msysGit-based Git for Windows 1.x was superseded by Git for Windows 2.x
[git/mingw/4msysgit.git] / sha1_file.c
blob3f70b1d86aaa8e0648d8a3c6ebb6aca701ae7475
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "blob.h"
14 #include "commit.h"
15 #include "run-command.h"
16 #include "tag.h"
17 #include "tree.h"
18 #include "tree-walk.h"
19 #include "refs.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
24 #include "dir.h"
26 #ifndef O_NOATIME
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
29 #else
30 #define O_NOATIME 0
31 #endif
32 #endif
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s) { return s; }
37 const unsigned char null_sha1[20];
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
45 static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50 } *cached_objects;
51 static int cached_object_nr, cached_object_alloc;
53 static struct cached_object empty_tree = {
54 EMPTY_TREE_SHA1_BIN_LITERAL,
55 OBJ_TREE,
56 "",
61 * A pointer to the last packed_git in which an object was found.
62 * When an object is sought, we look in this packfile first, because
63 * objects that are looked up at similar times are often in the same
64 * packfile as one another.
66 static struct packed_git *last_found_pack;
68 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 int i;
71 struct cached_object *co = cached_objects;
73 for (i = 0; i < cached_object_nr; i++, co++) {
74 if (!hashcmp(co->sha1, sha1))
75 return co;
77 if (!hashcmp(sha1, empty_tree.sha1))
78 return &empty_tree;
79 return NULL;
82 int mkdir_in_gitdir(const char *path)
84 if (mkdir(path, 0777)) {
85 int saved_errno = errno;
86 struct stat st;
87 struct strbuf sb = STRBUF_INIT;
89 if (errno != EEXIST)
90 return -1;
92 * Are we looking at a path in a symlinked worktree
93 * whose original repository does not yet have it?
94 * e.g. .git/rr-cache pointing at its original
95 * repository in which the user hasn't performed any
96 * conflict resolution yet?
98 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
99 strbuf_readlink(&sb, path, st.st_size) ||
100 !is_absolute_path(sb.buf) ||
101 mkdir(sb.buf, 0777)) {
102 strbuf_release(&sb);
103 errno = saved_errno;
104 return -1;
106 strbuf_release(&sb);
108 return adjust_shared_perm(path);
111 enum scld_error safe_create_leading_directories(char *path)
113 char *next_component = path + offset_1st_component(path);
114 enum scld_error ret = SCLD_OK;
116 while (ret == SCLD_OK && next_component) {
117 struct stat st;
118 char *slash = next_component, slash_character;
120 while (*slash && !is_dir_sep(*slash))
121 slash++;
123 if (!*slash)
124 break;
126 next_component = slash + 1;
127 while (is_dir_sep(*next_component))
128 next_component++;
129 if (!*next_component)
130 break;
132 slash_character = *slash;
133 *slash = '\0';
134 if (!stat(path, &st)) {
135 /* path exists */
136 if (!S_ISDIR(st.st_mode))
137 ret = SCLD_EXISTS;
138 } else if (mkdir(path, 0777)) {
139 if (errno == EEXIST &&
140 !stat(path, &st) && S_ISDIR(st.st_mode))
141 ; /* somebody created it since we checked */
142 else if (errno == ENOENT)
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
150 * trying again:
152 ret = SCLD_VANISHED;
153 else
154 ret = SCLD_FAILED;
155 } else if (adjust_shared_perm(path)) {
156 ret = SCLD_PERMS;
158 *slash = slash_character;
160 return ret;
163 enum scld_error safe_create_leading_directories_const(const char *path)
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf = xstrdup(path);
167 enum scld_error result = safe_create_leading_directories(buf);
168 free(buf);
169 return result;
172 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 int i;
175 for (i = 0; i < 20; i++) {
176 static char hex[] = "0123456789abcdef";
177 unsigned int val = sha1[i];
178 char *pos = pathbuf + i*2 + (i > 0);
179 *pos++ = hex[val >> 4];
180 *pos = hex[val & 0xf];
184 const char *sha1_file_name(const unsigned char *sha1)
186 static char buf[PATH_MAX];
187 const char *objdir;
188 int len;
190 objdir = get_object_directory();
191 len = strlen(objdir);
193 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
194 if (len + 43 > PATH_MAX)
195 die("insanely long object directory %s", objdir);
196 memcpy(buf, objdir, len);
197 buf[len] = '/';
198 buf[len+3] = '/';
199 buf[len+42] = '\0';
200 fill_sha1_path(buf + len + 1, sha1);
201 return buf;
205 * Return the name of the pack or index file with the specified sha1
206 * in its filename. *base and *name are scratch space that must be
207 * provided by the caller. which should be "pack" or "idx".
