submodule: Use cat instead of echo to avoid DOS line-endings
[git/dscho.git] / sha1_file.c
blob34013014442e18bd02ae0ce33df34a89ec8d6171
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 int pfxlen, entlen;
252 struct strbuf pathbuf = STRBUF_INIT;
254 if (!is_absolute_path(entry) && relative_base) {
255 strbuf_addstr(&pathbuf, real_path(relative_base));
256 strbuf_addch(&pathbuf, '/');
258 strbuf_add(&pathbuf, entry, len);
260 normalize_path_copy(pathbuf.buf, pathbuf.buf);
262 pfxlen = strlen(pathbuf.buf);
265 * The trailing slash after the directory name is given by
266 * this function at the end. Remove duplicates.
268 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
269 pfxlen -= 1;
271 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
272 ent = xmalloc(sizeof(*ent) + entlen);
273 memcpy(ent->base, pathbuf.buf, pfxlen);
274 strbuf_release(&pathbuf);
276 ent->name = ent->base + pfxlen + 1;
277 ent->base[pfxlen + 3] = '/';
278 ent->base[pfxlen] = ent->base[entlen-1] = 0;
280 /* Detect cases where alternate disappeared */
281 if (!is_directory(ent->base)) {
282 error("object directory %s does not exist; "
283 "check .git/objects/info/alternates.",
284 ent->base);
285 free(ent);
286 return -1;
289 /* Prevent the common mistake of listing the same
290 * thing twice, or object directory itself.
292 for (alt = alt_odb_list; alt; alt = alt->next) {
293 if (!memcmp(ent->base, alt->base, pfxlen)) {
294 free(ent);
295 return -1;
298 if (!memcmp(ent->base, objdir, pfxlen)) {
299 free(ent);
300 return -1;
303 /* add the alternate entry */
304 *alt_odb_tail = ent;
305 alt_odb_tail = &(ent->next);
306 ent->next = NULL;
308 /* recursively add alternates */
309 read_info_alternates(ent->base, depth + 1);
311 ent->base[pfxlen] = '/';
313 return 0;
316 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
317 const char *relative_base, int depth)
319 const char *cp, *last;
321 if (depth > 5) {
322 error("%s: ignoring alternate object stores, nesting too deep.",
323 relative_base);
324 return;
327 last = alt;
328 while (last < ep) {
329 cp = last;
330 if (cp < ep && *cp == '#') {
331 while (cp < ep && *cp != sep)
332 cp++;
333 last = cp + 1;
334 continue;
336 while (cp < ep && *cp != sep)
337 cp++;
338 if (last != cp) {
339 if (!is_absolute_path(last) && depth) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, last);
342 } else {
343 link_alt_odb_entry(last, cp - last,
344 relative_base, depth);
347 while (cp < ep && *cp == sep)
348 cp++;
349 last = cp;
353 static void read_info_alternates(const char * relative_base, int depth)
355 char *map;
356 size_t mapsz;
357 struct stat st;
358 const char alt_file_name[] = "info/alternates";
359 /* Given that relative_base is no longer than PATH_MAX,
360 ensure that "path" has enough space to append "/", the
361 file name, "info/alternates", and a trailing NUL. */
362 char path[PATH_MAX + 1 + sizeof alt_file_name];
363 int fd;
365 sprintf(path, "%s/%s", relative_base, alt_file_name);
366 fd = git_open_noatime(path);
367 if (fd < 0)
368 return;
369 if (fstat(fd, &st) || (st.st_size == 0)) {
370 close(fd);
371 return;
373 mapsz = xsize_t(st.st_size);
374 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
375 close(fd);
377 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
379 munmap(map, mapsz);
382 void add_to_alternates_file(const char *reference)
384 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
385 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
386 char *alt = mkpath("%s\n", reference);
387 write_or_die(fd, alt, strlen(alt));
388 if (commit_lock_file(lock))
389 die("could not close alternates file");
390 if (alt_odb_tail)
391 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
394 void foreach_alt_odb(alt_odb_fn fn, void *cb)
396 struct alternate_object_database *ent;
398 prepare_alt_odb();
399 for (ent = alt_odb_list; ent; ent = ent->next)
400 if (fn(ent, cb))
401 return;
404 void prepare_alt_odb(void)
406 const char *alt;
408 if (alt_odb_tail)
409 return;
411 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
412 if (!alt) alt = "";
414 alt_odb_tail = &alt_odb_list;
415 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
417 read_info_alternates(get_object_directory(), 0);
420 static int has_loose_object_local(const unsigned char *sha1)
422 char *name = sha1_file_name(sha1);
423 return !access(name, F_OK);
426 int has_loose_object_nonlocal(const unsigned char *sha1)
428 struct alternate_object_database *alt;
429 prepare_alt_odb();
430 for (alt = alt_odb_list; alt; alt = alt->next) {
431 fill_sha1_path(alt->name, sha1);
432 if (!access(alt->base, F_OK))
433 return 1;
435 return 0;
438 static int has_loose_object(const unsigned char *sha1)
440 return has_loose_object_local(sha1) ||
441 has_loose_object_nonlocal(sha1);
444 static unsigned int pack_used_ctr;
445 static unsigned int pack_mmap_calls;
446 static unsigned int peak_pack_open_windows;
447 static unsigned int pack_open_windows;
448 static unsigned int pack_open_fds;
449 static unsigned int pack_max_fds;
450 static size_t peak_pack_mapped;
451 static size_t pack_mapped;
452 struct packed_git *packed_git;
454 void pack_report(void)
456 fprintf(stderr,
457 "pack_report: getpagesize() = %10" SZ_FMT "\n"
458 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
459 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
460 sz_fmt(getpagesize()),
461 sz_fmt(packed_git_window_size),
462 sz_fmt(packed_git_limit));
463 fprintf(stderr,
464 "pack_report: pack_used_ctr = %10u\n"
465 "pack_report: pack_mmap_calls = %10u\n"
466 "pack_report: pack_open_windows = %10u / %10u\n"
467 "pack_report: pack_mapped = "
468 "%10" SZ_FMT " / %10" SZ_FMT "\n",
469 pack_used_ctr,
470 pack_mmap_calls,
471 pack_open_windows, peak_pack_open_windows,
472 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
475 static int check_packed_git_idx(const char *path, struct packed_git *p)
477 void *idx_map;
478 struct pack_idx_header *hdr;
479 size_t idx_size;
480 uint32_t version, nr, i, *index;
481 int fd = git_open_noatime(path);
482 struct stat st;
484 if (fd < 0)
485 return -1;
486 if (fstat(fd, &st)) {
487 close(fd);
488 return -1;
490 idx_size = xsize_t(st.st_size);
491 if (idx_size < 4 * 256 + 20 + 20) {
492 close(fd);
493 return error("index file %s is too small", path);
495 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
496 close(fd);
498 hdr = idx_map;
499 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
500 version = ntohl(hdr->idx_version);
501 if (version < 2 || version > 2) {
502 munmap(idx_map, idx_size);
503 return error("index file %s is version %"PRIu32
504 " and is not supported by this binary"
505 " (try upgrading GIT to a newer version)",
506 path, version);
508 } else
509 version = 1;
511 nr = 0;
512 index = idx_map;
513 if (version > 1)
514 index += 2; /* skip index header */
515 for (i = 0; i < 256; i++) {
516 uint32_t n = ntohl(index[i]);
517 if (n < nr) {
518 munmap(idx_map, idx_size);
519 return error("non-monotonic index %s", path);
521 nr = n;
524 if (version == 1) {
526 * Total size:
527 * - 256 index entries 4 bytes each
528 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
529 * - 20-byte SHA1 of the packfile
530 * - 20-byte SHA1 file checksum
532 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
533 munmap(idx_map, idx_size);
534 return error("wrong index v1 file size in %s", path);
536 } else if (version == 2) {
538 * Minimum size:
539 * - 8 bytes of header
540 * - 256 index entries 4 bytes each
541 * - 20-byte sha1 entry * nr
542 * - 4-byte crc entry * nr
543 * - 4-byte offset entry * nr
544 * - 20-byte SHA1 of the packfile
545 * - 20-byte SHA1 file checksum
546 * And after the 4-byte offset table might be a
547 * variable sized table containing 8-byte entries
548 * for offsets larger than 2^31.
550 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
551 unsigned long max_size = min_size;
552 if (nr)
553 max_size += (nr - 1)*8;
554 if (idx_size < min_size || idx_size > max_size) {
555 munmap(idx_map, idx_size);
556 return error("wrong index v2 file size in %s", path);
558 if (idx_size != min_size &&
560 * make sure we can deal with large pack offsets.
561 * 31-bit signed offset won't be enough, neither
562 * 32-bit unsigned one will be.
