am: Use cat instead of echo to avoid DOS line-endings (fixes t4150)
[git/dscho.git] / sha1_file.c
blob3c4f1652f13479238dee56674cbb86d299363be1
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"
21 #include "bulk-checkin.h"
22 #include "streaming.h"
24 #ifndef O_NOATIME
25 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
26 #define O_NOATIME 01000000
27 #else
28 #define O_NOATIME 0
29 #endif
30 #endif
32 #define SZ_FMT PRIuMAX
33 static inline uintmax_t sz_fmt(size_t s) { return s; }
35 const unsigned char null_sha1[20];
38 * This is meant to hold a *small* number of objects that you would
39 * want read_sha1_file() to be able to return, but yet you do not want
40 * to write them into the object store (e.g. a browse-only
41 * application).
43 static struct cached_object {
44 unsigned char sha1[20];
45 enum object_type type;
46 void *buf;
47 unsigned long size;
48 } *cached_objects;
49 static int cached_object_nr, cached_object_alloc;
51 static struct cached_object empty_tree = {
52 EMPTY_TREE_SHA1_BIN_LITERAL,
53 OBJ_TREE,
54 "",
58 static struct packed_git *last_found_pack;
60 static struct cached_object *find_cached_object(const unsigned char *sha1)
62 int i;
63 struct cached_object *co = cached_objects;
65 for (i = 0; i < cached_object_nr; i++, co++) {
66 if (!hashcmp(co->sha1, sha1))
67 return co;
69 if (!hashcmp(sha1, empty_tree.sha1))
70 return &empty_tree;
71 return NULL;
74 int mkdir_in_gitdir(const char *path)
76 if (mkdir(path, 0777)) {
77 int saved_errno = errno;
78 struct stat st;
79 struct strbuf sb = STRBUF_INIT;
81 if (errno != EEXIST)
82 return -1;
84 * Are we looking at a path in a symlinked worktree
85 * whose original repository does not yet have it?
86 * e.g. .git/rr-cache pointing at its original
87 * repository in which the user hasn't performed any
88 * conflict resolution yet?
90 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
91 strbuf_readlink(&sb, path, st.st_size) ||
92 !is_absolute_path(sb.buf) ||
93 mkdir(sb.buf, 0777)) {
94 strbuf_release(&sb);
95 errno = saved_errno;
96 return -1;
98 strbuf_release(&sb);
100 return adjust_shared_perm(path);
103 int safe_create_leading_directories(char *path)
105 char *pos = path + offset_1st_component(path);
106 struct stat st;
108 while (pos) {
109 pos = strchr(pos, '/');
110 if (!pos)
111 break;
112 while (*++pos == '/')
114 if (!*pos)
115 break;
116 *--pos = '\0';
117 if (!stat(path, &st)) {
118 /* path exists */
119 if (!S_ISDIR(st.st_mode)) {
120 *pos = '/';
121 return -3;
124 else if (mkdir(path, 0777)) {
125 *pos = '/';
126 return -1;
128 else if (adjust_shared_perm(path)) {
129 *pos = '/';
130 return -2;
132 *pos++ = '/';
134 return 0;
137 int safe_create_leading_directories_const(const char *path)
139 /* path points to cache entries, so xstrdup before messing with it */
140 char *buf = xstrdup(path);
141 int result = safe_create_leading_directories(buf);
142 free(buf);
143 return result;
146 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
148 int i;
149 for (i = 0; i < 20; i++) {
150 static char hex[] = "0123456789abcdef";
151 unsigned int val = sha1[i];
152 char *pos = pathbuf + i*2 + (i > 0);
153 *pos++ = hex[val >> 4];
154 *pos = hex[val & 0xf];
159 * NOTE! This returns a statically allocated buffer, so you have to be
160 * careful about using it. Do an "xstrdup()" if you need to save the
161 * filename.
163 * Also note that this returns the location for creating. Reading
164 * SHA1 file can happen from any alternate directory listed in the
165 * DB_ENVIRONMENT environment variable if it is not found in
166 * the primary object database.
168 char *sha1_file_name(const unsigned char *sha1)
170 static char buf[PATH_MAX];
171 const char *objdir;
172 int len;
174 objdir = get_object_directory();
175 len = strlen(objdir);
177 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
178 if (len + 43 > PATH_MAX)
179 die("insanely long object directory %s", objdir);
180 memcpy(buf, objdir, len);
181 buf[len] = '/';
182 buf[len+3] = '/';
183 buf[len+42] = '\0';
184 fill_sha1_path(buf + len + 1, sha1);
185 return buf;
188 static char *sha1_get_pack_name(const unsigned char *sha1,
189 char **name, char **base, const char *which)
191 static const char hex[] = "0123456789abcdef";
192 char *buf;
193 int i;
195 if (!*base) {
196 const char *sha1_file_directory = get_object_directory();
197 int len = strlen(sha1_file_directory);
198 *base = xmalloc(len + 60);
199 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
200 sha1_file_directory, which);
201 *name = *base + len + 11;
204 buf = *name;
206 for (i = 0; i < 20; i++) {
207 unsigned int val = *sha1++;
208 *buf++ = hex[val >> 4];
209 *buf++ = hex[val & 0xf];
212 return *base;
215 char *sha1_pack_name(const unsigned char *sha1)
217 static char *name, *base;
219 return sha1_get_pack_name(sha1, &name, &base, "pack");
222 char *sha1_pack_index_name(const unsigned char *sha1)
224 static char *name, *base;
226 return sha1_get_pack_name(sha1, &name, &base, "idx");
229 struct alternate_object_database *alt_odb_list;
230 static struct alternate_object_database **alt_odb_tail;
232 static void read_info_alternates(const char * alternates, int depth);
233 static int git_open_noatime(const char *name);
236 * Prepare alternate object database registry.
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
250 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
252 const char *objdir = get_object_directory();
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
258 if (!is_absolute_path(entry) && relative_base) {
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
262 strbuf_add(&pathbuf, entry, len);
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
266 pfxlen = strlen(pathbuf.buf);
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284 /* Detect cases where alternate disappeared */
285 if (!is_directory(ent->base)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (!memcmp(ent->base, alt->base, pfxlen)) {
298 free(ent);
299 return -1;
302 if (!memcmp(ent->base, objdir, pfxlen)) {
303 free(ent);
304 return -1;
307 /* add the alternate entry */
308 *alt_odb_tail = ent;
309 alt_odb_tail = &(ent->next);
310 ent->next = NULL;
312 /* recursively add alternates */
313 read_info_alternates(ent->base, depth + 1);
315 ent->base[pfxlen] = '/';
317 return 0;
320 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
321 const char *relative_base, int depth)
323 const char *cp, *last;
325 if (depth > 5) {
326 error("%s: ignoring alternate object stores, nesting too deep.",
327 relative_base);
328 return;
331 last = alt;
332 while (last < ep) {
333 cp = last;
334 if (cp < ep && *cp == '#') {
335 while (cp < ep && *cp != sep)
336 cp++;
337 last = cp + 1;
338 continue;
340 while (cp < ep && *cp != sep)
341 cp++;
342 if (last != cp) {
343 if (!is_absolute_path(last) && depth) {
344 error("%s: ignoring relative alternate object store %s",
345 relative_base, last);
346 } else {
347 link_alt_odb_entry(last, cp - last,
348 relative_base, depth);
351 while (cp < ep && *cp == sep)
352 cp++;
353 last = cp;
357 static void read_info_alternates(const char * relative_base, int depth)
359 char *map;
360 size_t mapsz;
361 struct stat st;
362 const char alt_file_name[] = "info/alternates";
363 /* Given that relative_base is no longer than PATH_MAX,
364 ensure that "path" has enough space to append "/", the
365 file name, "info/alternates", and a trailing NUL. */
366 char path[PATH_MAX + 1 + sizeof alt_file_name];
367 int fd;
369 sprintf(path, "%s/%s", relative_base, alt_file_name);
370 fd = git_open_noatime(path);
371 if (fd < 0)
372 return;
373 if (fstat(fd, &st) || (st.st_size == 0)) {
374 close(fd);
375 return;
377 mapsz = xsize_t(st.st_size);
378 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
379 close(fd);
381 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
383 munmap(map, mapsz);
386 void add_to_alternates_file(const char *reference)
388 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
389 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
390 char *alt = mkpath("%s\n", reference);
391 write_or_die(fd, alt, strlen(alt));
392 if (commit_lock_file(lock))
393 die("could not close alternates file");
394 if (alt_odb_tail)
395 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
398 void foreach_alt_odb(alt_odb_fn fn, void *cb)
400 struct alternate_object_database *ent;
402 prepare_alt_odb();
403 for (ent = alt_odb_list; ent; ent = ent->next)
404 if (fn(ent, cb))
405 return;
408 void prepare_alt_odb(void)
410 const char *alt;
412 if (alt_odb_tail)
413 return;
415 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
416 if (!alt) alt = "";
418 alt_odb_tail = &alt_odb_list;
419 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
421 read_info_alternates(get_object_directory(), 0);
424 static int has_loose_object_local(const unsigned char *sha1)
426 char *name = sha1_file_name(sha1);
427 return !access(name, F_OK);
430 int has_loose_object_nonlocal(const unsigned char *sha1)
432 struct alternate_object_database *alt;