209 static char *sha1_get_pack_name(const unsigned char *sha1,
210 char **name, char **base, const char *which)
212 static const char hex[] = "0123456789abcdef";
213 char *buf;
214 int i;
216 if (!*base) {
217 const char *sha1_file_directory = get_object_directory();
218 int len = strlen(sha1_file_directory);
219 *base = xmalloc(len + 60);
220 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
221 sha1_file_directory, which);
222 *name = *base + len + 11;
225 buf = *name;
227 for (i = 0; i < 20; i++) {
228 unsigned int val = *sha1++;
229 *buf++ = hex[val >> 4];
230 *buf++ = hex[val & 0xf];
233 return *base;
236 char *sha1_pack_name(const unsigned char *sha1)
238 static char *name, *base;
240 return sha1_get_pack_name(sha1, &name, &base, "pack");
243 char *sha1_pack_index_name(const unsigned char *sha1)
245 static char *name, *base;
247 return sha1_get_pack_name(sha1, &name, &base, "idx");
250 struct alternate_object_database *alt_odb_list;
251 static struct alternate_object_database **alt_odb_tail;
254 * Prepare alternate object database registry.
256 * The variable alt_odb_list points at the list of struct
257 * alternate_object_database. The elements on this list come from
258 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
259 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
260 * whose contents is similar to that environment variable but can be
261 * LF separated. Its base points at a statically allocated buffer that
262 * contains "/the/directory/corresponding/to/.git/objects/...", while
263 * its name points just after the slash at the end of ".git/objects/"
264 * in the example above, and has enough space to hold 40-byte hex
265 * SHA1, an extra slash for the first level indirection, and the
266 * terminating NUL.
268 static int link_alt_odb_entry(const char *entry, const char *relative_base,
269 int depth, const char *normalized_objdir)
271 struct alternate_object_database *ent;
272 struct alternate_object_database *alt;
273 int pfxlen, entlen;
274 struct strbuf pathbuf = STRBUF_INIT;
276 if (!is_absolute_path(entry) && relative_base) {
277 strbuf_addstr(&pathbuf, real_path(relative_base));
278 strbuf_addch(&pathbuf, '/');
280 strbuf_addstr(&pathbuf, entry);
282 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284 pfxlen = strlen(pathbuf.buf);
287 * The trailing slash after the directory name is given by
288 * this function at the end. Remove duplicates.
290 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
291 pfxlen -= 1;
293 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
294 ent = xmalloc(sizeof(*ent) + entlen);
295 memcpy(ent->base, pathbuf.buf, pfxlen);
296 strbuf_release(&pathbuf);
298 ent->name = ent->base + pfxlen + 1;
299 ent->base[pfxlen + 3] = '/';
300 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302 /* Detect cases where alternate disappeared */
303 if (!is_directory(ent->base)) {
304 error("object directory %s does not exist; "
305 "check .git/objects/info/alternates.",
306 ent->base);
307 free(ent);
308 return -1;
311 /* Prevent the common mistake of listing the same
312 * thing twice, or object directory itself.