564 (sizeof(off_t) <= 4)) {
565 munmap(idx_map, idx_size);
566 return error("pack too large for current definition of off_t in %s", path);
570 p->index_version = version;
571 p->index_data = idx_map;
572 p->index_size = idx_size;
573 p->num_objects = nr;
574 return 0;
577 int open_pack_index(struct packed_git *p)
579 char *idx_name;
580 int ret;
582 if (p->index_data)
583 return 0;
585 idx_name = xstrdup(p->pack_name);
586 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
587 ret = check_packed_git_idx(idx_name, p);
588 free(idx_name);
589 return ret;
592 static void scan_windows(struct packed_git *p,
593 struct packed_git **lru_p,
594 struct pack_window **lru_w,
595 struct pack_window **lru_l)
597 struct pack_window *w, *w_l;
599 for (w_l = NULL, w = p->windows; w; w = w->next) {
600 if (!w->inuse_cnt) {
601 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
602 *lru_p = p;
603 *lru_w = w;
604 *lru_l = w_l;
607 w_l = w;
611 static int unuse_one_window(struct packed_git *current, int keep_fd)
613 struct packed_git *p, *lru_p = NULL;
614 struct pack_window *lru_w = NULL, *lru_l = NULL;
616 if (current)
617 scan_windows(current, &lru_p, &lru_w, &lru_l);
618 for (p = packed_git; p; p = p->next)
619 scan_windows(p, &lru_p, &lru_w, &lru_l);
620 if (lru_p) {
621 munmap(lru_w->base, lru_w->len);
622 pack_mapped -= lru_w->len;
623 if (lru_l)
624 lru_l->next = lru_w->next;
625 else {
626 lru_p->windows = lru_w->next;
627 if (!lru_p->windows && lru_p->pack_fd != -1
628 && lru_p->pack_fd != keep_fd) {
629 close(lru_p->pack_fd);
630 pack_open_fds--;
631 lru_p->pack_fd = -1;
634 free(lru_w);
635 pack_open_windows--;
636 return 1;
638 return 0;
641 void release_pack_memory(size_t need, int fd)
643 size_t cur = pack_mapped;
644 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
645 ; /* nothing */
648 void *xmmap(void *start, size_t length,
649 int prot, int flags, int fd, off_t offset)
651 void *ret = mmap(start, length, prot, flags, fd, offset);
652 if (ret == MAP_FAILED) {
653 if (!length)
654 return NULL;
655 release_pack_memory(length, fd);
656 ret = mmap(start, length, prot, flags, fd, offset);
657 if (ret == MAP_FAILED)
658 die_errno("Out of memory? mmap failed");
660 return ret;
663 void close_pack_windows(struct packed_git *p)
665 while (p->windows) {
666 struct pack_window *w = p->windows;
668 if (w->inuse_cnt)
669 die("pack '%s' still has open windows to it",
670 p->pack_name);
671 munmap(w->base, w->len);
672 pack_mapped -= w->len;
673 pack_open_windows--;
674 p->windows = w->next;
675 free(w);
679 void unuse_pack(struct pack_window **w_cursor)
681 struct pack_window *w = *w_cursor;
682 if (w) {
683 w->inuse_cnt--;
684 *w_cursor = NULL;
688 void close_pack_index(struct packed_git *p)
690 if (p->index_data) {
691 munmap((void *)p->index_data, p->index_size);
692 p->index_data = NULL;
697 * This is used by git-repack in case a newly created pack happens to
698 * contain the same set of objects as an existing one. In that case
699 * the resulting file might be different even if its name would be the
700 * same. It is best to close any reference to the old pack before it is
701 * replaced on disk. Of course no index pointers nor windows for given pack
702 * must subsist at this point. If ever objects from this pack are requested
703 * again, the new version of the pack will be reinitialized through
704 * reprepare_packed_git().
706 void free_pack_by_name(const char *pack_name)
708 struct packed_git *p, **pp = &packed_git;
710 while (*pp) {
711 p = *pp;
712 if (strcmp(pack_name, p->pack_name) == 0) {
713 clear_delta_base_cache();
714 close_pack_windows(p);
715 if (p->pack_fd != -1) {
716 close(p->pack_fd);
717 pack_open_fds--;
719 close_pack_index(p);
720 free(p->bad_object_sha1);
721 *pp = p->next;
722 free(p);
723 return;
725 pp = &p->next;
730 * Do not call this directly as this leaks p->pack_fd on error return;
731 * call open_packed_git() instead.
733 static int open_packed_git_1(struct packed_git *p)
735 struct stat st;
736 struct pack_header hdr;
737 unsigned char sha1[20];
738 unsigned char *idx_sha1;
739 long fd_flag;
741 if (!p->index_data && open_pack_index(p))
742 return error("packfile %s index unavailable", p->pack_name);
744 if (!pack_max_fds) {
745 struct rlimit lim;
746 unsigned int max_fds;
748 if (getrlimit(RLIMIT_NOFILE, &lim))
749 die_errno("cannot get RLIMIT_NOFILE");
751 max_fds = lim.rlim_cur;
753 /* Save 3 for stdin/stdout/stderr, 22 for work */
754 if (25 < max_fds)
755 pack_max_fds = max_fds - 25;
756 else
757 pack_max_fds = 1;
760 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
761 ; /* nothing */
763 p->pack_fd = git_open_noatime(p->pack_name);
764 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
765 return -1;
766 pack_open_fds++;
768 /* If we created the struct before we had the pack we lack size. */
769 if (!p->pack_size) {
770 if (!S_ISREG(st.st_mode))
771 return error("packfile %s not a regular file", p->pack_name);
772 p->pack_size = st.st_size;
773 } else if (p->pack_size != st.st_size)
774 return error("packfile %s size changed", p->pack_name);
776 /* We leave these file descriptors open with sliding mmap;
777 * there is no point keeping them open across exec(), though.
779 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
780 if (fd_flag < 0)
781 return error("cannot determine file descriptor flags");
782 fd_flag |= FD_CLOEXEC;
783 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
784 return error("cannot set FD_CLOEXEC");
786 /* Verify we recognize this pack file format. */
787 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
788 return error("file %s is far too short to be a packfile", p->pack_name);
789 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
790 return error("file %s is not a GIT packfile", p->pack_name);
791 if (!pack_version_ok(hdr.hdr_version))
792 return error("packfile %s is version %"PRIu32" and not"
793 " supported (try upgrading GIT to a newer version)",
794 p->pack_name, ntohl(hdr.hdr_version));
796 /* Verify the pack matches its index. */
797 if (p->num_objects != ntohl(hdr.hdr_entries))
798 return error("packfile %s claims to have %"PRIu32" objects"
799 " while index indicates %"PRIu32" objects",
800 p->pack_name, ntohl(hdr.hdr_entries),
801 p->num_objects);
802 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
803 return error("end of packfile %s is unavailable", p->pack_name);
804 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
805 return error("packfile %s signature is unavailable", p->pack_name);
806 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
807 if (hashcmp(sha1, idx_sha1))
808 return error("packfile %s does not match index", p->pack_name);
809 return 0;
812 static int open_packed_git(struct packed_git *p)
814 if (!open_packed_git_1(p))
815 return 0;
816 if (p->pack_fd != -1) {
817 close(p->pack_fd);
818 pack_open_fds--;
819 p->pack_fd = -1;
821 return -1;
824 static int in_window(struct pack_window *win, off_t offset)
826 /* We must promise at least 20 bytes (one hash) after the
827 * offset is available from this window, otherwise the offset
828 * is not actually in this window and a different window (which
829 * has that one hash excess) must be used. This is to support
830 * the object header and delta base parsing routines below.
832 off_t win_off = win->offset;
833 return win_off <= offset
834 && (offset + 20) <= (win_off + win->len);
837 unsigned char *use_pack(struct packed_git *p,
838 struct pack_window **w_cursor,
839 off_t offset,
840 unsigned long *left)
842 struct pack_window *win = *w_cursor;
844 /* Since packfiles end in a hash of their content and it's
845 * pointless to ask for an offset into the middle of that
846 * hash, and the in_window function above wouldn't match
847 * don't allow an offset too close to the end of the file.
