapsize);
2202 return buf;
2204 reprepare_packed_git();
2205 return read_packed_sha1(sha1, type, size);
2209 * This function dies on corrupt objects; the callers who want to
2210 * deal with them should arrange to call read_object() and give error
2211 * messages themselves.
2213 void *read_sha1_file_extended(const unsigned char *sha1,
2214 enum object_type *type,
2215 unsigned long *size,
2216 unsigned flag)
2218 void *data;
2219 char *path;
2220 const struct packed_git *p;
2221 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2222 ? lookup_replace_object(sha1) : sha1;
2224 errno = 0;
2225 data = read_object(repl, type, size);
2226 if (data)
2227 return data;
2229 if (errno && errno != ENOENT)
2230 die_errno("failed to read object %s", sha1_to_hex(sha1));
2232 /* die if we replaced an object with one that does not exist */
2233 if (repl != sha1)
2234 die("replacement %s not found for %s",
2235 sha1_to_hex(repl), sha1_to_hex(sha1));
2237 if (has_loose_object(repl)) {
2238 path = sha1_file_name(sha1);
2239 die("loose object %s (stored in %s) is corrupt",
2240 sha1_to_hex(repl), path);
2243 if ((p = has_packed_and_bad(repl)) != NULL)
2244 die("packed object %s (stored in %s) is corrupt",
2245 sha1_to_hex(repl), p->pack_name);
2247 return NULL;
2250 void *read_object_with_reference(const unsigned char *sha1,
2251 const char *required_type_name,
2252 unsigned long *size,
2253 unsigned char *actual_sha1_return)
2255 enum object_type type, required_type;
2256 void *buffer;
2257 unsigned long isize;
2258 unsigned char actual_sha1[20];
2260 required_type = type_from_string(required_type_name);
2261 hashcpy(actual_sha1, sha1);
2262 while (1) {
2263 int ref_length = -1;
2264 const char *ref_type = NULL;
2266 buffer = read_sha1_file(actual_sha1, &type, &isize);
2267 if (!buffer)
2268 return NULL;
2269 if (type == required_type) {
2270 *size = isize;
2271 if (actual_sha1_return)
2272 hashcpy(actual_sha1_return, actual_sha1);
2273 return buffer;
2275 /* Handle references */
2276 else if (type == OBJ_COMMIT)
2277 ref_type = "tree ";
2278 else if (type == OBJ_TAG)
2279 ref_type = "object ";
2280 else {
2281 free(buffer);
2282 return NULL;
2284 ref_length = strlen(ref_type);
2286 if (ref_length + 40 > isize ||
2287 memcmp(buffer, ref_type, ref_length) ||
2288 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2289 free(buffer);
2290 return NULL;
2292 free(buffer);
2293 /* Now we have the ID of the referred-to object in
2294 * actual_sha1. Check again. */
2298 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2299 const char *type, unsigned char *sha1,
2300 char *hdr, int *hdrlen)
2302 git_SHA_CTX c;
2304 /* Generate the header */
2305 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2307 /* Sha1.. */
2308 git_SHA1_Init(&c);
2309 git_SHA1_Update(&c, hdr, *hdrlen);
2310 git_SHA1_Update(&c, buf, len);
2311 git_SHA1_Final(sha1, &c);
2315 * Move the just written object into its final resting place.
2316 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2317 * "moving" is only a part of what it does, when no patch between
2318 * master to pu changes the call sites of this function.
2320 int move_temp_to_file(const char *tmpfile, const char *filename)
2322 int ret = 0;
2324 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2325 goto try_rename;
2326 else if (link(tmpfile, filename))
2327 ret = errno;
2330 * Coda hack - coda doesn't like cross-directory links,
2331 * so we fall back to a rename, which will mean that it
2332 * won't be able to check collisions, but that's not a
2333 * big deal.
2335 * The same holds for FAT formatted media.
2337 * When this succeeds, we just return. We have nothing
2338 * left to unlink.