314 for (alt = alt_odb_list; alt; alt = alt->next) {
315 if (pfxlen == alt->name - alt->base - 1 &&
316 !memcmp(ent->base, alt->base, pfxlen)) {
317 free(ent);
318 return -1;
321 if (!strcmp_icase(ent->base, normalized_objdir)) {
322 free(ent);
323 return -1;
326 /* add the alternate entry */
327 *alt_odb_tail = ent;
328 alt_odb_tail = &(ent->next);
329 ent->next = NULL;
331 /* recursively add alternates */
332 read_info_alternates(ent->base, depth + 1);
334 ent->base[pfxlen] = '/';
336 return 0;
339 static void link_alt_odb_entries(const char *alt, int len, int sep,
340 const char *relative_base, int depth)
342 struct string_list entries = STRING_LIST_INIT_NODUP;
343 char *alt_copy;
344 int i;
345 struct strbuf objdirbuf = STRBUF_INIT;
347 if (depth > 5) {
348 error("%s: ignoring alternate object stores, nesting too deep.",
349 relative_base);
350 return;
353 strbuf_addstr(&objdirbuf, absolute_path(get_object_directory()));
354 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356 alt_copy = xmemdupz(alt, len);
357 string_list_split_in_place(&entries, alt_copy, sep, -1);
358 for (i = 0; i < entries.nr; i++) {
359 const char *entry = entries.items[i].string;
360 if (entry[0] == '\0' || entry[0] == '#')
361 continue;
362 if (!is_absolute_path(entry) && depth) {
363 error("%s: ignoring relative alternate object store %s",
364 relative_base, entry);
365 } else {
366 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
369 string_list_clear(&entries, 0);
370 free(alt_copy);
371 strbuf_release(&objdirbuf);
374 void read_info_alternates(const char * relative_base, int depth)
376 char *map;
377 size_t mapsz;
378 struct stat st;
379 const char alt_file_name[] = "info/alternates";
380 /* Given that relative_base is no longer than PATH_MAX,
381 ensure that "path" has enough space to append "/", the
382 file name, "info/alternates", and a trailing NUL. */
383 char path[PATH_MAX + 1 + sizeof alt_file_name];
384 int fd;
386 sprintf(path, "%s/%s", relative_base, alt_file_name);
387 fd = git_open_noatime(path);
388 if (fd < 0)
389 return;
390 if (fstat(fd, &st) || (st.st_size == 0)) {
391 close(fd);
392 return;
394 mapsz = xsize_t(st.st_size);
395 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
396 close(fd);
398 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
400 munmap(map, mapsz);
403 void add_to_alternates_file(const char *reference)
405 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
406 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
407 char *alt = mkpath("%s\n", reference);
408 write_or_die(fd, alt, strlen(alt));
409 if (commit_lock_file(lock))
410 die("could not close alternates file");
411 if (alt_odb_tail)
412 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
415 void foreach_alt_odb(alt_odb_fn fn, void *cb)
417 struct alternate_object_database *ent;
419 prepare_alt_odb();
420 for (ent = alt_odb_list; ent; ent = ent->next)
421 if (fn(ent, cb))
422 return;
425 void prepare_alt_odb(void)
427 const char *alt;
429 if (alt_odb_tail)
430 return;
432 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
433 if (!alt) alt = "";
435 alt_odb_tail = &alt_odb_list;
436 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
438 read_info_alternates(get_object_directory(), 0);
441 static int has_loose_object_local(const unsigned char *sha1)
443 return !access(sha1_file_name(sha1), F_OK);
446 int has_loose_object_nonlocal(const unsigned char *sha1)
448 struct alternate_object_database *alt;
449 prepare_alt_odb();
450 for (alt = alt_odb_list; alt; alt = alt->next) {
451 fill_sha1_path(alt->name, sha1);
452 if (!access(alt->base, F_OK))
453 return 1;
455 return 0;
458 static int has_loose_object(const unsigned char *sha1)
460 return has_loose_object_local(sha1) ||
461 has_loose_object_nonlocal(sha1);
464 static unsigned int pack_used_ctr;
465 static unsigned int pack_mmap_calls;
466 static unsigned int peak_pack_open_windows;
467 static unsigned int pack_open_windows;
468 static unsigned int pack_open_fds;
469 static unsigned int pack_max_fds;
470 static size_t peak_pack_mapped;
471 static size_t pack_mapped;
472 struct packed_git *packed_git;
474 void pack_report(void)
476 fprintf(stderr,
477 "pack_report: getpagesize() = %10" SZ_FMT "\n"
478 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
479 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
480 sz_fmt(getpagesize()),
481 sz_fmt(packed_git_window_size),
482 sz_fmt(packed_git_limit));
483 fprintf(stderr,
484 "pack_report: pack_used_ctr = %10u\n"
485 "pack_report: pack_mmap_calls = %10u\n"
486 "pack_report: pack_open_windows = %10u / %10u\n"
487 "pack_report: pack_mapped = "
488 "%10" SZ_FMT " / %10" SZ_FMT "\n",
489 pack_used_ctr,
490 pack_mmap_calls,
491 pack_open_windows, peak_pack_open_windows,
492 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
496 * Open and mmap the index file at path, perform a couple of
497 * consistency checks, then record its information to p. Return 0 on
498 * success.