849 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
850 die("packfile %s cannot be accessed", p->pack_name);
851 if (offset > (p->pack_size - 20))
852 die("offset beyond end of packfile (truncated pack?)");
854 if (!win || !in_window(win, offset)) {
855 if (win)
856 win->inuse_cnt--;
857 for (win = p->windows; win; win = win->next) {
858 if (in_window(win, offset))
859 break;
861 if (!win) {
862 size_t window_align = packed_git_window_size / 2;
863 off_t len;
865 if (p->pack_fd == -1 && open_packed_git(p))
866 die("packfile %s cannot be accessed", p->pack_name);
868 win = xcalloc(1, sizeof(*win));
869 win->offset = (offset / window_align) * window_align;
870 len = p->pack_size - win->offset;
871 if (len > packed_git_window_size)
872 len = packed_git_window_size;
873 win->len = (size_t)len;
874 pack_mapped += win->len;
875 while (packed_git_limit < pack_mapped
876 && unuse_one_window(p, p->pack_fd))
877 ; /* nothing */
878 win->base = xmmap(NULL, win->len,
879 PROT_READ, MAP_PRIVATE,
880 p->pack_fd, win->offset);
881 if (win->base == MAP_FAILED)
882 die("packfile %s cannot be mapped: %s",
883 p->pack_name,
884 strerror(errno));
885 if (!win->offset && win->len == p->pack_size
886 && !p->do_not_close) {
887 close(p->pack_fd);
888 pack_open_fds--;
889 p->pack_fd = -1;
891 pack_mmap_calls++;
892 pack_open_windows++;
893 if (pack_mapped > peak_pack_mapped)
894 peak_pack_mapped = pack_mapped;
895 if (pack_open_windows > peak_pack_open_windows)
896 peak_pack_open_windows = pack_open_windows;
897 win->next = p->windows;
898 p->windows = win;
901 if (win != *w_cursor) {
902 win->last_used = pack_used_ctr++;
903 win->inuse_cnt++;
904 *w_cursor = win;
906 offset -= win->offset;
907 if (left)
908 *left = win->len - xsize_t(offset);
909 return win->base + offset;
912 static struct packed_git *alloc_packed_git(int extra)
914 struct packed_git *p = xmalloc(sizeof(*p) + extra);
915 memset(p, 0, sizeof(*p));
916 p->pack_fd = -1;
917 return p;
920 static void try_to_free_pack_memory(size_t size)
922 release_pack_memory(size, -1);
925 struct packed_git *add_packed_git(const char *path, int path_len, int local)
927 static int have_set_try_to_free_routine;
928 struct stat st;
929 struct packed_git *p = alloc_packed_git(path_len + 2);
931 if (!have_set_try_to_free_routine) {
932 have_set_try_to_free_routine = 1;
933 set_try_to_free_routine(try_to_free_pack_memory);
937 * Make sure a corresponding .pack file exists and that
938 * the index looks sane.
940 path_len -= strlen(".idx");
941 if (path_len < 1) {
942 free(p);
943 return NULL;
945 memcpy(p->pack_name, path, path_len);
947 strcpy(p->pack_name + path_len, ".keep");
948 if (!access(p->pack_name, F_OK))
949 p->pack_keep = 1;
951 strcpy(p->pack_name + path_len, ".pack");
952 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
953 free(p);
954 return NULL;
957 /* ok, it looks sane as far as we can check without
958 * actually mapping the pack file.
960 p->pack_size = st.st_size;
961 p->pack_local = local;
962 p->mtime = st.st_mtime;
963 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
964 hashclr(p->sha1);
965 return p;
968 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
970 const char *path = sha1_pack_name(sha1);
971 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
973 strcpy(p->pack_name, path);
974 hashcpy(p->sha1, sha1);
975 if (check_packed_git_idx(idx_path, p)) {
976 free(p);
977 return NULL;
980 return p;
983 void install_packed_git(struct packed_git *pack)
985 if (pack->pack_fd != -1)
986 pack_open_fds++;
988 pack->next = packed_git;
989 packed_git = pack;
992 static void prepare_packed_git_one(char *objdir, int local)
994 /* Ensure that this buffer is large enough so that we can
995 append "/pack/" without clobbering the stack even if
996 strlen(objdir) were PATH_MAX. */
997 char path[PATH_MAX + 1 + 4 + 1 + 1];
998 int len;
999 DIR *dir;
1000 struct dirent *de;
1002 sprintf(path, "%s/pack", objdir);
1003 len = strlen(path);
1004 dir = opendir(path);
1005 if (!dir) {
1006 if (errno != ENOENT)
1007 error("unable to open object pack directory: %s: %s",
1008 path, strerror(errno));
1009 return;
1011 path[len++] = '/';
1012 while ((de = readdir(dir)) != NULL) {
1013 int namelen = strlen(de->d_name);
1014 struct packed_git *p;
1016 if (!has_extension(de->d_name, ".idx"))
1017 continue;
1019 if (len + namelen + 1 > sizeof(path))
1020 continue;
1022 /* Don't reopen a pack we already have. */
1023 strcpy(path + len, de->d_name);
1024 for (p = packed_git; p; p = p->next) {
1025 if (!memcmp(path, p->pack_name, len + namelen - 4))
1026 break;
1028 if (p)
1029 continue;
1030 /* See if it really is a valid .idx file with corresponding
1031 * .pack file that we can map.
1033 p = add_packed_git(path, len + namelen, local);
1034 if (!p)
1035 continue;
1036 install_packed_git(p);
1038 closedir(dir);
1041 static int sort_pack(const void *a_, const void *b_)
1043 struct packed_git *a = *((struct packed_git **)a_);
1044 struct packed_git *b = *((struct packed_git **)b_);
1045 int st;
1048 * Local packs tend to contain objects specific to our
1049 * variant of the project than remote ones. In addition,
1050 * remote ones could be on a network mounted filesystem.
1051 * Favor local ones for these reasons.
1053 st = a->pack_local - b->pack_local;
1054 if (st)
1055 return -st;
1058 * Younger packs tend to contain more recent objects,
1059 * and more recent objects tend to get accessed more
1060 * often.
1062 if (a->mtime < b->mtime)
1063 return 1;
1064 else if (a->mtime == b->mtime)
1065 return 0;
1066 return -1;
1069 static void rearrange_packed_git(void)
1071 struct packed_git **ary, *p;
1072 int i, n;
1074 for (n = 0, p = packed_git; p; p = p->next)
1075 n++;
1076 if (n < 2)
1077 return;
1079 /* prepare an array of packed_git for easier sorting */
1080 ary = xcalloc(n, sizeof(struct packed_git *));
1081 for (n = 0, p = packed_git; p; p = p->next)
1082 ary[n++] = p;
1084 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1086 /* link them back again */
1087 for (i = 0; i < n - 1; i++)
1088 ary[i]->next = ary[i + 1];
1089 ary[n - 1]->next = NULL;
1090 packed_git = ary[0];
1092 free(ary);
1095 static int prepare_packed_git_run_once = 0;
1096 void prepare_packed_git(void)
1098 struct alternate_object_database *alt;
1100 if (prepare_packed_git_run_once)
1101 return;
1102 prepare_packed_git_one(get_object_directory(), 1);
1103 prepare_alt_odb();
1104 for (alt = alt_odb_list; alt; alt = alt->next) {
1105 alt->name[-1] = 0;
1106 prepare_packed_git_one(alt->base, 0);
1107 alt->name[-1] = '/';
1109 rearrange_packed_git();
1110 prepare_packed_git_run_once = 1;
1113 void reprepare_packed_git(void)
1115 discard_revindex();
1116 prepare_packed_git_run_once = 0;
1117 prepare_packed_git();
1120 static void mark_bad_packed_object(struct packed_git *p,
1121 const unsigned char *sha1)
1123 unsigned i;
1124 for (i = 0; i < p->num_bad_objects; i++)
1125 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1126 return;
1127 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1128 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1129 p->num_bad_objects++;
1132 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1134 struct packed_git *p;
1135 unsigned i;
1137 for (p = packed_git; p; p = p->next)
1138 for (i = 0; i < p->num_bad_objects; i++)
1139 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1140 return p;
1141 return NULL;
1144 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1146 unsigned char real_sha1[20];
1147 hash_sha1_file(map, size, type, real_sha1);
1148 return hashcmp(sha1, real_sha1) ? -1 : 0;
1151 static int git_open_noatime(const char *name)
1153 static int sha1_file_open_flag = O_NOATIME;
1155 for (;;) {
1156 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1157 if (fd >= 0)
1158 return fd;
1160 /* Might the failure be due to O_NOATIME? */
1161 if (errno != ENOENT && sha1_file_open_flag) {
1162 sha1_file_open_flag = 0;
1163 continue;
1166 return -1;
1170 static int open_sha1_file(const unsigned char *sha1)
1172 int fd;
1173 char *name = sha1_file_name(sha1);
1174 struct alternate_object_database *alt;
1176 fd = git_open_noatime(name);
1177 if (fd >= 0)
1178 return fd;
1180 prepare_alt_odb();
1181 errno = ENOENT;
1182 for (alt = alt_odb_list; alt; alt = alt->next) {
1183 name = alt->name;
1184 fill_sha1_path(name, sha1);
1185 fd = git_open_noatime(alt->base);
1186 if (fd >= 0)
1187 return fd;
1189 return -1;
1192 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1194 void *map;
1195 int fd;
1197 fd = open_sha1_file(sha1);
1198 map = NULL;
1199 if (fd >= 0) {
1200 struct stat st;
1202 if (!fstat(fd, &st)) {
1203 *size = xsize_t(st.st_size);
1204 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1206 close(fd);
1208 return map;
1212 * There used to be a second loose object header format which
1213 * was meant to mimic the in-pack format, allowing for direct
1214 * copy of the object data. This format turned up not to be
1215 * really worth it and we no longer write loose objects in that
1216 * format.
1218 static int experimental_loose_object(unsigned char *map)
1220 unsigned int word;
1223 * We must determine if the buffer contains the standard
1224 * zlib-deflated stream or the experimental format based
1225 * on the in-pack object format. Compare the header byte
1226 * for each format:
1228 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1229 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1231 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1232 * in standard loose-object format, UNLESS it is a Git-pack
1233 * format object *exactly* 8 bytes in size when inflated.