433 prepare_alt_odb();
434 for (alt = alt_odb_list; alt; alt = alt->next) {
435 fill_sha1_path(alt->name, sha1);
436 if (!access(alt->base, F_OK))
437 return 1;
439 return 0;
442 static int has_loose_object(const unsigned char *sha1)
444 return has_loose_object_local(sha1) ||
445 has_loose_object_nonlocal(sha1);
448 static unsigned int pack_used_ctr;
449 static unsigned int pack_mmap_calls;
450 static unsigned int peak_pack_open_windows;
451 static unsigned int pack_open_windows;
452 static unsigned int pack_open_fds;
453 static unsigned int pack_max_fds;
454 static size_t peak_pack_mapped;
455 static size_t pack_mapped;
456 struct packed_git *packed_git;
458 void pack_report(void)
460 fprintf(stderr,
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
479 static int check_packed_git_idx(const char *path, struct packed_git *p)
481 void *idx_map;
482 struct pack_idx_header *hdr;
483 size_t idx_size;
484 uint32_t version, nr, i, *index;
485 int fd = git_open_noatime(path);
486 struct stat st;
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
494 idx_size = xsize_t(st.st_size);
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
500 close(fd);
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
512 } else
513 version = 1;
515 nr = 0;
516 index = idx_map;
517 if (version > 1)
518 index += 2; /* skip index header */
519 for (i = 0; i < 256; i++) {
520 uint32_t n = ntohl(index[i]);
521 if (n < nr) {
522 munmap(idx_map, idx_size);
523 return error("non-monotonic index %s", path);
525 nr = n;
528 if (version == 1) {
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
538 return error("wrong index v1 file size in %s", path);
540 } else if (version == 2) {
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
559 munmap(idx_map, idx_size);
560 return error("wrong index v2 file size in %s", path);
562 if (idx_size != min_size &&
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
574 p->index_version = version;
575 p->index_data = idx_map;
576 p->index_size = idx_size;
577 p->num_objects = nr;
578 return 0;
581 int open_pack_index(struct packed_git *p)
583 char *idx_name;
584 int ret;
586 if (p->index_data)
587 return 0;
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
596 static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
601 struct pack_window *w, *w_l;
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
611 w_l = w;
615 static int unuse_one_window(struct packed_git *current, int keep_fd)
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
629 else {
630 lru_p->windows = lru_w->next;
631 if (!lru_p->windows && lru_p->pack_fd != -1
632 && lru_p->pack_fd != keep_fd) {
633 close(lru_p->pack_fd);
634 pack_open_fds--;
635 lru_p->pack_fd = -1;
638 free(lru_w);
639 pack_open_windows--;
640 return 1;
642 return 0;
645 void release_pack_memory(size_t need, int fd)
647 size_t cur = pack_mapped;
648 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
649 ; /* nothing */
652 void *xmmap(void *start, size_t length,
653 int prot, int flags, int fd, off_t offset)
655 void *ret = mmap(start, length, prot, flags, fd, offset);
656 if (ret == MAP_FAILED) {
657 if (!length)
658 return NULL;
659 release_pack_memory(length, fd);
660 ret = mmap(start, length, prot, flags, fd, offset);
661 if (ret == MAP_FAILED)
662 die_errno("Out of memory? mmap failed");
664 return ret;
667 void close_pack_windows(struct packed_git *p)
669 while (p->windows) {
670 struct pack_window *w = p->windows;
672 if (w->inuse_cnt)
673 die("pack '%s' still has open windows to it",
674 p->pack_name);
675 munmap(w->base, w->len);
676 pack_mapped -= w->len;
677 pack_open_windows--;
678 p->windows = w->next;
679 free(w);
683 void unuse_pack(struct pack_window **w_cursor)
685 struct pack_window *w = *w_cursor;
686 if (w) {
687 w->inuse_cnt--;
688 *w_cursor = NULL;
692 void close_pack_index(struct packed_git *p)
694 if (p->index_data) {
695 munmap((void *)p->index_data, p->index_size);
696 p->index_data = NULL;
701 * This is used by git-repack in case a newly created pack happens to
702 * contain the same set of objects as an existing one. In that case
703 * the resulting file might be different even if its name would be the
704 * same. It is best to close any reference to the old pack before it is
705 * replaced on disk. Of course no index pointers nor windows for given pack
706 * must subsist at this point. If ever objects from this pack are requested
707 * again, the new version of the pack will be reinitialized through
708 * reprepare_packed_git().
710 void free_pack_by_name(const char *pack_name)
712 struct packed_git *p, **pp = &packed_git;
714 while (*pp) {
715 p = *pp;
716 if (strcmp(pack_name, p->pack_name) == 0) {
717 clear_delta_base_cache();
718 close_pack_windows(p);
719 if (p->pack_fd != -1) {
720 close(p->pack_fd);
721 pack_open_fds--;
723 close_pack_index(p);
724 free(p->bad_object_sha1);
725 *pp = p->next;
726 if (last_found_pack == p)
727 last_found_pack = NULL;
728 free(p);
729 return;
731 pp = &p->next;
736 * Do not call this directly as this leaks p->pack_fd on error return;
737 * call open_packed_git() instead.
739 static int open_packed_git_1(struct packed_git *p)
741 struct stat st;
742 struct pack_header hdr;
743 unsigned char sha1[20];
744 unsigned char *idx_sha1;
745 long fd_flag;
747 if (!p->index_data && open_pack_index(p))
748 return error("packfile %s index unavailable", p->pack_name);
750 if (!pack_max_fds) {
751 struct rlimit lim;
752 unsigned int max_fds;
754 if (getrlimit(RLIMIT_NOFILE, &lim))
755 die_errno("cannot get RLIMIT_NOFILE");
757 max_fds = lim.rlim_cur;
759 /* Save 3 for stdin/stdout/stderr, 22 for work */
760 if (25 < max_fds)
761 pack_max_fds = max_fds - 25;
762 else
763 pack_max_fds = 1;
766 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
767 ; /* nothing */
769 p->pack_fd = git_open_noatime(p->pack_name);
770 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
771 return -1;
772 pack_open_fds++;
774 /* If we created the struct before we had the pack we lack size. */
775 if (!p->pack_size) {
776 if (!S_ISREG(st.st_mode))
777 return error("packfile %s not a regular file", p->pack_name);
778 p->pack_size = st.st_size;
779 } else if (p->pack_size != st.st_size)
780 return error("packfile %s size changed", p->pack_name);
782 /* We leave these file descriptors open with sliding mmap;
783 * there is no point keeping them open across exec(), though.
785 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
786 if (fd_flag < 0)
787 return error("cannot determine file descriptor flags");
788 fd_flag |= FD_CLOEXEC;
789 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
790 return error("cannot set FD_CLOEXEC");
792 /* Verify we recognize this pack file format. */
793 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
794 return error("file %s is far too short to be a packfile", p->pack_name);
795 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
796 return error("file %s is not a GIT packfile", p->pack_name);
797 if (!pack_version_ok(hdr.hdr_version))
798 return error("packfile %s is version %"PRIu32" and not"
799 " supported (try upgrading GIT to a newer version)",
800 p->pack_name, ntohl(hdr.hdr_version));
802 /* Verify the pack matches its index. */
803 if (p->num_objects != ntohl(hdr.hdr_entries))
804 return error("packfile %s claims to have %"PRIu32" objects"
805 " while index indicates %"PRIu32" objects",
806 p->pack_name, ntohl(hdr.hdr_entries),
807 p->num_objects);
808 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
809 return error("end of packfile %s is unavailable", p->pack_name);
810 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
811 return error("packfile %s signature is unavailable", p->pack_name);
812 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
813 if (hashcmp(sha1, idx_sha1))
814 return error("packfile %s does not match index", p->pack_name);
815 return 0;
818 static int open_packed_git(struct packed_git *p)
820 if (!open_packed_git_1(p))
821 return 0;
822 if (p->pack_fd != -1) {
823 close(p->pack_fd);
824 pack_open_fds--;
825 p->pack_fd = -1;
827 return -1;
830 static int in_window(struct pack_window *win, off_t offset)
832 /* We must promise at least 20 bytes (one hash) after the
833 * offset is available from this window, otherwise the offset
834 * is not actually in this window and a different window (which
835 * has that one hash excess) must be used. This is to support
836 * the object header and delta base parsing routines below.
838 off_t win_off = win->offset;
839 return win_off <= offset
840 && (offset + 20) <= (win_off + win->len);
843 unsigned char *use_pack(struct packed_git *p,
844 struct pack_window **w_cursor,
845 off_t offset,
846 unsigned long *left)
848 struct pack_window *win = *w_cursor;
850 /* Since packfiles end in a hash of their content and it's
851 * pointless to ask for an offset into the middle of that
852 * hash, and the in_window function above wouldn't match
853 * don't allow an offset too close to the end of the file.