500 static int check_packed_git_idx(const char *path, struct packed_git *p)
502 void *idx_map;
503 struct pack_idx_header *hdr;
504 size_t idx_size;
505 uint32_t version, nr, i, *index;
506 int fd = git_open_noatime(path);
507 struct stat st;
509 if (fd < 0)
510 return -1;
511 if (fstat(fd, &st)) {
512 close(fd);
513 return -1;
515 idx_size = xsize_t(st.st_size);
516 if (idx_size < 4 * 256 + 20 + 20) {
517 close(fd);
518 return error("index file %s is too small", path);
520 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
521 close(fd);
523 hdr = idx_map;
524 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
525 version = ntohl(hdr->idx_version);
526 if (version < 2 || version > 2) {
527 munmap(idx_map, idx_size);
528 return error("index file %s is version %"PRIu32
529 " and is not supported by this binary"
530 " (try upgrading GIT to a newer version)",
531 path, version);
533 } else
534 version = 1;
536 nr = 0;
537 index = idx_map;
538 if (version > 1)
539 index += 2; /* skip index header */
540 for (i = 0; i < 256; i++) {
541 uint32_t n = ntohl(index[i]);
542 if (n < nr) {
543 munmap(idx_map, idx_size);
544 return error("non-monotonic index %s", path);
546 nr = n;
549 if (version == 1) {
551 * Total size:
552 * - 256 index entries 4 bytes each
553 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
554 * - 20-byte SHA1 of the packfile
555 * - 20-byte SHA1 file checksum
557 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
558 munmap(idx_map, idx_size);
559 return error("wrong index v1 file size in %s", path);
561 } else if (version == 2) {
563 * Minimum size:
564 * - 8 bytes of header
565 * - 256 index entries 4 bytes each
566 * - 20-byte sha1 entry * nr
567 * - 4-byte crc entry * nr
568 * - 4-byte offset entry * nr
569 * - 20-byte SHA1 of the packfile
570 * - 20-byte SHA1 file checksum
571 * And after the 4-byte offset table might be a
572 * variable sized table containing 8-byte entries
573 * for offsets larger than 2^31.
575 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
576 unsigned long max_size = min_size;
577 if (nr)
578 max_size += (nr - 1)*8;
579 if (idx_size < min_size || idx_size > max_size) {
580 munmap(idx_map, idx_size);
581 return error("wrong index v2 file size in %s", path);
583 if (idx_size != min_size &&
585 * make sure we can deal with large pack offsets.
586 * 31-bit signed offset won't be enough, neither
587 * 32-bit unsigned one will be.