1235 * However, RFC1950 also specifies that the 1st 16-bit word
1236 * must be divisible by 31 - this checksum tells us our buffer
1237 * is in the standard format, giving a false positive only if
1238 * the 1st word of the Git-pack format object happens to be
1239 * divisible by 31, ie:
1240 * ((byte0 * 256) + byte1) % 31 = 0
1241 * => 0ttt10000www1000 % 31 = 0
1243 * As it happens, this case can only arise for www=3 & ttt=1
1244 * - ie, a Commit object, which would have to be 8 bytes in
1245 * size. As no Commit can be that small, we find that the
1246 * combination of these two criteria (bitmask & checksum)
1247 * can always correctly determine the buffer format.
1249 word = (map[0] << 8) + map[1];
1250 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1251 return 0;
1252 else
1253 return 1;
1256 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1257 unsigned long len, enum object_type *type, unsigned long *sizep)
1259 unsigned shift;
1260 unsigned long size, c;
1261 unsigned long used = 0;
1263 c = buf[used++];
1264 *type = (c >> 4) & 7;
1265 size = c & 15;
1266 shift = 4;
1267 while (c & 0x80) {
1268 if (len <= used || bitsizeof(long) <= shift) {
1269 error("bad object header");
1270 return 0;
1272 c = buf[used++];
1273 size += (c & 0x7f) << shift;
1274 shift += 7;
1276 *sizep = size;
1277 return used;
1280 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1282 unsigned long size, used;
1283 static const char valid_loose_object_type[8] = {
1284 0, /* OBJ_EXT */
1285 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1286 0, /* "delta" and others are invalid in a loose object */
1288 enum object_type type;
1290 /* Get the data stream */
1291 memset(stream, 0, sizeof(*stream));
1292 stream->next_in = map;
1293 stream->avail_in = mapsize;
1294 stream->next_out = buffer;
1295 stream->avail_out = bufsiz;
1297 if (experimental_loose_object(map)) {
1299 * The old experimental format we no longer produce;
1300 * we can still read it.
1302 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1303 if (!used || !valid_loose_object_type[type])
1304 return -1;
1305 map += used;
1306 mapsize -= used;
1308 /* Set up the stream for the rest.. */
1309 stream->next_in = map;
1310 stream->avail_in = mapsize;
1311 git_inflate_init(stream);
1313 /* And generate the fake traditional header */
1314 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1315 typename(type), size);
1316 return 0;
1318 git_inflate_init(stream);
1319 return git_inflate(stream, 0);
1322 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1324 int bytes = strlen(buffer) + 1;
1325 unsigned char *buf = xmallocz(size);
1326 unsigned long n;
1327 int status = Z_OK;
1329 n = stream->total_out - bytes;
1330 if (n > size)
1331 n = size;
1332 memcpy(buf, (char *) buffer + bytes, n);
1333 bytes = n;
1334 if (bytes <= size) {
1336 * The above condition must be (bytes <= size), not
1337 * (bytes < size). In other words, even though we
1338 * expect no more output and set avail_out to zero,
1339 * the input zlib stream may have bytes that express
1340 * "this concludes the stream", and we *do* want to
1341 * eat that input.
1343 * Otherwise we would not be able to test that we
1344 * consumed all the input to reach the expected size;
1345 * we also want to check that zlib tells us that all
1346 * went well with status == Z_STREAM_END at the end.
1348 stream->next_out = buf + bytes;
1349 stream->avail_out = size - bytes;
1350 while (status == Z_OK)
1351 status = git_inflate(stream, Z_FINISH);
1353 if (status == Z_STREAM_END && !stream->avail_in) {
1354 git_inflate_end(stream);
1355 return buf;
1358 if (status < 0)
1359 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1360 else if (stream->avail_in)
1361 error("garbage at end of loose object '%s'",
1362 sha1_to_hex(sha1));
1363 free(buf);
1364 return NULL;
1368 * We used to just use "sscanf()", but that's actually way
1369 * too permissive for what we want to check. So do an anal
1370 * object header parse by hand.
1372 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1374 char type[10];
1375 int i;
1376 unsigned long size;
1379 * The type can be at most ten bytes (including the
1380 * terminating '\0' that we add), and is followed by
1381 * a space.
1383 i = 0;
1384 for (;;) {
1385 char c = *hdr++;
1386 if (c == ' ')
1387 break;
1388 type[i++] = c;
1389 if (i >= sizeof(type))
1390 return -1;
1392 type[i] = 0;
1395 * The length must follow immediately, and be in canonical
1396 * decimal format (ie "010" is not valid).
1398 size = *hdr++ - '0';
1399 if (size > 9)
1400 return -1;
1401 if (size) {
1402 for (;;) {
1403 unsigned long c = *hdr - '0';
1404 if (c > 9)
1405 break;
1406 hdr++;
1407 size = size * 10 + c;
1410 *sizep = size;
1413 * The length must be followed by a zero byte
1415 return *hdr ? -1 : type_from_string(type);
1418 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1420 int ret;
1421 git_zstream stream;
1422 char hdr[8192];
1424 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1425 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1426 return NULL;
1428 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1431 unsigned long get_size_from_delta(struct packed_git *p,
1432 struct pack_window **w_curs,
1433 off_t curpos)
1435 const unsigned char *data;
1436 unsigned char delta_head[20], *in;
1437 git_zstream stream;
1438 int st;
1440 memset(&stream, 0, sizeof(stream));
1441 stream.next_out = delta_head;
1442 stream.avail_out = sizeof(delta_head);
1444 git_inflate_init(&stream);
1445 do {
1446 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1447 stream.next_in = in;
1448 st = git_inflate(&stream, Z_FINISH);
1449 curpos += stream.next_in - in;
1450 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1451 stream.total_out < sizeof(delta_head));
1452 git_inflate_end(&stream);
1453 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1454 error("delta data unpack-initial failed");
1455 return 0;
1458 /* Examine the initial part of the delta to figure out
1459 * the result size.
1461 data = delta_head;
1463 /* ignore base size */
1464 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1466 /* Read the result size */
1467 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1470 static off_t get_delta_base(struct packed_git *p,
1471 struct pack_window **w_curs,
1472 off_t *curpos,
1473 enum object_type type,
1474 off_t delta_obj_offset)
1476 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1477 off_t base_offset;
1479 /* use_pack() assured us we have [base_info, base_info + 20)
1480 * as a range that we can look at without walking off the
1481 * end of the mapped window. Its actually the hash size
1482 * that is assured. An OFS_DELTA longer than the hash size
1483 * is stupid, as then a REF_DELTA would be smaller to store.
1485 if (type == OBJ_OFS_DELTA) {
1486 unsigned used = 0;
1487 unsigned char c = base_info[used++];
1488 base_offset = c & 127;
1489 while (c & 128) {
1490 base_offset += 1;
1491 if (!base_offset || MSB(base_offset, 7))
1492 return 0; /* overflow */
1493 c = base_info[used++];
1494 base_offset = (base_offset << 7) + (c & 127);
1496 base_offset = delta_obj_offset - base_offset;
1497 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1498 return 0; /* out of bound */
1499 *curpos += used;
1500 } else if (type == OBJ_REF_DELTA) {
1501 /* The base entry _must_ be in the same pack */
1502 base_offset = find_pack_entry_one(base_info, p);
1503 *curpos += 20;
1504 } else
1505 die("I am totally screwed");
1506 return base_offset;
1509 /* forward declaration for a mutually recursive function */
1510 static int packed_object_info(struct packed_git *p, off_t offset,
1511 unsigned long *sizep, int *rtype);
1513 static int packed_delta_info(struct packed_git *p,
1514 struct pack_window **w_curs,
1515 off_t curpos,
1516 enum object_type type,
1517 off_t obj_offset,
1518 unsigned long *sizep)
1520 off_t base_offset;
1522 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1523 if (!base_offset)
1524 return OBJ_BAD;
1525 type = packed_object_info(p, base_offset, NULL, NULL);
1526 if (type <= OBJ_NONE) {
1527 struct revindex_entry *revidx;
1528 const unsigned char *base_sha1;
1529 revidx = find_pack_revindex(p, base_offset);
1530 if (!revidx)
1531 return OBJ_BAD;
1532 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1533 mark_bad_packed_object(p, base_sha1);
1534 type = sha1_object_info(base_sha1, NULL);
1535 if (type <= OBJ_NONE)
1536 return OBJ_BAD;
1539 /* We choose to only get the type of the base object and
1540 * ignore potentially corrupt pack file that expects the delta
1541 * based on a base with a wrong size. This saves tons of
1542 * inflate() calls.
1544 if (sizep) {
1545 *sizep = get_size_from_delta(p, w_curs, curpos);
1546 if (*sizep == 0)
1547 type = OBJ_BAD;
1550 return type;
1553 int unpack_object_header(struct packed_git *p,
1554 struct pack_window **w_curs,
1555 off_t *curpos,
1556 unsigned long *sizep)
1558 unsigned char *base;
1559 unsigned long left;
1560 unsigned long used;
1561 enum object_type type;
1563 /* use_pack() assures us we have [base, base + 20) available
1564 * as a range that we can look at. (Its actually the hash
1565 * size that is assured.) With our object header encoding
1566 * the maximum deflated object size is 2^137, which is just
1567 * insane, so we know won't exceed what we have been given.