855 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
856 die("packfile %s cannot be accessed", p->pack_name);
857 if (offset > (p->pack_size - 20))
858 die("offset beyond end of packfile (truncated pack?)");
860 if (!win || !in_window(win, offset)) {
861 if (win)
862 win->inuse_cnt--;
863 for (win = p->windows; win; win = win->next) {
864 if (in_window(win, offset))
865 break;
867 if (!win) {
868 size_t window_align = packed_git_window_size / 2;
869 off_t len;
871 if (p->pack_fd == -1 && open_packed_git(p))
872 die("packfile %s cannot be accessed", p->pack_name);
874 win = xcalloc(1, sizeof(*win));
875 win->offset = (offset / window_align) * window_align;
876 len = p->pack_size - win->offset;
877 if (len > packed_git_window_size)
878 len = packed_git_window_size;
879 win->len = (size_t)len;
880 pack_mapped += win->len;
881 while (packed_git_limit < pack_mapped
882 && unuse_one_window(p, p->pack_fd))
883 ; /* nothing */
884 win->base = xmmap(NULL, win->len,
885 PROT_READ, MAP_PRIVATE,
886 p->pack_fd, win->offset);
887 if (win->base == MAP_FAILED)
888 die("packfile %s cannot be mapped: %s",
889 p->pack_name,
890 strerror(errno));
891 if (!win->offset && win->len == p->pack_size
892 && !p->do_not_close) {
893 close(p->pack_fd);
894 pack_open_fds--;
895 p->pack_fd = -1;
897 pack_mmap_calls++;
898 pack_open_windows++;
899 if (pack_mapped > peak_pack_mapped)
900 peak_pack_mapped = pack_mapped;
901 if (pack_open_windows > peak_pack_open_windows)
902 peak_pack_open_windows = pack_open_windows;
903 win->next = p->windows;
904 p->windows = win;
907 if (win != *w_cursor) {
908 win->last_used = pack_used_ctr++;
909 win->inuse_cnt++;
910 *w_cursor = win;
912 offset -= win->offset;
913 if (left)
914 *left = win->len - xsize_t(offset);
915 return win->base + offset;
918 static struct packed_git *alloc_packed_git(int extra)
920 struct packed_git *p = xmalloc(sizeof(*p) + extra);
921 memset(p, 0, sizeof(*p));
922 p->pack_fd = -1;
923 return p;
926 static void try_to_free_pack_memory(size_t size)
928 release_pack_memory(size, -1);
931 struct packed_git *add_packed_git(const char *path, int path_len, int local)
933 static int have_set_try_to_free_routine;
934 struct stat st;
935 struct packed_git *p = alloc_packed_git(path_len + 2);
937 if (!have_set_try_to_free_routine) {
938 have_set_try_to_free_routine = 1;
939 set_try_to_free_routine(try_to_free_pack_memory);
943 * Make sure a corresponding .pack file exists and that
944 * the index looks sane.
946 path_len -= strlen(".idx");
947 if (path_len < 1) {
948 free(p);
949 return NULL;
951 memcpy(p->pack_name, path, path_len);
953 strcpy(p->pack_name + path_len, ".keep");
954 if (!access(p->pack_name, F_OK))
955 p->pack_keep = 1;
957 strcpy(p->pack_name + path_len, ".pack");
958 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
959 free(p);
960 return NULL;
963 /* ok, it looks sane as far as we can check without
964 * actually mapping the pack file.
966 p->pack_size = st.st_size;
967 p->pack_local = local;
968 p->mtime = st.st_mtime;
969 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
970 hashclr(p->sha1);
971 return p;
974 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
976 const char *path = sha1_pack_name(sha1);
977 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
979 strcpy(p->pack_name, path);
980 hashcpy(p->sha1, sha1);
981 if (check_packed_git_idx(idx_path, p)) {
982 free(p);
983 return NULL;
986 return p;
989 void install_packed_git(struct packed_git *pack)
991 if (pack->pack_fd != -1)
992 pack_open_fds++;
994 pack->next = packed_git;
995 packed_git = pack;
998 static void prepare_packed_git_one(char *objdir, int local)
1000 /* Ensure that this buffer is large enough so that we can
1001 append "/pack/" without clobbering the stack even if
1002 strlen(objdir) were PATH_MAX. */
1003 char path[PATH_MAX + 1 + 4 + 1 + 1];
1004 int len;
1005 DIR *dir;
1006 struct dirent *de;
1008 sprintf(path, "%s/pack", objdir);
1009 len = strlen(path);
1010 dir = opendir(path);
1011 if (!dir) {
1012 if (errno != ENOENT)
1013 error("unable to open object pack directory: %s: %s",
1014 path, strerror(errno));
1015 return;
1017 path[len++] = '/';
1018 while ((de = readdir(dir)) != NULL) {
1019 int namelen = strlen(de->d_name);
1020 struct packed_git *p;
1022 if (!has_extension(de->d_name, ".idx"))
1023 continue;
1025 if (len + namelen + 1 > sizeof(path))
1026 continue;
1028 /* Don't reopen a pack we already have. */
1029 strcpy(path + len, de->d_name);
1030 for (p = packed_git; p; p = p->next) {
1031 if (!memcmp(path, p->pack_name, len + namelen - 4))
1032 break;
1034 if (p)
1035 continue;
1036 /* See if it really is a valid .idx file with corresponding
1037 * .pack file that we can map.
1039 p = add_packed_git(path, len + namelen, local);
1040 if (!p)
1041 continue;
1042 install_packed_git(p);
1044 closedir(dir);
1047 static int sort_pack(const void *a_, const void *b_)
1049 struct packed_git *a = *((struct packed_git **)a_);
1050 struct packed_git *b = *((struct packed_git **)b_);
1051 int st;
1054 * Local packs tend to contain objects specific to our
1055 * variant of the project than remote ones. In addition,
1056 * remote ones could be on a network mounted filesystem.
1057 * Favor local ones for these reasons.
1059 st = a->pack_local - b->pack_local;
1060 if (st)
1061 return -st;
1064 * Younger packs tend to contain more recent objects,
1065 * and more recent objects tend to get accessed more
1066 * often.
1068 if (a->mtime < b->mtime)
1069 return 1;
1070 else if (a->mtime == b->mtime)
1071 return 0;
1072 return -1;
1075 static void rearrange_packed_git(void)
1077 struct packed_git **ary, *p;
1078 int i, n;
1080 for (n = 0, p = packed_git; p; p = p->next)
1081 n++;
1082 if (n < 2)
1083 return;
1085 /* prepare an array of packed_git for easier sorting */
1086 ary = xcalloc(n, sizeof(struct packed_git *));
1087 for (n = 0, p = packed_git; p; p = p->next)
1088 ary[n++] = p;
1090 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1092 /* link them back again */
1093 for (i = 0; i < n - 1; i++)
1094 ary[i]->next = ary[i + 1];
1095 ary[n - 1]->next = NULL;
1096 packed_git = ary[0];
1098 free(ary);
1101 static int prepare_packed_git_run_once = 0;
1102 void prepare_packed_git(void)
1104 struct alternate_object_database *alt;
1106 if (prepare_packed_git_run_once)
1107 return;
1108 prepare_packed_git_one(get_object_directory(), 1);
1109 prepare_alt_odb();
1110 for (alt = alt_odb_list; alt; alt = alt->next) {
1111 alt->name[-1] = 0;
1112 prepare_packed_git_one(alt->base, 0);
1113 alt->name[-1] = '/';
1115 rearrange_packed_git();
1116 prepare_packed_git_run_once = 1;
1119 void reprepare_packed_git(void)
1121 discard_revindex();
1122 prepare_packed_git_run_once = 0;
1123 prepare_packed_git();
1126 static void mark_bad_packed_object(struct packed_git *p,
1127 const unsigned char *sha1)
1129 unsigned i;
1130 for (i = 0; i < p->num_bad_objects; i++)
1131 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1132 return;
1133 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1134 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1135 p->num_bad_objects++;
1138 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1140 struct packed_git *p;
1141 unsigned i;
1143 for (p = packed_git; p; p = p->next)
1144 for (i = 0; i < p->num_bad_objects; i++)
1145 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1146 return p;
1147 return NULL;
1151 * With an in-core object data in "map", rehash it to make sure the
1152 * object name actually matches "sha1" to detect object corruption.
1153 * With "map" == NULL, try reading the object named with "sha1" using
1154 * the streaming interface and rehash it to do the same.