2340 if (ret && ret != EEXIST) {
2341 try_rename:
2342 if (!rename(tmpfile, filename))
2343 goto out;
2344 ret = errno;
2346 unlink_or_warn(tmpfile);
2347 if (ret) {
2348 if (ret != EEXIST) {
2349 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2351 /* FIXME!!! Collision check here ? */
2354 out:
2355 if (adjust_shared_perm(filename))
2356 return error("unable to set permission to '%s'", filename);
2357 return 0;
2360 static int write_buffer(int fd, const void *buf, size_t len)
2362 if (write_in_full(fd, buf, len) < 0)
2363 return error("file write error (%s)", strerror(errno));
2364 return 0;
2367 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2368 unsigned char *sha1)
2370 char hdr[32];
2371 int hdrlen;
2372 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2373 return 0;
2376 /* Finalize a file on disk, and close it. */
2377 static void close_sha1_file(int fd)
2379 if (fsync_object_files)
2380 fsync_or_die(fd, "sha1 file");
2381 if (close(fd) != 0)
2382 die_errno("error when closing sha1 file");
2385 /* Size of directory component, including the ending '/' */
2386 static inline int directory_size(const char *filename)
2388 const char *s = strrchr(filename, '/');
2389 if (!s)
2390 return 0;
2391 return s - filename + 1;
2395 * This creates a temporary file in the same directory as the final
2396 * 'filename'
2398 * We want to avoid cross-directory filename renames, because those
2399 * can have problems on various filesystems (FAT, NFS, Coda).
2401 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2403 int fd, dirlen = directory_size(filename);
2405 if (dirlen + 20 > bufsiz) {
2406 errno = ENAMETOOLONG;
2407 return -1;
2409 memcpy(buffer, filename, dirlen);
2410 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2411 fd = git_mkstemp_mode(buffer, 0444);
2412 if (fd < 0 && dirlen && errno == ENOENT) {
2413 /* Make sure the directory exists */
2414 memcpy(buffer, filename, dirlen);
2415 buffer[dirlen-1] = 0;
2416 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2417 return -1;
2419 /* Try again */
2420 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2421 fd = git_mkstemp_mode(buffer, 0444);
2423 return fd;
2426 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2427 const void *buf, unsigned long len, time_t mtime)
2429 int fd, ret;
2430 unsigned char compressed[4096];
2431 git_zstream stream;
2432 git_SHA_CTX c;
2433 unsigned char parano_sha1[20];
2434 char *filename;
2435 static char tmpfile[PATH_MAX];
2437 filename = sha1_file_name(sha1);
2438 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2439 if (fd < 0) {
2440 if (errno == EACCES)
2441 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2442 else
2443 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2446 /* Set it up */
2447 memset(&stream, 0, sizeof(stream));
2448 git_deflate_init(&stream, zlib_compression_level);
2449 stream.next_out = compressed;
2450 stream.avail_out = sizeof(compressed);
2451 git_SHA1_Init(&c);
2453 /* First header.. */
2454 stream.next_in = (unsigned char *)hdr;
2455 stream.avail_in = hdrlen;
2456 while (git_deflate(&stream, 0) == Z_OK)
2457 ; /* nothing */
2458 git_SHA1_Update(&c, hdr, hdrlen);
2460 /* Then the data itself.. */
2461 stream.next_in = (void *)buf;
2462 stream.avail_in = len;
2463 do {
2464 unsigned char *in0 = stream.next_in;
2465 ret = git_deflate(&stream, Z_FINISH);
2466 git_SHA1_Update(&c, in0, stream.next_in - in0);
2467 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2468 die("unable to write sha1 file");
2469 stream.next_out = compressed;
2470 stream.avail_out = sizeof(compressed);
2471 } while (ret == Z_OK);
2473 if (ret != Z_STREAM_END)
2474 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2475 ret = git_deflate_end_gently(&stream);
2476 if (ret != Z_OK)
2477 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2478 git_SHA1_Final(parano_sha1, &c);
2479 if (hashcmp(sha1, parano_sha1) != 0)
2480 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2482 close_sha1_file(fd);
2484 if (mtime) {
2485 struct utimbuf utb;
2486 utb.actime = mtime;
2487 utb.modtime = mtime;
2488 if (utime(tmpfile, &utb) < 0)
2489 warning("failed utime() on %s: %s",
2490 tmpfile, strerror(errno));
2493 return move_temp_to_file(tmpfile, filename);
2496 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2498 unsigned char sha1[20];
2499 char hdr[32];
2500 int hdrlen;
2502 /* Normally if we have it in the pack then we do not bother writing
2503 * it out into .git/objects/??/?{38} file.