1569 base = use_pack(p, w_curs, *curpos, &left);
1570 used = unpack_object_header_buffer(base, left, &type, sizep);
1571 if (!used) {
1572 type = OBJ_BAD;
1573 } else
1574 *curpos += used;
1576 return type;
1579 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1580 unsigned long *sizep, int *rtype)
1582 struct pack_window *w_curs = NULL;
1583 unsigned long size;
1584 off_t curpos = obj_offset;
1585 enum object_type type;
1587 type = unpack_object_header(p, &w_curs, &curpos, &size);
1588 if (rtype)
1589 *rtype = type; /* representation type */
1591 switch (type) {
1592 case OBJ_OFS_DELTA:
1593 case OBJ_REF_DELTA:
1594 type = packed_delta_info(p, &w_curs, curpos,
1595 type, obj_offset, sizep);
1596 break;
1597 case OBJ_COMMIT:
1598 case OBJ_TREE:
1599 case OBJ_BLOB:
1600 case OBJ_TAG:
1601 if (sizep)
1602 *sizep = size;
1603 break;
1604 default:
1605 error("unknown object type %i at offset %"PRIuMAX" in %s",
1606 type, (uintmax_t)obj_offset, p->pack_name);
1607 type = OBJ_BAD;
1609 unuse_pack(&w_curs);
1610 return type;
1613 static void *unpack_compressed_entry(struct packed_git *p,
1614 struct pack_window **w_curs,
1615 off_t curpos,
1616 unsigned long size)
1618 int st;
1619 git_zstream stream;
1620 unsigned char *buffer, *in;
1622 buffer = xmallocz(size);
1623 memset(&stream, 0, sizeof(stream));
1624 stream.next_out = buffer;
1625 stream.avail_out = size + 1;
1627 git_inflate_init(&stream);
1628 do {
1629 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1630 stream.next_in = in;
1631 st = git_inflate(&stream, Z_FINISH);
1632 if (!stream.avail_out)
1633 break; /* the payload is larger than it should be */
1634 curpos += stream.next_in - in;
1635 } while (st == Z_OK || st == Z_BUF_ERROR);
1636 git_inflate_end(&stream);
1637 if ((st != Z_STREAM_END) || stream.total_out != size) {
1638 free(buffer);
1639 return NULL;
1642 return buffer;
1645 #define MAX_DELTA_CACHE (256)
1647 static size_t delta_base_cached;
1649 static struct delta_base_cache_lru_list {
1650 struct delta_base_cache_lru_list *prev;
1651 struct delta_base_cache_lru_list *next;
1652 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1654 static struct delta_base_cache_entry {
1655 struct delta_base_cache_lru_list lru;
1656 void *data;
1657 struct packed_git *p;
1658 off_t base_offset;
1659 unsigned long size;
1660 enum object_type type;
1661 } delta_base_cache[MAX_DELTA_CACHE];
1663 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1665 unsigned long hash;
1667 hash = (unsigned long)p + (unsigned long)base_offset;
1668 hash += (hash >> 8) + (hash >> 16);
1669 return hash % MAX_DELTA_CACHE;
1672 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1674 unsigned long hash = pack_entry_hash(p, base_offset);
1675 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1676 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1679 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1680 unsigned long *base_size, enum object_type *type, int keep_cache)
1682 void *ret;
1683 unsigned long hash = pack_entry_hash(p, base_offset);
1684 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1686 ret = ent->data;
1687 if (!ret || ent->p != p || ent->base_offset != base_offset)
1688 return unpack_entry(p, base_offset, type, base_size);
1690 if (!keep_cache) {
1691 ent->data = NULL;
1692 ent->lru.next->prev = ent->lru.prev;
1693 ent->lru.prev->next = ent->lru.next;
1694 delta_base_cached -= ent->size;
1695 } else {
1696 ret = xmemdupz(ent->data, ent->size);
1698 *type = ent->type;
1699 *base_size = ent->size;
1700 return ret;
1703 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1705 if (ent->data) {
1706 free(ent->data);
1707 ent->data = NULL;
1708 ent->lru.next->prev = ent->lru.prev;
1709 ent->lru.prev->next = ent->lru.next;
1710 delta_base_cached -= ent->size;
1714 void clear_delta_base_cache(void)
1716 unsigned long p;
1717 for (p = 0; p < MAX_DELTA_CACHE; p++)
1718 release_delta_base_cache(&delta_base_cache[p]);
1721 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1722 void *base, unsigned long base_size, enum object_type type)
1724 unsigned long hash = pack_entry_hash(p, base_offset);
1725 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1726 struct delta_base_cache_lru_list *lru;
1728 release_delta_base_cache(ent);
1729 delta_base_cached += base_size;
1731 for (lru = delta_base_cache_lru.next;
1732 delta_base_cached > delta_base_cache_limit
1733 && lru != &delta_base_cache_lru;
1734 lru = lru->next) {
1735 struct delta_base_cache_entry *f = (void *)lru;
1736 if (f->type == OBJ_BLOB)
1737 release_delta_base_cache(f);
1739 for (lru = delta_base_cache_lru.next;
1740 delta_base_cached > delta_base_cache_limit
1741 && lru != &delta_base_cache_lru;
1742 lru = lru->next) {
1743 struct delta_base_cache_entry *f = (void *)lru;
1744 release_delta_base_cache(f);
1747 ent->p = p;
1748 ent->base_offset = base_offset;
1749 ent->type = type;
1750 ent->data = base;
1751 ent->size = base_size;
1752 ent->lru.next = &delta_base_cache_lru;
1753 ent->lru.prev = delta_base_cache_lru.prev;
1754 delta_base_cache_lru.prev->next = &ent->lru;
1755 delta_base_cache_lru.prev = &ent->lru;
1758 static void *read_object(const unsigned char *sha1, enum object_type *type,
1759 unsigned long *size);
1761 static void *unpack_delta_entry(struct packed_git *p,
1762 struct pack_window **w_curs,
1763 off_t curpos,
1764 unsigned long delta_size,
1765 off_t obj_offset,
1766 enum object_type *type,
1767 unsigned long *sizep)
1769 void *delta_data, *result, *base;
1770 unsigned long base_size;
1771 off_t base_offset;
1773 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1774 if (!base_offset) {
1775 error("failed to validate delta base reference "
1776 "at offset %"PRIuMAX" from %s",
1777 (uintmax_t)curpos, p->pack_name);
1778 return NULL;
1780 unuse_pack(w_curs);
1781 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1782 if (!base) {
1784 * We're probably in deep shit, but let's try to fetch
1785 * the required base anyway from another pack or loose.
1786 * This is costly but should happen only in the presence
1787 * of a corrupted pack, and is better than failing outright.