1156 int check_sha1_signature(const unsigned char *sha1, void *map,
1157 unsigned long size, const char *type)
1159 unsigned char real_sha1[20];
1160 enum object_type obj_type;
1161 struct git_istream *st;
1162 git_SHA_CTX c;
1163 char hdr[32];
1164 int hdrlen;
1166 if (map) {
1167 hash_sha1_file(map, size, type, real_sha1);
1168 return hashcmp(sha1, real_sha1) ? -1 : 0;
1171 st = open_istream(sha1, &obj_type, &size, NULL);
1172 if (!st)
1173 return -1;
1175 /* Generate the header */
1176 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1178 /* Sha1.. */
1179 git_SHA1_Init(&c);
1180 git_SHA1_Update(&c, hdr, hdrlen);
1181 for (;;) {
1182 char buf[1024 * 16];
1183 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1185 if (!readlen)
1186 break;
1187 git_SHA1_Update(&c, buf, readlen);
1189 git_SHA1_Final(real_sha1, &c);
1190 close_istream(st);
1191 return hashcmp(sha1, real_sha1) ? -1 : 0;
1194 static int git_open_noatime(const char *name)
1196 static int sha1_file_open_flag = O_NOATIME;
1198 for (;;) {
1199 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1200 if (fd >= 0)
1201 return fd;
1203 /* Might the failure be due to O_NOATIME? */
1204 if (errno != ENOENT && sha1_file_open_flag) {
1205 sha1_file_open_flag = 0;
1206 continue;
1209 return -1;
1213 static int open_sha1_file(const unsigned char *sha1)
1215 int fd;
1216 char *name = sha1_file_name(sha1);
1217 struct alternate_object_database *alt;
1219 fd = git_open_noatime(name);
1220 if (fd >= 0)
1221 return fd;
1223 prepare_alt_odb();
1224 errno = ENOENT;
1225 for (alt = alt_odb_list; alt; alt = alt->next) {
1226 name = alt->name;
1227 fill_sha1_path(name, sha1);
1228 fd = git_open_noatime(alt->base);
1229 if (fd >= 0)
1230 return fd;
1232 return -1;
1235 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1237 void *map;
1238 int fd;
1240 fd = open_sha1_file(sha1);
1241 map = NULL;
1242 if (fd >= 0) {
1243 struct stat st;
1245 if (!fstat(fd, &st)) {
1246 *size = xsize_t(st.st_size);
1247 if (!*size) {
1248 /* mmap() is forbidden on empty files */
1249 error("object file %s is empty", sha1_file_name(sha1));
1250 return NULL;
1252 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1254 close(fd);
1256 return map;
1260 * There used to be a second loose object header format which
1261 * was meant to mimic the in-pack format, allowing for direct
1262 * copy of the object data. This format turned up not to be
1263 * really worth it and we no longer write loose objects in that
1264 * format.
1266 static int experimental_loose_object(unsigned char *map)
1268 unsigned int word;
1271 * We must determine if the buffer contains the standard
1272 * zlib-deflated stream or the experimental format based
1273 * on the in-pack object format. Compare the header byte
1274 * for each format:
1276 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1277 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1279 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1280 * in standard loose-object format, UNLESS it is a Git-pack
1281 * format object *exactly* 8 bytes in size when inflated.
1283 * However, RFC1950 also specifies that the 1st 16-bit word
1284 * must be divisible by 31 - this checksum tells us our buffer
1285 * is in the standard format, giving a false positive only if
1286 * the 1st word of the Git-pack format object happens to be
1287 * divisible by 31, ie:
1288 * ((byte0 * 256) + byte1) % 31 = 0
1289 * => 0ttt10000www1000 % 31 = 0
1291 * As it happens, this case can only arise for www=3 & ttt=1
1292 * - ie, a Commit object, which would have to be 8 bytes in
1293 * size. As no Commit can be that small, we find that the
1294 * combination of these two criteria (bitmask & checksum)
1295 * can always correctly determine the buffer format.
1297 word = (map[0] << 8) + map[1];
1298 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1299 return 0;
1300 else
1301 return 1;
1304 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1305 unsigned long len, enum object_type *type, unsigned long *sizep)
1307 unsigned shift;
1308 unsigned long size, c;
1309 unsigned long used = 0;
1311 c = buf[used++];
1312 *type = (c >> 4) & 7;
1313 size = c & 15;
1314 shift = 4;
1315 while (c & 0x80) {
1316 if (len <= used || bitsizeof(long) <= shift) {
1317 error("bad object header");
1318 size = used = 0;
1319 break;
1321 c = buf[used++];
1322 size += (c & 0x7f) << shift;
1323 shift += 7;
1325 *sizep = size;
1326 return used;
1329 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1331 unsigned long size, used;
1332 static const char valid_loose_object_type[8] = {
1333 0, /* OBJ_EXT */
1334 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1335 0, /* "delta" and others are invalid in a loose object */
1337 enum object_type type;
1339 /* Get the data stream */
1340 memset(stream, 0, sizeof(*stream));
1341 stream->next_in = map;
1342 stream->avail_in = mapsize;
1343 stream->next_out = buffer;
1344 stream->avail_out = bufsiz;
1346 if (experimental_loose_object(map)) {
1348 * The old experimental format we no longer produce;
1349 * we can still read it.
1351 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1352 if (!used || !valid_loose_object_type[type])
1353 return -1;
1354 map += used;
1355 mapsize -= used;
1357 /* Set up the stream for the rest.. */
1358 stream->next_in = map;
1359 stream->avail_in = mapsize;
1360 git_inflate_init(stream);
1362 /* And generate the fake traditional header */
1363 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1364 typename(type), size);
1365 return 0;
1367 git_inflate_init(stream);
1368 return git_inflate(stream, 0);
1371 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1373 int bytes = strlen(buffer) + 1;
1374 unsigned char *buf = xmallocz(size);
1375 unsigned long n;
1376 int status = Z_OK;
1378 n = stream->total_out - bytes;
1379 if (n > size)
1380 n = size;
1381 memcpy(buf, (char *) buffer + bytes, n);
1382 bytes = n;
1383 if (bytes <= size) {
1385 * The above condition must be (bytes <= size), not
1386 * (bytes < size). In other words, even though we
1387 * expect no more output and set avail_out to zero,
1388 * the input zlib stream may have bytes that express
1389 * "this concludes the stream", and we *do* want to
1390 * eat that input.
1392 * Otherwise we would not be able to test that we
1393 * consumed all the input to reach the expected size;
1394 * we also want to check that zlib tells us that all
1395 * went well with status == Z_STREAM_END at the end.
1397 stream->next_out = buf + bytes;
1398 stream->avail_out = size - bytes;
1399 while (status == Z_OK)
1400 status = git_inflate(stream, Z_FINISH);
1402 if (status == Z_STREAM_END && !stream->avail_in) {
1403 git_inflate_end(stream);
1404 return buf;
1407 if (status < 0)
1408 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1409 else if (stream->avail_in)
1410 error("garbage at end of loose object '%s'",
1411 sha1_to_hex(sha1));
1412 free(buf);
1413 return NULL;
1417 * We used to just use "sscanf()", but that's actually way
1418 * too permissive for what we want to check. So do an anal
1419 * object header parse by hand.
1421 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1423 char type[10];
1424 int i;
1425 unsigned long size;
1428 * The type can be at most ten bytes (including the
1429 * terminating '\0' that we add), and is followed by
1430 * a space.
1432 i = 0;
1433 for (;;) {
1434 char c = *hdr++;
1435 if (c == ' ')
1436 break;
1437 type[i++] = c;
1438 if (i >= sizeof(type))
1439 return -1;
1441 type[i] = 0;
1444 * The length must follow immediately, and be in canonical
1445 * decimal format (ie "010" is not valid).
1447 size = *hdr++ - '0';
1448 if (size > 9)
1449 return -1;
1450 if (size) {
1451 for (;;) {
1452 unsigned long c = *hdr - '0';
1453 if (c > 9)
1454 break;
1455 hdr++;
1456 size = size * 10 + c;
1459 *sizep = size;
1462 * The length must be followed by a zero byte
1464 return *hdr ? -1 : type_from_string(type);
1467 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1469 int ret;
1470 git_zstream stream;
1471 char hdr[8192];
1473 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1474 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1475 return NULL;
1477 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1480 unsigned long get_size_from_delta(struct packed_git *p,
1481 struct pack_window **w_curs,
1482 off_t curpos)
1484 const unsigned char *data;
1485 unsigned char delta_head[20], *in;
1486 git_zstream stream;
1487 int st;
1489 memset(&stream, 0, sizeof(stream));
1490 stream.next_out = delta_head;
1491 stream.avail_out = sizeof(delta_head);
1493 git_inflate_init(&stream);
1494 do {
1495 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1496 stream.next_in = in;
1497 st = git_inflate(&stream, Z_FINISH);
1498 curpos += stream.next_in - in;
1499 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1500 stream.total_out < sizeof(delta_head));
1501 git_inflate_end(&stream);
1502 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1503 error("delta data unpack-initial failed");
1504 return 0;
1507 /* Examine the initial part of the delta to figure out
1508 * the result size.
1510 data = delta_head;
1512 /* ignore base size */
1513 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1515 /* Read the result size */
1516 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1519 static off_t get_delta_base(struct packed_git *p,
1520 struct pack_window **w_curs,
1521 off_t *curpos,
1522 enum object_type type,
1523 off_t delta_obj_offset)
1525 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1526 off_t base_offset;
1528 /* use_pack() assured us we have [base_info, base_info + 20)
1529 * as a range that we can look at without walking off the
1530 * end of the mapped window. Its actually the hash size
1531 * that is assured. An OFS_DELTA longer than the hash size
1532 * is stupid, as then a REF_DELTA would be smaller to store.