2505 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2506 if (returnsha1)
2507 hashcpy(returnsha1, sha1);
2508 if (has_sha1_file(sha1))
2509 return 0;
2510 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2513 int force_object_loose(const unsigned char *sha1, time_t mtime)
2515 void *buf;
2516 unsigned long len;
2517 enum object_type type;
2518 char hdr[32];
2519 int hdrlen;
2520 int ret;
2522 if (has_loose_object(sha1))
2523 return 0;
2524 buf = read_packed_sha1(sha1, &type, &len);
2525 if (!buf)
2526 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2527 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2528 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2529 free(buf);
2531 return ret;
2534 int has_pack_index(const unsigned char *sha1)
2536 struct stat st;
2537 if (stat(sha1_pack_index_name(sha1), &st))
2538 return 0;
2539 return 1;
2542 int has_sha1_pack(const unsigned char *sha1)
2544 struct pack_entry e;
2545 return find_pack_entry(sha1, &e);
2548 int has_sha1_file(const unsigned char *sha1)
2550 struct pack_entry e;
2552 if (find_pack_entry(sha1, &e))
2553 return 1;
2554 return has_loose_object(sha1);
2557 static void check_tree(const void *buf, size_t size)
2559 struct tree_desc desc;
2560 struct name_entry entry;
2562 init_tree_desc(&desc, buf, size);
2563 while (tree_entry(&desc, &entry))
2564 /* do nothing
2565 * tree_entry() will die() on malformed entries */
2569 static void check_commit(const void *buf, size_t size)
2571 struct commit c;
2572 memset(&c, 0, sizeof(c));
2573 if (parse_commit_buffer(&c, buf, size))
2574 die("corrupt commit");
2577 static void check_tag(const void *buf, size_t size)
2579 struct tag t;
2580 memset(&t, 0, sizeof(t));
2581 if (parse_tag_buffer(&t, buf, size))
2582 die("corrupt tag");
2585 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2586 enum object_type type,
2587 const char *path, unsigned flags)
2589 int ret, re_allocated = 0;
2590 int write_object = flags & HASH_WRITE_OBJECT;
2592 if (!type)
2593 type = OBJ_BLOB;
2596 * Convert blobs to git internal format
2598 if ((type == OBJ_BLOB) && path) {
2599 struct strbuf nbuf = STRBUF_INIT;
2600 if (convert_to_git(path, buf, size, &nbuf,
2601 write_object ? safe_crlf : 0)) {
2602 buf = strbuf_detach(&nbuf, &size);
2603 re_allocated = 1;
2606 if (flags & HASH_FORMAT_CHECK) {
2607 if (type == OBJ_TREE)
2608 check_tree(buf, size);
2609 if (type == OBJ_COMMIT)
2610 check_commit(buf, size);
2611 if (type == OBJ_TAG)
2612 check_tag(buf, size);
2615 if (write_object)
2616 ret = write_sha1_file(buf, size, typename(type), sha1);
2617 else
2618 ret = hash_sha1_file(buf, size, typename(type), sha1);
2619 if (re_allocated)
2620 free(buf);
2621 return ret;
2624 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2625 const char *path, unsigned flags)
2627 struct strbuf sbuf = STRBUF_INIT;
2628 int ret;
2630 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2631 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2632 else
2633 ret = -1;
2634 strbuf_release(&sbuf);
2635 return ret;
2638 #define SMALL_FILE_SIZE (32*1024)
2640 static int index_core(unsigned char *sha1, int fd, size_t size,
2641 enum object_type type, const char *path,
2642 unsigned flags)
2644 int ret;
2646 if (!size) {
2647 ret = index_mem(sha1, NULL, size, type, path, flags);
2648 } else if (size <= SMALL_FILE_SIZE) {
2649 char *buf = xmalloc(size);
2650 if (size == read_in_full(fd, buf, size))
2651 ret = index_mem(sha1, buf, size, type, path, flags);
2652 else
2653 ret = error("short read %s", strerror(errno));
2654 free(buf);
2655 } else {
2656 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2657 ret = index_mem(sha1, buf, size, type, path, flags);
2658 munmap(buf, size);
2660 return ret;
2664 * This creates one packfile per large blob, because the caller
2665 * immediately wants the result sha1, and fast-import can report the
2666 * object name via marks mechanism only by closing the created
2667 * packfile.
2669 * This also bypasses the usual "convert-to-git" dance, and that is on
2670 * purpose. We could write a streaming version of the converting
2671 * functions and insert that before feeding the data to fast-import
2672 * (or equivalent in-core API described above), but the primary
2673 * motivation for trying to stream from the working tree file and to
2674 * avoid mmaping it in core is to deal with large binary blobs, and
2675 * by definition they do _not_ want to get any conversion.