1789 struct revindex_entry *revidx;
1790 const unsigned char *base_sha1;
1791 revidx = find_pack_revindex(p, base_offset);
1792 if (!revidx)
1793 return NULL;
1794 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1795 error("failed to read delta base object %s"
1796 " at offset %"PRIuMAX" from %s",
1797 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1798 p->pack_name);
1799 mark_bad_packed_object(p, base_sha1);
1800 base = read_object(base_sha1, type, &base_size);
1801 if (!base)
1802 return NULL;
1805 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1806 if (!delta_data) {
1807 error("failed to unpack compressed delta "
1808 "at offset %"PRIuMAX" from %s",
1809 (uintmax_t)curpos, p->pack_name);
1810 free(base);
1811 return NULL;
1813 result = patch_delta(base, base_size,
1814 delta_data, delta_size,
1815 sizep);
1816 if (!result)
1817 die("failed to apply delta");
1818 free(delta_data);
1819 add_delta_base_cache(p, base_offset, base, base_size, *type);
1820 return result;
1823 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1825 static FILE *log_file;
1827 if (!log_file) {
1828 log_file = fopen(log_pack_access, "w");
1829 if (!log_file) {
1830 error("cannot open pack access log '%s' for writing: %s",
1831 log_pack_access, strerror(errno));
1832 log_pack_access = NULL;
1833 return;
1836 fprintf(log_file, "%s %"PRIuMAX"\n",
1837 p->pack_name, (uintmax_t)obj_offset);
1838 fflush(log_file);
1841 int do_check_packed_object_crc;
1843 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1844 enum object_type *type, unsigned long *sizep)
1846 struct pack_window *w_curs = NULL;
1847 off_t curpos = obj_offset;
1848 void *data;
1850 if (log_pack_access)
1851 write_pack_access_log(p, obj_offset);
1853 if (do_check_packed_object_crc && p->index_version > 1) {
1854 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1855 unsigned long len = revidx[1].offset - obj_offset;
1856 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1857 const unsigned char *sha1 =
1858 nth_packed_object_sha1(p, revidx->nr);
1859 error("bad packed object CRC for %s",
1860 sha1_to_hex(sha1));
1861 mark_bad_packed_object(p, sha1);
1862 unuse_pack(&w_curs);
1863 return NULL;
1867 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1868 switch (*type) {
1869 case OBJ_OFS_DELTA:
1870 case OBJ_REF_DELTA:
1871 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1872 obj_offset, type, sizep);
1873 break;
1874 case OBJ_COMMIT:
1875 case OBJ_TREE:
1876 case OBJ_BLOB:
1877 case OBJ_TAG:
1878 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1879 break;
1880 default:
1881 data = NULL;
1882 error("unknown object type %i at offset %"PRIuMAX" in %s",
1883 *type, (uintmax_t)obj_offset, p->pack_name);
1885 unuse_pack(&w_curs);
1886 return data;
1889 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1890 uint32_t n)
1892 const unsigned char *index = p->index_data;
1893 if (!index) {
1894 if (open_pack_index(p))
1895 return NULL;
1896 index = p->index_data;
1898 if (n >= p->num_objects)
1899 return NULL;
1900 index += 4 * 256;
1901 if (p->index_version == 1) {
1902 return index + 24 * n + 4;
1903 } else {
1904 index += 8;
1905 return index + 20 * n;
1909 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1911 const unsigned char *index = p->index_data;
1912 index += 4 * 256;
1913 if (p->index_version == 1) {
1914 return ntohl(*((uint32_t *)(index + 24 * n)));
1915 } else {
1916 uint32_t off;
1917 index += 8 + p->num_objects * (20 + 4);
1918 off = ntohl(*((uint32_t *)(index + 4 * n)));
1919 if (!(off & 0x80000000))
1920 return off;
1921 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1922 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1923 ntohl(*((uint32_t *)(index + 4)));
1927 off_t find_pack_entry_one(const unsigned char *sha1,
1928 struct packed_git *p)
1930 const uint32_t *level1_ofs = p->index_data;
1931 const unsigned char *index = p->index_data;
1932 unsigned hi, lo, stride;
1933 static int use_lookup = -1;
1934 static int debug_lookup = -1;
1936 if (debug_lookup < 0)
1937 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1939 if (!index) {
1940 if (open_pack_index(p))
1941 return 0;
1942 level1_ofs = p->index_data;
1943 index = p->index_data;
1945 if (p->index_version > 1) {
1946 level1_ofs += 2;
1947 index += 8;
1949 index += 4 * 256;
1950 hi = ntohl(level1_ofs[*sha1]);
1951 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1952 if (p->index_version > 1) {
1953 stride = 20;
1954 } else {
1955 stride = 24;
1956 index += 4;
1959 if (debug_lookup)
1960 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1961 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1963 if (use_lookup < 0)
1964 use_lookup = !!getenv("GIT_USE_LOOKUP");
1965 if (use_lookup) {
1966 int pos = sha1_entry_pos(index, stride, 0,
1967 lo, hi, p->num_objects, sha1);
1968 if (pos < 0)
1969 return 0;
1970 return nth_packed_object_offset(p, pos);
1973 do {
1974 unsigned mi = (lo + hi) / 2;
1975 int cmp = hashcmp(index + mi * stride, sha1);
1977 if (debug_lookup)
1978 printf("lo %u hi %u rg %u mi %u\n",
1979 lo, hi, hi - lo, mi);
1980 if (!cmp)
1981 return nth_packed_object_offset(p, mi);
1982 if (cmp > 0)
1983 hi = mi;
1984 else
1985 lo = mi+1;
1986 } while (lo < hi);
1987 return 0;
1990 static int is_pack_valid(struct packed_git *p)
1992 /* An already open pack is known to be valid. */
1993 if (p->pack_fd != -1)
1994 return 1;
1996 /* If the pack has one window completely covering the
1997 * file size, the pack is known to be valid even if
1998 * the descriptor is not currently open.
2000 if (p->windows) {
2001 struct pack_window *w = p->windows;
2003 if (!w->offset && w->len == p->pack_size)
2004 return 1;
2007 /* Force the pack to open to prove its valid. */
2008 return !open_packed_git(p);
2011 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2013 static struct packed_git *last_found = (void *)1;
2014 struct packed_git *p;
2015 off_t offset;
2017 prepare_packed_git();
2018 if (!packed_git)
2019 return 0;
2020 p = (last_found == (void *)1) ? packed_git : last_found;
2022 do {
2023 if (p->num_bad_objects) {
2024 unsigned i;
2025 for (i = 0; i < p->num_bad_objects; i++)
2026 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2027 goto next;
2030 offset = find_pack_entry_one(sha1, p);
2031 if (offset) {
2033 * We are about to tell the caller where they can
2034 * locate the requested object. We better make
2035 * sure the packfile is still here and can be
2036 * accessed before supplying that answer, as
2037 * it may have been deleted since the index
2038 * was loaded!
2040 if (!is_pack_valid(p)) {
2041 error("packfile %s cannot be accessed", p->pack_name);
2042 goto next;
2044 e->offset = offset;
2045 e->p = p;
2046 hashcpy(e->sha1, sha1);
2047 last_found = p;
2048 return 1;
2051 next:
2052 if (p == last_found)
2053 p = packed_git;
2054 else
2055 p = p->next;
2056 if (p == last_found)
2057 p = p->next;
2058 } while (p);
2059 return 0;
2062 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2063 struct packed_git *packs)
2065 struct packed_git *p;
2067 for (p = packs; p; p = p->next) {
2068 if (find_pack_entry_one(sha1, p))
2069 return p;
2071 return NULL;
2075 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2077 int status;
2078 unsigned long mapsize, size;
2079 void *map;
2080 git_zstream stream;
2081 char hdr[32];
2083 map = map_sha1_file(sha1, &mapsize);
2084 if (!map)
2085 return error("unable to find %s", sha1_to_hex(sha1));
2086 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2087 status = error("unable to unpack %s header",
2088 sha1_to_hex(sha1));
2089 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2090 status = error("unable to parse %s header", sha1_to_hex(sha1));
2091 else if (sizep)
2092 *sizep = size;
2093 git_inflate_end(&stream);
2094 munmap(map, mapsize);
2095 return status;
2098 /* returns enum object_type or negative */
2099 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2101 struct cached_object *co;
2102 struct pack_entry e;
2103 int status, rtype;
2105 co = find_cached_object(sha1);
2106 if (co) {
2107 if (oi->sizep)
2108 *(oi->sizep) = co->size;
2109 oi->whence = OI_CACHED;
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, oi->sizep);
2116 if (status >= 0) {
2117 oi->whence = OI_LOOSE;
2118 return status;
2121 /* Not a loose object; someone else may have just packed it. */
2122 reprepare_packed_git();
2123 if (!find_pack_entry(sha1, &e))
2124 return status;
2127 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2128 if (status < 0) {
2129 mark_bad_packed_object(e.p, sha1);
2130 status = sha1_object_info_extended(sha1, oi);
2131 } else if (in_delta_base_cache(e.p, e.offset)) {
2132 oi->whence = OI_DBCACHED;
2133 } else {
2134 oi->whence = OI_PACKED;
2135 oi->u.packed.offset = e.offset;
2136 oi->u.packed.pack = e.p;
2137 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2138 rtype == OBJ_OFS_DELTA);
2141 return status;
2144 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2146 struct object_info oi;
2148 oi.sizep = sizep;
2149 return sha1_object_info_extended(sha1, &oi);
2152 static void *read_packed_sha1(const unsigned char *sha1,
2153 enum object_type *type, unsigned long *size)
2155 struct pack_entry e;
2156 void *data;
2158 if (!find_pack_entry(sha1, &e))
2159 return NULL;
2160 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2161 if (!data) {
2163 * We're probably in deep shit, but let's try to fetch
2164 * the required object anyway from another pack or loose.
2165 * This should happen only in the presence of a corrupted
2166 * pack, and is better than failing outright.
2168 error("failed to read object %s at offset %"PRIuMAX" from %s",
2169 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2170 mark_bad_packed_object(e.p, sha1);
2171 data = read_object(sha1, type, size);
2173 return data;
2176 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2177 unsigned char *sha1)
2179 struct cached_object *co;
2181 hash_sha1_file(buf, len, typename(type), sha1);
2182 if (has_sha1_file(sha1) || find_cached_object(sha1))
2183 return 0;
2184 if (cached_object_alloc <= cached_object_nr) {
2185 cached_object_alloc = alloc_nr(cached_object_alloc);
2186 cached_objects = xrealloc(cached_objects,
2187 sizeof(*cached_objects) *
2188 cached_object_alloc);
2190 co = &cached_objects[cached_object_nr++];
2191 co->size = len;
2192 co->type = type;
2193 co->buf = xmalloc(len);
2194 memcpy(co->buf, buf, len);
2195 hashcpy(co->sha1, sha1);
2196 return 0;
2199 static void *read_object(const unsigned char *sha1, enum object_type *type,
2200 unsigned long *size)
2202 unsigned long mapsize;
2203 void *map, *buf;
2204 struct cached_object *co;
2206 co = find_cached_object(sha1);
2207 if (co) {
2208 *type = co->type;
2209 *size = co->size;
2210 return xmemdupz(co->buf, co->size);
2213 buf = read_packed_sha1(sha1, type, size);
2214 if (buf)
2215 return buf;
2216 map = map_sha1_file(sha1, &mapsize);
2217 if (map) {
2218 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2219 munmap(map, mapsize);
2220 return buf;
2222 reprepare_packed_git();
2223 return read_packed_sha1(sha1, type, size);
2227 * This function dies on corrupt objects; the callers who want to
2228 * deal with them should arrange to call read_object() and give error
2229 * messages themselves.