1534 if (type == OBJ_OFS_DELTA) {
1535 unsigned used = 0;
1536 unsigned char c = base_info[used++];
1537 base_offset = c & 127;
1538 while (c & 128) {
1539 base_offset += 1;
1540 if (!base_offset || MSB(base_offset, 7))
1541 return 0; /* overflow */
1542 c = base_info[used++];
1543 base_offset = (base_offset << 7) + (c & 127);
1545 base_offset = delta_obj_offset - base_offset;
1546 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1547 return 0; /* out of bound */
1548 *curpos += used;
1549 } else if (type == OBJ_REF_DELTA) {
1550 /* The base entry _must_ be in the same pack */
1551 base_offset = find_pack_entry_one(base_info, p);
1552 *curpos += 20;
1553 } else
1554 die("I am totally screwed");
1555 return base_offset;
1558 /* forward declaration for a mutually recursive function */
1559 static int packed_object_info(struct packed_git *p, off_t offset,
1560 unsigned long *sizep, int *rtype);
1562 static int packed_delta_info(struct packed_git *p,
1563 struct pack_window **w_curs,
1564 off_t curpos,
1565 enum object_type type,
1566 off_t obj_offset,
1567 unsigned long *sizep)
1569 off_t base_offset;
1571 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1572 if (!base_offset)
1573 return OBJ_BAD;
1574 type = packed_object_info(p, base_offset, NULL, NULL);
1575 if (type <= OBJ_NONE) {
1576 struct revindex_entry *revidx;
1577 const unsigned char *base_sha1;
1578 revidx = find_pack_revindex(p, base_offset);
1579 if (!revidx)
1580 return OBJ_BAD;
1581 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1582 mark_bad_packed_object(p, base_sha1);
1583 type = sha1_object_info(base_sha1, NULL);
1584 if (type <= OBJ_NONE)
1585 return OBJ_BAD;
1588 /* We choose to only get the type of the base object and
1589 * ignore potentially corrupt pack file that expects the delta
1590 * based on a base with a wrong size. This saves tons of
1591 * inflate() calls.
1593 if (sizep) {
1594 *sizep = get_size_from_delta(p, w_curs, curpos);
1595 if (*sizep == 0)
1596 type = OBJ_BAD;
1599 return type;
1602 int unpack_object_header(struct packed_git *p,
1603 struct pack_window **w_curs,
1604 off_t *curpos,
1605 unsigned long *sizep)
1607 unsigned char *base;
1608 unsigned long left;
1609 unsigned long used;
1610 enum object_type type;
1612 /* use_pack() assures us we have [base, base + 20) available
1613 * as a range that we can look at. (Its actually the hash
1614 * size that is assured.) With our object header encoding
1615 * the maximum deflated object size is 2^137, which is just
1616 * insane, so we know won't exceed what we have been given.
1618 base = use_pack(p, w_curs, *curpos, &left);
1619 used = unpack_object_header_buffer(base, left, &type, sizep);
1620 if (!used) {
1621 type = OBJ_BAD;
1622 } else
1623 *curpos += used;
1625 return type;
1628 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1629 unsigned long *sizep, int *rtype)
1631 struct pack_window *w_curs = NULL;
1632 unsigned long size;
1633 off_t curpos = obj_offset;
1634 enum object_type type;
1636 type = unpack_object_header(p, &w_curs, &curpos, &size);
1637 if (rtype)
1638 *rtype = type; /* representation type */
1640 switch (type) {
1641 case OBJ_OFS_DELTA:
1642 case OBJ_REF_DELTA:
1643 type = packed_delta_info(p, &w_curs, curpos,
1644 type, obj_offset, sizep);
1645 break;
1646 case OBJ_COMMIT:
1647 case OBJ_TREE:
1648 case OBJ_BLOB:
1649 case OBJ_TAG:
1650 if (sizep)
1651 *sizep = size;
1652 break;
1653 default:
1654 error("unknown object type %i at offset %"PRIuMAX" in %s",
1655 type, (uintmax_t)obj_offset, p->pack_name);
1656 type = OBJ_BAD;
1658 unuse_pack(&w_curs);
1659 return type;
1662 static void *unpack_compressed_entry(struct packed_git *p,
1663 struct pack_window **w_curs,
1664 off_t curpos,
1665 unsigned long size)
1667 int st;
1668 git_zstream stream;
1669 unsigned char *buffer, *in;
1671 buffer = xmallocz(size);
1672 memset(&stream, 0, sizeof(stream));
1673 stream.next_out = buffer;
1674 stream.avail_out = size + 1;
1676 git_inflate_init(&stream);
1677 do {
1678 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1679 stream.next_in = in;
1680 st = git_inflate(&stream, Z_FINISH);
1681 if (!stream.avail_out)
1682 break; /* the payload is larger than it should be */
1683 curpos += stream.next_in - in;
1684 } while (st == Z_OK || st == Z_BUF_ERROR);
1685 git_inflate_end(&stream);
1686 if ((st != Z_STREAM_END) || stream.total_out != size) {
1687 free(buffer);
1688 return NULL;
1691 return buffer;
1694 #define MAX_DELTA_CACHE (256)
1696 static size_t delta_base_cached;
1698 static struct delta_base_cache_lru_list {
1699 struct delta_base_cache_lru_list *prev;
1700 struct delta_base_cache_lru_list *next;
1701 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1703 static struct delta_base_cache_entry {
1704 struct delta_base_cache_lru_list lru;
1705 void *data;
1706 struct packed_git *p;
1707 off_t base_offset;
1708 unsigned long size;
1709 enum object_type type;
1710 } delta_base_cache[MAX_DELTA_CACHE];
1712 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1714 unsigned long hash;
1716 hash = (unsigned long)p + (unsigned long)base_offset;
1717 hash += (hash >> 8) + (hash >> 16);
1718 return hash % MAX_DELTA_CACHE;
1721 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1723 unsigned long hash = pack_entry_hash(p, base_offset);
1724 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1725 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1728 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1729 unsigned long *base_size, enum object_type *type, int keep_cache)
1731 void *ret;
1732 unsigned long hash = pack_entry_hash(p, base_offset);
1733 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1735 ret = ent->data;
1736 if (!ret || ent->p != p || ent->base_offset != base_offset)
1737 return unpack_entry(p, base_offset, type, base_size);
1739 if (!keep_cache) {
1740 ent->data = NULL;
1741 ent->lru.next->prev = ent->lru.prev;
1742 ent->lru.prev->next = ent->lru.next;
1743 delta_base_cached -= ent->size;
1744 } else {
1745 ret = xmemdupz(ent->data, ent->size);
1747 *type = ent->type;
1748 *base_size = ent->size;
1749 return ret;
1752 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1754 if (ent->data) {
1755 free(ent->data);
1756 ent->data = NULL;
1757 ent->lru.next->prev = ent->lru.prev;
1758 ent->lru.prev->next = ent->lru.next;
1759 delta_base_cached -= ent->size;
1763 void clear_delta_base_cache(void)
1765 unsigned long p;
1766 for (p = 0; p < MAX_DELTA_CACHE; p++)
1767 release_delta_base_cache(&delta_base_cache[p]);
1770 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1771 void *base, unsigned long base_size, enum object_type type)
1773 unsigned long hash = pack_entry_hash(p, base_offset);
1774 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1775 struct delta_base_cache_lru_list *lru;
1777 release_delta_base_cache(ent);
1778 delta_base_cached += base_size;
1780 for (lru = delta_base_cache_lru.next;
1781 delta_base_cached > delta_base_cache_limit
1782 && lru != &delta_base_cache_lru;
1783 lru = lru->next) {
1784 struct delta_base_cache_entry *f = (void *)lru;
1785 if (f->type == OBJ_BLOB)
1786 release_delta_base_cache(f);
1788 for (lru = delta_base_cache_lru.next;
1789 delta_base_cached > delta_base_cache_limit
1790 && lru != &delta_base_cache_lru;
1791 lru = lru->next) {
1792 struct delta_base_cache_entry *f = (void *)lru;
1793 release_delta_base_cache(f);
1796 ent->p = p;
1797 ent->base_offset = base_offset;
1798 ent->type = type;
1799 ent->data = base;
1800 ent->size = base_size;
1801 ent->lru.next = &delta_base_cache_lru;
1802 ent->lru.prev = delta_base_cache_lru.prev;
1803 delta_base_cache_lru.prev->next = &ent->lru;
1804 delta_base_cache_lru.prev = &ent->lru;
1807 static void *read_object(const unsigned char *sha1, enum object_type *type,
1808 unsigned long *size);
1810 static void *unpack_delta_entry(struct packed_git *p,
1811 struct pack_window **w_curs,
1812 off_t curpos,
1813 unsigned long delta_size,
1814 off_t obj_offset,
1815 enum object_type *type,
1816 unsigned long *sizep)
1818 void *delta_data, *result, *base;
1819 unsigned long base_size;
1820 off_t base_offset;
1822 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1823 if (!base_offset) {
1824 error("failed to validate delta base reference "
1825 "at offset %"PRIuMAX" from %s",
1826 (uintmax_t)curpos, p->pack_name);
1827 return NULL;
1829 unuse_pack(w_curs);
1830 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1831 if (!base) {
1833 * We're probably in deep shit, but let's try to fetch
1834 * the required base anyway from another pack or loose.