2677 static int index_stream(unsigned char *sha1, int fd, size_t size,
2678 enum object_type type, const char *path,
2679 unsigned flags)
2681 struct child_process fast_import;
2682 char export_marks[512];
2683 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2684 char tmpfile[512];
2685 char fast_import_cmd[512];
2686 char buf[512];
2687 int len, tmpfd;
2689 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2690 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2691 if (tmpfd < 0)
2692 die_errno("cannot create tempfile: %s", tmpfile);
2693 if (close(tmpfd))
2694 die_errno("cannot close tempfile: %s", tmpfile);
2695 sprintf(export_marks, "--export-marks=%s", tmpfile);
2697 memset(&fast_import, 0, sizeof(fast_import));
2698 fast_import.in = -1;
2699 fast_import.argv = argv;
2700 fast_import.git_cmd = 1;
2701 if (start_command(&fast_import))
2702 die_errno("index-stream: git fast-import failed");
2704 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2705 (unsigned long) size);
2706 write_or_whine(fast_import.in, fast_import_cmd, len,
2707 "index-stream: feeding fast-import");
2708 while (size) {
2709 char buf[10240];
2710 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2711 size_t actual;
2713 actual = read_in_full(fd, buf, sz);
2714 if (actual < 0)
2715 die_errno("index-stream: reading input");
2716 if (write_in_full(fast_import.in, buf, actual) != actual)
2717 die_errno("index-stream: feeding fast-import");
2718 size -= actual;
2720 if (close(fast_import.in))
2721 die_errno("index-stream: closing fast-import");
2722 if (finish_command(&fast_import))
2723 die_errno("index-stream: finishing fast-import");
2725 tmpfd = open(tmpfile, O_RDONLY);
2726 if (tmpfd < 0)
2727 die_errno("index-stream: cannot open fast-import mark");
2728 len = read(tmpfd, buf, sizeof(buf));
2729 if (len < 0)
2730 die_errno("index-stream: reading fast-import mark");
2731 if (close(tmpfd) < 0)
2732 die_errno("index-stream: closing fast-import mark");
2733 if (unlink(tmpfile))
2734 die_errno("index-stream: unlinking fast-import mark");
2735 if (len != 44 ||
2736 memcmp(":1 ", buf, 3) ||
2737 get_sha1_hex(buf + 3, sha1))
2738 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2739 return 0;
2742 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2743 enum object_type type, const char *path, unsigned flags)
2745 int ret;
2746 size_t size = xsize_t(st->st_size);
2748 if (!S_ISREG(st->st_mode))
2749 ret = index_pipe(sha1, fd, type, path, flags);
2750 else if (size <= big_file_threshold || type != OBJ_BLOB)
2751 ret = index_core(sha1, fd, size, type, path, flags);
2752 else
2753 ret = index_stream(sha1, fd, size, type, path, flags);
2754 close(fd);
2755 return ret;
2758 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2760 int fd;
2761 struct strbuf sb = STRBUF_INIT;
2763 switch (st->st_mode & S_IFMT) {
2764 case S_IFREG:
2765 fd = open(path, O_RDONLY);
2766 if (fd < 0)
2767 return error("open(\"%s\"): %s", path,
2768 strerror(errno));
2769 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2770 return error("%s: failed to insert into database",
2771 path);
2772 break;
2773 case S_IFLNK:
2774 if (strbuf_readlink(&sb, path, st->st_size)) {
2775 char *errstr = strerror(errno);
2776 return error("readlink(\"%s\"): %s", path,
2777 errstr);
2779 if (!(flags & HASH_WRITE_OBJECT))
2780 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2781 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2782 return error("%s: failed to insert into database",
2783 path);
2784 strbuf_release(&sb);
2785 break;
2786 case S_IFDIR:
2787 return resolve_gitlink_ref(path, "HEAD", sha1);
2788 default:
2789 return error("%s: unsupported file type", path);
2791 return 0;
2794 int read_pack_header(int fd, struct pack_header *header)
2796 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2797 /* "eof before pack header was fully read" */
2798 return PH_ERROR_EOF;
2800 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2801 /* "protocol error (pack signature mismatch detected)" */
2802 return PH_ERROR_PACK_SIGNATURE;
2803 if (!pack_version_ok(header->hdr_version))
2804 /* "protocol error (pack version unsupported)" */
2805 return PH_ERROR_PROTOCOL;
2806 return 0;
2809 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2811 enum object_type type = sha1_object_info(sha1, NULL);
2812 if (type < 0)
2813 die("%s is not a valid object", sha1_to_hex(sha1));
2814 if (type != expect)
2815 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2816 typename(expect));