2231 void *read_sha1_file_extended(const unsigned char *sha1,
2232 enum object_type *type,
2233 unsigned long *size,
2234 unsigned flag)
2236 void *data;
2237 char *path;
2238 const struct packed_git *p;
2239 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2240 ? lookup_replace_object(sha1) : sha1;
2242 errno = 0;
2243 data = read_object(repl, type, size);
2244 if (data)
2245 return data;
2247 if (errno && errno != ENOENT)
2248 die_errno("failed to read object %s", sha1_to_hex(sha1));
2250 /* die if we replaced an object with one that does not exist */
2251 if (repl != sha1)
2252 die("replacement %s not found for %s",
2253 sha1_to_hex(repl), sha1_to_hex(sha1));
2255 if (has_loose_object(repl)) {
2256 path = sha1_file_name(sha1);
2257 die("loose object %s (stored in %s) is corrupt",
2258 sha1_to_hex(repl), path);
2261 if ((p = has_packed_and_bad(repl)) != NULL)
2262 die("packed object %s (stored in %s) is corrupt",
2263 sha1_to_hex(repl), p->pack_name);
2265 return NULL;
2268 void *read_object_with_reference(const unsigned char *sha1,
2269 const char *required_type_name,
2270 unsigned long *size,
2271 unsigned char *actual_sha1_return)
2273 enum object_type type, required_type;
2274 void *buffer;
2275 unsigned long isize;
2276 unsigned char actual_sha1[20];
2278 required_type = type_from_string(required_type_name);
2279 hashcpy(actual_sha1, sha1);
2280 while (1) {
2281 int ref_length = -1;
2282 const char *ref_type = NULL;
2284 buffer = read_sha1_file(actual_sha1, &type, &isize);
2285 if (!buffer)
2286 return NULL;
2287 if (type == required_type) {
2288 *size = isize;
2289 if (actual_sha1_return)
2290 hashcpy(actual_sha1_return, actual_sha1);
2291 return buffer;
2293 /* Handle references */
2294 else if (type == OBJ_COMMIT)
2295 ref_type = "tree ";
2296 else if (type == OBJ_TAG)
2297 ref_type = "object ";
2298 else {
2299 free(buffer);
2300 return NULL;
2302 ref_length = strlen(ref_type);
2304 if (ref_length + 40 > isize ||
2305 memcmp(buffer, ref_type, ref_length) ||
2306 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2307 free(buffer);
2308 return NULL;
2310 free(buffer);
2311 /* Now we have the ID of the referred-to object in
2312 * actual_sha1. Check again. */
2316 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2317 const char *type, unsigned char *sha1,
2318 char *hdr, int *hdrlen)
2320 git_SHA_CTX c;
2322 /* Generate the header */
2323 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2325 /* Sha1.. */
2326 git_SHA1_Init(&c);
2327 git_SHA1_Update(&c, hdr, *hdrlen);
2328 git_SHA1_Update(&c, buf, len);
2329 git_SHA1_Final(sha1, &c);
2333 * Move the just written object into its final resting place.
2334 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2335 * "moving" is only a part of what it does, when no patch between
2336 * master to pu changes the call sites of this function.
2338 int move_temp_to_file(const char *tmpfile, const char *filename)
2340 int ret = 0;
2342 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2343 goto try_rename;
2344 else if (link(tmpfile, filename))
2345 ret = errno;
2348 * Coda hack - coda doesn't like cross-directory links,
2349 * so we fall back to a rename, which will mean that it
2350 * won't be able to check collisions, but that's not a
2351 * big deal.
2353 * The same holds for FAT formatted media.
2355 * When this succeeds, we just return. We have nothing
2356 * left to unlink.
2358 if (ret && ret != EEXIST) {
2359 try_rename:
2360 if (!rename(tmpfile, filename))
2361 goto out;
2362 ret = errno;
2364 unlink_or_warn(tmpfile);
2365 if (ret) {
2366 if (ret != EEXIST) {
2367 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2369 /* FIXME!!! Collision check here ? */
2372 out:
2373 if (adjust_shared_perm(filename))
2374 return error("unable to set permission to '%s'", filename);
2375 return 0;
2378 static int write_buffer(int fd, const void *buf, size_t len)
2380 if (write_in_full(fd, buf, len) < 0)
2381 return error("file write error (%s)", strerror(errno));
2382 return 0;
2385 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2386 unsigned char *sha1)
2388 char hdr[32];
2389 int hdrlen;
2390 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2391 return 0;
2394 /* Finalize a file on disk, and close it. */
2395 static void close_sha1_file(int fd)
2397 if (fsync_object_files)
2398 fsync_or_die(fd, "sha1 file");
2399 if (close(fd) != 0)
2400 die_errno("error when closing sha1 file");
2403 /* Size of directory component, including the ending '/' */
2404 static inline int directory_size(const char *filename)
2406 const char *s = strrchr(filename, '/');
2407 if (!s)
2408 return 0;
2409 return s - filename + 1;
2413 * This creates a temporary file in the same directory as the final
2414 * 'filename'
2416 * We want to avoid cross-directory filename renames, because those
2417 * can have problems on various filesystems (FAT, NFS, Coda).
2419 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2421 int fd, dirlen = directory_size(filename);
2423 if (dirlen + 20 > bufsiz) {
2424 errno = ENAMETOOLONG;
2425 return -1;
2427 memcpy(buffer, filename, dirlen);
2428 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2429 fd = git_mkstemp_mode(buffer, 0444);
2430 if (fd < 0 && dirlen && errno == ENOENT) {
2431 /* Make sure the directory exists */
2432 memcpy(buffer, filename, dirlen);
2433 buffer[dirlen-1] = 0;
2434 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2435 return -1;
2437 /* Try again */
2438 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2439 fd = git_mkstemp_mode(buffer, 0444);
2441 return fd;
2444 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2445 const void *buf, unsigned long len, time_t mtime)
2447 int fd, ret;
2448 unsigned char compressed[4096];
2449 git_zstream stream;
2450 git_SHA_CTX c;
2451 unsigned char parano_sha1[20];
2452 char *filename;
2453 static char tmpfile[PATH_MAX];
2455 filename = sha1_file_name(sha1);
2456 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2457 if (fd < 0) {
2458 if (errno == EACCES)
2459 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2460 else
2461 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2464 /* Set it up */
2465 memset(&stream, 0, sizeof(stream));
2466 git_deflate_init(&stream, zlib_compression_level);
2467 stream.next_out = compressed;
2468 stream.avail_out = sizeof(compressed);
2469 git_SHA1_Init(&c);
2471 /* First header.. */
2472 stream.next_in = (unsigned char *)hdr;
2473 stream.avail_in = hdrlen;
2474 while (git_deflate(&stream, 0) == Z_OK)
2475 ; /* nothing */
2476 git_SHA1_Update(&c, hdr, hdrlen);
2478 /* Then the data itself.. */
2479 stream.next_in = (void *)buf;
2480 stream.avail_in = len;
2481 do {
2482 unsigned char *in0 = stream.next_in;
2483 ret = git_deflate(&stream, Z_FINISH);
2484 git_SHA1_Update(&c, in0, stream.next_in - in0);
2485 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2486 die("unable to write sha1 file");
2487 stream.next_out = compressed;
2488 stream.avail_out = sizeof(compressed);
2489 } while (ret == Z_OK);
2491 if (ret != Z_STREAM_END)
2492 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2493 ret = git_deflate_end_gently(&stream);
2494 if (ret != Z_OK)
2495 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2496 git_SHA1_Final(parano_sha1, &c);
2497 if (hashcmp(sha1, parano_sha1) != 0)
2498 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2500 close_sha1_file(fd);
2502 if (mtime) {
2503 struct utimbuf utb;
2504 utb.actime = mtime;
2505 utb.modtime = mtime;
2506 if (utime(tmpfile, &utb) < 0)
2507 warning("failed utime() on %s: %s",
2508 tmpfile, strerror(errno));
2511 return move_temp_to_file(tmpfile, filename);
2514 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2516 unsigned char sha1[20];
2517 char hdr[32];
2518 int hdrlen;
2520 /* Normally if we have it in the pack then we do not bother writing
2521 * it out into .git/objects/??/?{38} file.