1835 * This is costly but should happen only in the presence
1836 * of a corrupted pack, and is better than failing outright.
1838 struct revindex_entry *revidx;
1839 const unsigned char *base_sha1;
1840 revidx = find_pack_revindex(p, base_offset);
1841 if (!revidx)
1842 return NULL;
1843 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1844 error("failed to read delta base object %s"
1845 " at offset %"PRIuMAX" from %s",
1846 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1847 p->pack_name);
1848 mark_bad_packed_object(p, base_sha1);
1849 base = read_object(base_sha1, type, &base_size);
1850 if (!base)
1851 return NULL;
1854 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1855 if (!delta_data) {
1856 error("failed to unpack compressed delta "
1857 "at offset %"PRIuMAX" from %s",
1858 (uintmax_t)curpos, p->pack_name);
1859 free(base);
1860 return NULL;
1862 result = patch_delta(base, base_size,
1863 delta_data, delta_size,
1864 sizep);
1865 if (!result)
1866 die("failed to apply delta");
1867 free(delta_data);
1868 add_delta_base_cache(p, base_offset, base, base_size, *type);
1869 return result;
1872 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1874 static FILE *log_file;
1876 if (!log_file) {
1877 log_file = fopen(log_pack_access, "w");
1878 if (!log_file) {
1879 error("cannot open pack access log '%s' for writing: %s",
1880 log_pack_access, strerror(errno));
1881 log_pack_access = NULL;
1882 return;
1885 fprintf(log_file, "%s %"PRIuMAX"\n",
1886 p->pack_name, (uintmax_t)obj_offset);
1887 fflush(log_file);
1890 int do_check_packed_object_crc;
1892 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1893 enum object_type *type, unsigned long *sizep)
1895 struct pack_window *w_curs = NULL;
1896 off_t curpos = obj_offset;
1897 void *data;
1899 if (log_pack_access)
1900 write_pack_access_log(p, obj_offset);
1902 if (do_check_packed_object_crc && p->index_version > 1) {
1903 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1904 unsigned long len = revidx[1].offset - obj_offset;
1905 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1906 const unsigned char *sha1 =
1907 nth_packed_object_sha1(p, revidx->nr);
1908 error("bad packed object CRC for %s",
1909 sha1_to_hex(sha1));
1910 mark_bad_packed_object(p, sha1);
1911 unuse_pack(&w_curs);
1912 return NULL;
1916 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1917 switch (*type) {
1918 case OBJ_OFS_DELTA:
1919 case OBJ_REF_DELTA:
1920 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1921 obj_offset, type, sizep);
1922 break;
1923 case OBJ_COMMIT:
1924 case OBJ_TREE:
1925 case OBJ_BLOB:
1926 case OBJ_TAG:
1927 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1928 break;
1929 default:
1930 data = NULL;
1931 error("unknown object type %i at offset %"PRIuMAX" in %s",
1932 *type, (uintmax_t)obj_offset, p->pack_name);
1934 unuse_pack(&w_curs);
1935 return data;
1938 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1939 uint32_t n)
1941 const unsigned char *index = p->index_data;
1942 if (!index) {
1943 if (open_pack_index(p))
1944 return NULL;
1945 index = p->index_data;
1947 if (n >= p->num_objects)
1948 return NULL;
1949 index += 4 * 256;
1950 if (p->index_version == 1) {
1951 return index + 24 * n + 4;
1952 } else {
1953 index += 8;
1954 return index + 20 * n;
1958 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1960 const unsigned char *index = p->index_data;
1961 index += 4 * 256;
1962 if (p->index_version == 1) {
1963 return ntohl(*((uint32_t *)(index + 24 * n)));
1964 } else {
1965 uint32_t off;
1966 index += 8 + p->num_objects * (20 + 4);
1967 off = ntohl(*((uint32_t *)(index + 4 * n)));
1968 if (!(off & 0x80000000))
1969 return off;
1970 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1971 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1972 ntohl(*((uint32_t *)(index + 4)));
1976 off_t find_pack_entry_one(const unsigned char *sha1,
1977 struct packed_git *p)
1979 const uint32_t *level1_ofs = p->index_data;
1980 const unsigned char *index = p->index_data;
1981 unsigned hi, lo, stride;
1982 static int use_lookup = -1;
1983 static int debug_lookup = -1;
1985 if (debug_lookup < 0)
1986 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1988 if (!index) {
1989 if (open_pack_index(p))
1990 return 0;
1991 level1_ofs = p->index_data;
1992 index = p->index_data;
1994 if (p->index_version > 1) {
1995 level1_ofs += 2;
1996 index += 8;
1998 index += 4 * 256;
1999 hi = ntohl(level1_ofs[*sha1]);
2000 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2001 if (p->index_version > 1) {
2002 stride = 20;
2003 } else {
2004 stride = 24;
2005 index += 4;
2008 if (debug_lookup)
2009 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2010 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2012 if (use_lookup < 0)
2013 use_lookup = !!getenv("GIT_USE_LOOKUP");
2014 if (use_lookup) {
2015 int pos = sha1_entry_pos(index, stride, 0,
2016 lo, hi, p->num_objects, sha1);
2017 if (pos < 0)
2018 return 0;
2019 return nth_packed_object_offset(p, pos);
2022 do {
2023 unsigned mi = (lo + hi) / 2;
2024 int cmp = hashcmp(index + mi * stride, sha1);
2026 if (debug_lookup)
2027 printf("lo %u hi %u rg %u mi %u\n",
2028 lo, hi, hi - lo, mi);
2029 if (!cmp)
2030 return nth_packed_object_offset(p, mi);
2031 if (cmp > 0)
2032 hi = mi;
2033 else
2034 lo = mi+1;
2035 } while (lo < hi);
2036 return 0;
2039 int is_pack_valid(struct packed_git *p)
2041 /* An already open pack is known to be valid. */
2042 if (p->pack_fd != -1)
2043 return 1;
2045 /* If the pack has one window completely covering the
2046 * file size, the pack is known to be valid even if
2047 * the descriptor is not currently open.
2049 if (p->windows) {
2050 struct pack_window *w = p->windows;
2052 if (!w->offset && w->len == p->pack_size)
2053 return 1;
2056 /* Force the pack to open to prove its valid. */
2057 return !open_packed_git(p);
2060 static int fill_pack_entry(const unsigned char *sha1,
2061 struct pack_entry *e,
2062 struct packed_git *p)
2064 off_t offset;
2066 if (p->num_bad_objects) {
2067 unsigned i;
2068 for (i = 0; i < p->num_bad_objects; i++)
2069 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2070 return 0;
2073 offset = find_pack_entry_one(sha1, p);
2074 if (!offset)
2075 return 0;
2078 * We are about to tell the caller where they can locate the
2079 * requested object. We better make sure the packfile is
2080 * still here and can be accessed before supplying that
2081 * answer, as it may have been deleted since the index was
2082 * loaded!
2084 if (!is_pack_valid(p)) {
2085 warning("packfile %s cannot be accessed", p->pack_name);
2086 return 0;
2088 e->offset = offset;
2089 e->p = p;
2090 hashcpy(e->sha1, sha1);
2091 return 1;
2094 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2096 struct packed_git *p;
2098 prepare_packed_git();
2099 if (!packed_git)
2100 return 0;
2102 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2103 return 1;
2105 for (p = packed_git; p; p = p->next) {
2106 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2107 continue;
2109 last_found_pack = p;
2110 return 1;
2112 return 0;
2115 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2116 struct packed_git *packs)
2118 struct packed_git *p;
2120 for (p = packs; p; p = p->next) {
2121 if (find_pack_entry_one(sha1, p))
2122 return p;
2124 return NULL;
2128 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2130 int status;
2131 unsigned long mapsize, size;
2132 void *map;
2133 git_zstream stream;
2134 char hdr[32];
2136 map = map_sha1_file(sha1, &mapsize);
2137 if (!map)
2138 return error("unable to find %s", sha1_to_hex(sha1));
2139 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2140 status = error("unable to unpack %s header",
2141 sha1_to_hex(sha1));
2142 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2143 status = error("unable to parse %s header", sha1_to_hex(sha1));
2144 else if (sizep)
2145 *sizep = size;
2146 git_inflate_end(&stream);
2147 munmap(map, mapsize);
2148 return status;
2151 /* returns enum object_type or negative */
2152 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2154 struct cached_object *co;
2155 struct pack_entry e;
2156 int status, rtype;
2158 co = find_cached_object(sha1);
2159 if (co) {
2160 if (oi->sizep)
2161 *(oi->sizep) = co->size;
2162 oi->whence = OI_CACHED;
2163 return co->type;
2166 if (!find_pack_entry(sha1, &e)) {
2167 /* Most likely it's a loose object. */
2168 status = sha1_loose_object_info(sha1, oi->sizep);
2169 if (status >= 0) {
2170 oi->whence = OI_LOOSE;
2171 return status;
2174 /* Not a loose object; someone else may have just packed it. */
2175 reprepare_packed_git();
2176 if (!find_pack_entry(sha1, &e))
2177 return status;
2180 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2181 if (status < 0) {
2182 mark_bad_packed_object(e.p, sha1);
2183 status = sha1_object_info_extended(sha1, oi);
2184 } else if (in_delta_base_cache(e.p, e.offset)) {
2185 oi->whence = OI_DBCACHED;
2186 } else {
2187 oi->whence = OI_PACKED;
2188 oi->u.packed.offset = e.offset;
2189 oi->u.packed.pack = e.p;
2190 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2191 rtype == OBJ_OFS_DELTA);
2194 return status;
2197 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2199 struct object_info oi;
2201 oi.sizep = sizep;
2202 return sha1_object_info_extended(sha1, &oi);
2205 static void *read_packed_sha1(const unsigned char *sha1,
2206 enum object_type *type, unsigned long *size)
2208 struct pack_entry e;
2209 void *data;
2211 if (!find_pack_entry(sha1, &e))
2212 return NULL;
2213 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2214 if (!data) {
2216 * We're probably in deep shit, but let's try to fetch
2217 * the required object anyway from another pack or loose.