2523 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2524 if (returnsha1)
2525 hashcpy(returnsha1, sha1);
2526 if (has_sha1_file(sha1))
2527 return 0;
2528 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2531 int force_object_loose(const unsigned char *sha1, time_t mtime)
2533 void *buf;
2534 unsigned long len;
2535 enum object_type type;
2536 char hdr[32];
2537 int hdrlen;
2538 int ret;
2540 if (has_loose_object(sha1))
2541 return 0;
2542 buf = read_packed_sha1(sha1, &type, &len);
2543 if (!buf)
2544 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2545 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2546 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2547 free(buf);
2549 return ret;
2552 int has_pack_index(const unsigned char *sha1)
2554 struct stat st;
2555 if (stat(sha1_pack_index_name(sha1), &st))
2556 return 0;
2557 return 1;
2560 int has_sha1_pack(const unsigned char *sha1)
2562 struct pack_entry e;
2563 return find_pack_entry(sha1, &e);
2566 int has_sha1_file(const unsigned char *sha1)
2568 struct pack_entry e;
2570 if (find_pack_entry(sha1, &e))
2571 return 1;
2572 return has_loose_object(sha1);
2575 static void check_tree(const void *buf, size_t size)
2577 struct tree_desc desc;
2578 struct name_entry entry;
2580 init_tree_desc(&desc, buf, size);
2581 while (tree_entry(&desc, &entry))
2582 /* do nothing
2583 * tree_entry() will die() on malformed entries */
2587 static void check_commit(const void *buf, size_t size)
2589 struct commit c;
2590 memset(&c, 0, sizeof(c));
2591 if (parse_commit_buffer(&c, buf, size))
2592 die("corrupt commit");
2595 static void check_tag(const void *buf, size_t size)
2597 struct tag t;
2598 memset(&t, 0, sizeof(t));
2599 if (parse_tag_buffer(&t, buf, size))
2600 die("corrupt tag");
2603 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2604 enum object_type type,
2605 const char *path, unsigned flags)
2607 int ret, re_allocated = 0;
2608 int write_object = flags & HASH_WRITE_OBJECT;
2610 if (!type)
2611 type = OBJ_BLOB;
2614 * Convert blobs to git internal format
2616 if ((type == OBJ_BLOB) && path) {
2617 struct strbuf nbuf = STRBUF_INIT;
2618 if (convert_to_git(path, buf, size, &nbuf,
2619 write_object ? safe_crlf : 0)) {
2620 buf = strbuf_detach(&nbuf, &size);
2621 re_allocated = 1;
2624 if (flags & HASH_FORMAT_CHECK) {
2625 if (type == OBJ_TREE)
2626 check_tree(buf, size);
2627 if (type == OBJ_COMMIT)
2628 check_commit(buf, size);
2629 if (type == OBJ_TAG)
2630 check_tag(buf, size);
2633 if (write_object)
2634 ret = write_sha1_file(buf, size, typename(type), sha1);
2635 else
2636 ret = hash_sha1_file(buf, size, typename(type), sha1);
2637 if (re_allocated)
2638 free(buf);
2639 return ret;
2642 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2643 const char *path, unsigned flags)
2645 struct strbuf sbuf = STRBUF_INIT;
2646 int ret;
2648 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2649 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2650 else
2651 ret = -1;
2652 strbuf_release(&sbuf);
2653 return ret;
2656 #define SMALL_FILE_SIZE (32*1024)
2658 static int index_core(unsigned char *sha1, int fd, size_t size,
2659 enum object_type type, const char *path,
2660 unsigned flags)
2662 int ret;
2664 if (!size) {
2665 ret = index_mem(sha1, NULL, size, type, path, flags);
2666 } else if (size <= SMALL_FILE_SIZE) {
2667 char *buf = xmalloc(size);
2668 if (size == read_in_full(fd, buf, size))
2669 ret = index_mem(sha1, buf, size, type, path, flags);
2670 else
2671 ret = error("short read %s", strerror(errno));
2672 free(buf);
2673 } else {
2674 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2675 ret = index_mem(sha1, buf, size, type, path, flags);
2676 munmap(buf, size);
2678 return ret;
2682 * This creates one packfile per large blob, because the caller
2683 * immediately wants the result sha1, and fast-import can report the
2684 * object name via marks mechanism only by closing the created
2685 * packfile.
2687 * This also bypasses the usual "convert-to-git" dance, and that is on
2688 * purpose. We could write a streaming version of the converting
2689 * functions and insert that before feeding the data to fast-import
2690 * (or equivalent in-core API described above), but the primary
2691 * motivation for trying to stream from the working tree file and to
2692 * avoid mmaping it in core is to deal with large binary blobs, and
2693 * by definition they do _not_ want to get any conversion.
2695 static int index_stream(unsigned char *sha1, int fd, size_t size,
2696 enum object_type type, const char *path,
2697 unsigned flags)
2699 struct child_process fast_import;
2700 char export_marks[512];
2701 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2702 char tmpfile[512];
2703 char fast_import_cmd[512];
2704 char buf[512];
2705 int len, tmpfd;
2707 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2708 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2709 if (tmpfd < 0)
2710 die_errno("cannot create tempfile: %s", tmpfile);
2711 if (close(tmpfd))
2712 die_errno("cannot close tempfile: %s", tmpfile);
2713 sprintf(export_marks, "--export-marks=%s", tmpfile);
2715 memset(&fast_import, 0, sizeof(fast_import));
2716 fast_import.in = -1;
2717 fast_import.argv = argv;
2718 fast_import.git_cmd = 1;
2719 if (start_command(&fast_import))
2720 die_errno("index-stream: git fast-import failed");
2722 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2723 (unsigned long) size);
2724 write_or_whine(fast_import.in, fast_import_cmd, len,
2725 "index-stream: feeding fast-import");
2726 while (size) {
2727 char buf[10240];
2728 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2729 ssize_t actual;
2731 actual = read_in_full(fd, buf, sz);
2732 if (actual < 0)
2733 die_errno("index-stream: reading input");
2734 if (write_in_full(fast_import.in, buf, actual) != actual)
2735 die_errno("index-stream: feeding fast-import");
2736 size -= actual;
2738 if (close(fast_import.in))
2739 die_errno("index-stream: closing fast-import");
2740 if (finish_command(&fast_import))
2741 die_errno("index-stream: finishing fast-import");
2743 tmpfd = open(tmpfile, O_RDONLY);
2744 if (tmpfd < 0)
2745 die_errno("index-stream: cannot open fast-import mark");
2746 len = read(tmpfd, buf, sizeof(buf));
2747 if (len < 0)
2748 die_errno("index-stream: reading fast-import mark");
2749 if (close(tmpfd) < 0)
2750 die_errno("index-stream: closing fast-import mark");
2751 if (unlink(tmpfile))
2752 die_errno("index-stream: unlinking fast-import mark");
2753 if (len != 44 ||
2754 memcmp(":1 ", buf, 3) ||
2755 get_sha1_hex(buf + 3, sha1))
2756 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2757 return 0;
2760 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2761 enum object_type type, const char *path, unsigned flags)
2763 int ret;
2764 size_t size = xsize_t(st->st_size);
2766 if (!S_ISREG(st->st_mode))
2767 ret = index_pipe(sha1, fd, type, path, flags);
2768 else if (size <= big_file_threshold || type != OBJ_BLOB)
2769 ret = index_core(sha1, fd, size, type, path, flags);
2770 else
2771 ret = index_stream(sha1, fd, size, type, path, flags);
2772 close(fd);
2773 return ret;
2776 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2778 int fd;
2779 struct strbuf sb = STRBUF_INIT;
2781 switch (st->st_mode & S_IFMT) {
2782 case S_IFREG:
2783 fd = open(path, O_RDONLY);
2784 if (fd < 0)
2785 return error("open(\"%s\"): %s", path,
2786 strerror(errno));
2787 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2788 return error("%s: failed to insert into database",
2789 path);
2790 break;
2791 case S_IFLNK:
2792 if (strbuf_readlink(&sb, path, st->st_size)) {
2793 char *errstr = strerror(errno);
2794 return error("readlink(\"%s\"): %s", path,
2795 errstr);
2797 if (!(flags & HASH_WRITE_OBJECT))
2798 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2799 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2800 return error("%s: failed to insert into database",
2801 path);
2802 strbuf_release(&sb);
2803 break;
2804 case S_IFDIR:
2805 return resolve_gitlink_ref(path, "HEAD", sha1);
2806 default:
2807 return error("%s: unsupported file type", path);
2809 return 0;
2812 int read_pack_header(int fd, struct pack_header *header)
2814 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2815 /* "eof before pack header was fully read" */
2816 return PH_ERROR_EOF;
2818 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2819 /* "protocol error (pack signature mismatch detected)" */
2820 return PH_ERROR_PACK_SIGNATURE;
2821 if (!pack_version_ok(header->hdr_version))
2822 /* "protocol error (pack version unsupported)" */
2823 return PH_ERROR_PROTOCOL;
2824 return 0;
2827 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2829 enum object_type type = sha1_object_info(sha1, NULL);
2830 if (type < 0)
2831 die("%s is not a valid object", sha1_to_hex(sha1));
2832 if (type != expect)
2833 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2834 typename(expect));