2218 * This should happen only in the presence of a corrupted
2219 * pack, and is better than failing outright.
2221 error("failed to read object %s at offset %"PRIuMAX" from %s",
2222 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2223 mark_bad_packed_object(e.p, sha1);
2224 data = read_object(sha1, type, size);
2226 return data;
2229 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2230 unsigned char *sha1)
2232 struct cached_object *co;
2234 hash_sha1_file(buf, len, typename(type), sha1);
2235 if (has_sha1_file(sha1) || find_cached_object(sha1))
2236 return 0;
2237 if (cached_object_alloc <= cached_object_nr) {
2238 cached_object_alloc = alloc_nr(cached_object_alloc);
2239 cached_objects = xrealloc(cached_objects,
2240 sizeof(*cached_objects) *
2241 cached_object_alloc);
2243 co = &cached_objects[cached_object_nr++];
2244 co->size = len;
2245 co->type = type;
2246 co->buf = xmalloc(len);
2247 memcpy(co->buf, buf, len);
2248 hashcpy(co->sha1, sha1);
2249 return 0;
2252 static void *read_object(const unsigned char *sha1, enum object_type *type,
2253 unsigned long *size)
2255 unsigned long mapsize;
2256 void *map, *buf;
2257 struct cached_object *co;
2259 co = find_cached_object(sha1);
2260 if (co) {
2261 *type = co->type;
2262 *size = co->size;
2263 return xmemdupz(co->buf, co->size);
2266 buf = read_packed_sha1(sha1, type, size);
2267 if (buf)
2268 return buf;
2269 map = map_sha1_file(sha1, &mapsize);
2270 if (map) {
2271 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2272 munmap(map, mapsize);
2273 return buf;
2275 reprepare_packed_git();
2276 return read_packed_sha1(sha1, type, size);
2280 * This function dies on corrupt objects; the callers who want to
2281 * deal with them should arrange to call read_object() and give error
2282 * messages themselves.
2284 void *read_sha1_file_extended(const unsigned char *sha1,
2285 enum object_type *type,
2286 unsigned long *size,
2287 unsigned flag)
2289 void *data;
2290 char *path;
2291 const struct packed_git *p;
2292 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2293 ? lookup_replace_object(sha1) : sha1;
2295 errno = 0;
2296 data = read_object(repl, type, size);
2297 if (data)
2298 return data;
2300 if (errno && errno != ENOENT)
2301 die_errno("failed to read object %s", sha1_to_hex(sha1));
2303 /* die if we replaced an object with one that does not exist */
2304 if (repl != sha1)
2305 die("replacement %s not found for %s",
2306 sha1_to_hex(repl), sha1_to_hex(sha1));
2308 if (has_loose_object(repl)) {
2309 path = sha1_file_name(sha1);
2310 die("loose object %s (stored in %s) is corrupt",
2311 sha1_to_hex(repl), path);
2314 if ((p = has_packed_and_bad(repl)) != NULL)
2315 die("packed object %s (stored in %s) is corrupt",
2316 sha1_to_hex(repl), p->pack_name);
2318 return NULL;
2321 void *read_object_with_reference(const unsigned char *sha1,
2322 const char *required_type_name,
2323 unsigned long *size,
2324 unsigned char *actual_sha1_return)
2326 enum object_type type, required_type;
2327 void *buffer;
2328 unsigned long isize;
2329 unsigned char actual_sha1[20];
2331 required_type = type_from_string(required_type_name);
2332 hashcpy(actual_sha1, sha1);
2333 while (1) {
2334 int ref_length = -1;
2335 const char *ref_type = NULL;
2337 buffer = read_sha1_file(actual_sha1, &type, &isize);
2338 if (!buffer)
2339 return NULL;
2340 if (type == required_type) {
2341 *size = isize;
2342 if (actual_sha1_return)
2343 hashcpy(actual_sha1_return, actual_sha1);
2344 return buffer;
2346 /* Handle references */
2347 else if (type == OBJ_COMMIT)
2348 ref_type = "tree ";
2349 else if (type == OBJ_TAG)
2350 ref_type = "object ";
2351 else {
2352 free(buffer);
2353 return NULL;
2355 ref_length = strlen(ref_type);
2357 if (ref_length + 40 > isize ||
2358 memcmp(buffer, ref_type, ref_length) ||
2359 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2360 free(buffer);
2361 return NULL;
2363 free(buffer);
2364 /* Now we have the ID of the referred-to object in
2365 * actual_sha1. Check again. */
2369 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2370 const char *type, unsigned char *sha1,
2371 char *hdr, int *hdrlen)
2373 git_SHA_CTX c;
2375 /* Generate the header */
2376 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2378 /* Sha1.. */
2379 git_SHA1_Init(&c);
2380 git_SHA1_Update(&c, hdr, *hdrlen);
2381 git_SHA1_Update(&c, buf, len);
2382 git_SHA1_Final(sha1, &c);
2386 * Move the just written object into its final resting place.
2387 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2388 * "moving" is only a part of what it does, when no patch between
2389 * master to pu changes the call sites of this function.
2391 int move_temp_to_file(const char *tmpfile, const char *filename)
2393 int ret = 0;
2395 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2396 goto try_rename;
2397 else if (link(tmpfile, filename))
2398 ret = errno;
2401 * Coda hack - coda doesn't like cross-directory links,
2402 * so we fall back to a rename, which will mean that it
2403 * won't be able to check collisions, but that's not a
2404 * big deal.
2406 * The same holds for FAT formatted media.
2408 * When this succeeds, we just return. We have nothing
2409 * left to unlink.
2411 if (ret && ret != EEXIST) {
2412 try_rename:
2413 if (!rename(tmpfile, filename))
2414 goto out;
2415 ret = errno;
2417 unlink_or_warn(tmpfile);
2418 if (ret) {
2419 if (ret != EEXIST) {
2420 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2422 /* FIXME!!! Collision check here ? */
2425 out:
2426 if (adjust_shared_perm(filename))
2427 return error("unable to set permission to '%s'", filename);
2428 return 0;
2431 static int write_buffer(int fd, const void *buf, size_t len)
2433 if (write_in_full(fd, buf, len) < 0)
2434 return error("file write error (%s)", strerror(errno));
2435 return 0;
2438 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2439 unsigned char *sha1)
2441 char hdr[32];
2442 int hdrlen;
2443 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2444 return 0;
2447 /* Finalize a file on disk, and close it. */
2448 static void close_sha1_file(int fd)
2450 if (fsync_object_files)
2451 fsync_or_die(fd, "sha1 file");
2452 if (close(fd) != 0)
2453 die_errno("error when closing sha1 file");
2456 /* Size of directory component, including the ending '/' */
2457 static inline int directory_size(const char *filename)
2459 const char *s = strrchr(filename, '/');
2460 if (!s)
2461 return 0;
2462 return s - filename + 1;
2466 * This creates a temporary file in the same directory as the final
2467 * 'filename'
2469 * We want to avoid cross-directory filename renames, because those
2470 * can have problems on various filesystems (FAT, NFS, Coda).
2472 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2474 int fd, dirlen = directory_size(filename);
2476 if (dirlen + 20 > bufsiz) {
2477 errno = ENAMETOOLONG;
2478 return -1;
2480 memcpy(buffer, filename, dirlen);
2481 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2482 fd = git_mkstemp_mode(buffer, 0444);
2483 if (fd < 0 && dirlen && errno == ENOENT) {
2484 /* Make sure the directory exists */
2485 memcpy(buffer, filename, dirlen);
2486 buffer[dirlen-1] = 0;
2487 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2488 return -1;
2490 /* Try again */
2491 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2492 fd = git_mkstemp_mode(buffer, 0444);
2494 return fd;
2497 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2498 const void *buf, unsigned long len, time_t mtime)
2500 int fd, ret;
2501 unsigned char compressed[4096];
2502 git_zstream stream;
2503 git_SHA_CTX c;
2504 unsigned char parano_sha1[20];
2505 char *filename;
2506 static char tmp_file[PATH_MAX];
2508 filename = sha1_file_name(sha1);
2509 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2510 if (fd < 0) {
2511 if (errno == EACCES)
2512 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2513 else
2514 return error("unable to create temporary file: %s", strerror(errno));
2517 /* Set it up */
2518 memset(&stream, 0, sizeof(stream));
2519 git_deflate_init(&stream, zlib_compression_level);
2520 stream.next_out = compressed;
2521 stream.avail_out = sizeof(compressed);
2522 git_SHA1_Init(&c);
2524 /* First header.. */
2525 stream.next_in = (unsigned char *)hdr;
2526 stream.avail_in = hdrlen;
2527 while (git_deflate(&stream, 0) == Z_OK)
2528 ; /* nothing */
2529 git_SHA1_Update(&c, hdr, hdrlen);
2531 /* Then the data itself.. */
2532 stream.next_in = (void *)buf;
2533 stream.avail_in = len;
2534 do {
2535 unsigned char *in0 = stream.next_in;
2536 ret = git_deflate(&stream, Z_FINISH);
2537 git_SHA1_Update(&c, in0, stream.next_in - in0);
2538 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2539 die("unable to write sha1 file");
2540 stream.next_out = compressed;
2541 stream.avail_out = sizeof(compressed);
2542 } while (ret == Z_OK);
2544 if (ret != Z_STREAM_END)
2545 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2546 ret = git_deflate_end_gently(&stream);
2547 if (ret != Z_OK)
2548 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2549 git_SHA1_Final(parano_sha1, &c);
2550 if (hashcmp(sha1, parano_sha1) != 0)
2551 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2553 close_sha1_file(fd);
2555 if (mtime) {
2556 struct utimbuf utb;
2557 utb.actime = mtime;
2558 utb.modtime = mtime;
2559 if (utime(tmp_file, &utb) < 0)
2560 warning("failed utime() on %s: %s",
2561 tmp_file, strerror(errno));
2564 return move_temp_to_file(tmp_file, filename);
2567 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2569 unsigned char sha1[20];
2570 char hdr[32];
2571 int hdrlen;
2573 /* Normally if we have it in the pack then we do not bother writing
2574 * it out into .git/objects/??/?{38} file.
2576 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2577 if (returnsha1)
2578 hashcpy(returnsha1, sha1);
2579 if (has_sha1_file(sha1))
2580 return 0;
2581 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2584 int force_object_loose(const unsigned char *sha1, time_t mtime)
2586 void *buf;
2587 unsigned long len;
2588 enum object_type type;
2589 char hdr[32];
2590 int hdrlen;
2591 int ret;
2593 if (has_loose_object(sha1))
2594 return 0;
2595 buf = read_packed_sha1(sha1, &type, &len);
2596 if (!buf)
2597 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2598 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2599 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2600 free(buf);
2602 return ret;
2605 int has_pack_index(const unsigned char *sha1)
2607 struct stat st;
2608 if (stat(sha1_pack_index_name(sha1), &st))
2609 return 0;
2610 return 1;
2613 int has_sha1_pack(const unsigned char *sha1)
2615 struct pack_entry e;
2616 return find_pack_entry(sha1, &e);
2619 int has_sha1_file(const unsigned char *sha1)
2621 struct pack_entry e;
2623 if (find_pack_entry(sha1, &e))
2624 return 1;
2625 return has_loose_object(sha1);
2628 static void check_tree(const void *buf, size_t size)
2630 struct tree_desc desc;
2631 struct name_entry entry;
2633 init_tree_desc(&desc, buf, size);
2634 while (tree_entry(&desc, &entry))
2635 /* do nothing
2636 * tree_entry() will die() on malformed entries */
2640 static void check_commit(const void *buf, size_t size)
2642 struct commit c;
2643 memset(&c, 0, sizeof(c));
2644 if (parse_commit_buffer(&c, buf, size))
2645 die("corrupt commit");
2648 static void check_tag(const void *buf, size_t size)
2650 struct tag t;
2651 memset(&t, 0, sizeof(t));
2652 if (parse_tag_buffer(&t, buf, size))
2653 die("corrupt tag");
2656 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2657 enum object_type type,
2658 const char *path, unsigned flags)
2660 int ret, re_allocated = 0;
2661 int write_object = flags & HASH_WRITE_OBJECT;
2663 if (!type)
2664 type = OBJ_BLOB;
2667 * Convert blobs to git internal format
2669 if ((type == OBJ_BLOB) && path) {
2670 struct strbuf nbuf = STRBUF_INIT;
2671 if (convert_to_git(path, buf, size, &nbuf,
2672 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2673 buf = strbuf_detach(&nbuf, &size);
2674 re_allocated = 1;
2677 if (flags & HASH_FORMAT_CHECK) {
2678 if (type == OBJ_TREE)
2679 check_tree(buf, size);
2680 if (type == OBJ_COMMIT)
2681 check_commit(buf, size);
2682 if (type == OBJ_TAG)
2683 check_tag(buf, size);
2686 if (write_object)
2687 ret = write_sha1_file(buf, size, typename(type), sha1);
2688 else
2689 ret = hash_sha1_file(buf, size, typename(type), sha1);
2690 if (re_allocated)
2691 free(buf);
2692 return ret;
2695 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2696 const char *path, unsigned flags)
2698 struct strbuf sbuf = STRBUF_INIT;
2699 int ret;
2701 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2702 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2703 else
2704 ret = -1;
2705 strbuf_release(&sbuf);
2706 return ret;
2709 #define SMALL_FILE_SIZE (32*1024)
2711 static int index_core(unsigned char *sha1, int fd, size_t size,
2712 enum object_type type, const char *path,
2713 unsigned flags)
2715 int ret;
2717 if (!size) {
2718 ret = index_mem(sha1, NULL, size, type, path, flags);
2719 } else if (size <= SMALL_FILE_SIZE) {
2720 char *buf = xmalloc(size);
2721 if (size == read_in_full(fd, buf, size))
2722 ret = index_mem(sha1, buf, size, type, path, flags);
2723 else
2724 ret = error("short read %s", strerror(errno));
2725 free(buf);
2726 } else {
2727 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2728 ret = index_mem(sha1, buf, size, type, path, flags);
2729 munmap(buf, size);
2731 return ret;
2735 * This creates one packfile per large blob unless bulk-checkin
2736 * machinery is "plugged".
2738 * This also bypasses the usual "convert-to-git" dance, and that is on
2739 * purpose. We could write a streaming version of the converting
2740 * functions and insert that before feeding the data to fast-import
2741 * (or equivalent in-core API described above). However, that is
2742 * somewhat complicated, as we do not know the size of the filter
2743 * result, which we need to know beforehand when writing a git object.
2744 * Since the primary motivation for trying to stream from the working
2745 * tree file and to avoid mmaping it in core is to deal with large
2746 * binary blobs, they generally do not want to get any conversion, and
2747 * callers should avoid this code path when filters are requested.
2749 static int index_stream(unsigned char *sha1, int fd, size_t size,
2750 enum object_type type, const char *path,
2751 unsigned flags)
2753 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2756 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2757 enum object_type type, const char *path, unsigned flags)
2759 int ret;
2760 size_t size = xsize_t(st->st_size);
2762 if (!S_ISREG(st->st_mode))
2763 ret = index_pipe(sha1, fd, type, path, flags);
2764 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2765 (path && would_convert_to_git(path, NULL, 0, 0)))
2766 ret = index_core(sha1, fd, size, type, path, flags);
2767 else
2768 ret = index_stream(sha1, fd, size, type, path, flags);
2769 close(fd);
2770 return ret;
2773 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2775 int fd;
2776 struct strbuf sb = STRBUF_INIT;
2778 switch (st->st_mode & S_IFMT) {
2779 case S_IFREG:
2780 fd = open(path, O_RDONLY);
2781 if (fd < 0)
2782 return error("open(\"%s\"): %s", path,
2783 strerror(errno));
2784 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2785 return error("%s: failed to insert into database",
2786 path);
2787 break;
2788 case S_IFLNK:
2789 if (strbuf_readlink(&sb, path, st->st_size)) {
2790 char *errstr = strerror(errno);
2791 return error("readlink(\"%s\"): %s", path,
2792 errstr);
2794 if (!(flags & HASH_WRITE_OBJECT))
2795 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2796 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2797 return error("%s: failed to insert into database",
2798 path);
2799 strbuf_release(&sb);
2800 break;
2801 case S_IFDIR:
2802 return resolve_gitlink_ref(path, "HEAD", sha1);
2803 default:
2804 return error("%s: unsupported file type", path);
2806 return 0;
2809 int read_pack_header(int fd, struct pack_header *header)
2811 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2812 /* "eof before pack header was fully read" */
2813 return PH_ERROR_EOF;
2815 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2816 /* "protocol error (pack signature mismatch detected)" */
2817 return PH_ERROR_PACK_SIGNATURE;
2818 if (!pack_version_ok(header->hdr_version))
2819 /* "protocol error (pack version unsupported)" */
2820 return PH_ERROR_PROTOCOL;
2821 return 0;
2824 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2826 enum object_type type = sha1_object_info(sha1, NULL);
2827 if (type < 0)
2828 die("%s is not a valid object", sha1_to_hex(sha1));
2829 if (type != expect)
2830 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2831 typename(expect));