589 (sizeof(off_t) <= 4)) {
590 munmap(idx_map, idx_size);
591 return error("pack too large for current definition of off_t in %s", path);
595 p->index_version = version;
596 p->index_data = idx_map;
597 p->index_size = idx_size;
598 p->num_objects = nr;
599 return 0;
602 int open_pack_index(struct packed_git *p)
604 char *idx_name;
605 int ret;
607 if (p->index_data)
608 return 0;
610 idx_name = xstrdup(p->pack_name);
611 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
612 ret = check_packed_git_idx(idx_name, p);
613 free(idx_name);
614 return ret;
617 static void scan_windows(struct packed_git *p,
618 struct packed_git **lru_p,
619 struct pack_window **lru_w,
620 struct pack_window **lru_l)
622 struct pack_window *w, *w_l;
624 for (w_l = NULL, w = p->windows; w; w = w->next) {
625 if (!w->inuse_cnt) {
626 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
627 *lru_p = p;
628 *lru_w = w;
629 *lru_l = w_l;
632 w_l = w;
636 static int unuse_one_window(struct packed_git *current)
638 struct packed_git *p, *lru_p = NULL;
639 struct pack_window *lru_w = NULL, *lru_l = NULL;
641 if (current)
642 scan_windows(current, &lru_p, &lru_w, &lru_l);
643 for (p = packed_git; p; p = p->next)
644 scan_windows(p, &lru_p, &lru_w, &lru_l);
645 if (lru_p) {
646 munmap(lru_w->base, lru_w->len);
647 pack_mapped -= lru_w->len;
648 if (lru_l)
649 lru_l->next = lru_w->next;
650 else
651 lru_p->windows = lru_w->next;
652 free(lru_w);
653 pack_open_windows--;
654 return 1;
656 return 0;
659 void release_pack_memory(size_t need)
661 size_t cur = pack_mapped;
662 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
663 ; /* nothing */
666 void *xmmap(void *start, size_t length,
667 int prot, int flags, int fd, off_t offset)
669 void *ret = mmap(start, length, prot, flags, fd, offset);
670 if (ret == MAP_FAILED) {
671 if (!length)
672 return NULL;
673 release_pack_memory(length);
674 ret = mmap(start, length, prot, flags, fd, offset);
675 if (ret == MAP_FAILED)
676 die_errno("Out of memory? mmap failed");
678 return ret;
681 void close_pack_windows(struct packed_git *p)
683 while (p->windows) {
684 struct pack_window *w = p->windows;
686 if (w->inuse_cnt)
687 die("pack '%s' still has open windows to it",
688 p->pack_name);
689 munmap(w->base, w->len);
690 pack_mapped -= w->len;
691 pack_open_windows--;
692 p->windows = w->next;
693 free(w);
698 * The LRU pack is the one with the oldest MRU window, preferring packs
699 * with no used windows, or the oldest mtime if it has no windows allocated.
701 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
703 struct pack_window *w, *this_mru_w;
704 int has_windows_inuse = 0;
707 * Reject this pack if it has windows and the previously selected
708 * one does not. If this pack does not have windows, reject
709 * it if the pack file is newer than the previously selected one.
711 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
712 return;
714 for (w = this_mru_w = p->windows; w; w = w->next) {
716 * Reject this pack if any of its windows are in use,
717 * but the previously selected pack did not have any
718 * inuse windows. Otherwise, record that this pack
719 * has windows in use.
721 if (w->inuse_cnt) {
722 if (*accept_windows_inuse)
723 has_windows_inuse = 1;
724 else
725 return;
728 if (w->last_used > this_mru_w->last_used)
729 this_mru_w = w;
732 * Reject this pack if it has windows that have been
733 * used more recently than the previously selected pack.
734 * If the previously selected pack had windows inuse and
735 * we have not encountered a window in this pack that is
736 * inuse, skip this check since we prefer a pack with no
737 * inuse windows to one that has inuse windows.
739 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
740 this_mru_w->last_used > (*mru_w)->last_used)
741 return;
745 * Select this pack.
747 *mru_w = this_mru_w;
748 *lru_p = p;
749 *accept_windows_inuse = has_windows_inuse;
752 static int close_one_pack(void)
754 struct packed_git *p, *lru_p = NULL;
755 struct pack_window *mru_w = NULL;
756 int accept_windows_inuse = 1;
758 for (p = packed_git; p; p = p->next) {
759 if (p->pack_fd == -1)
760 continue;
761 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
764 if (lru_p) {
765 close(lru_p->pack_fd);
766 pack_open_fds--;
767 lru_p->pack_fd = -1;
768 return 1;
771 return 0;
774 void unuse_pack(struct pack_window **w_cursor)
776 struct pack_window *w = *w_cursor;
777 if (w) {
778 w->inuse_cnt--;
779 *w_cursor = NULL;
783 void close_pack_index(struct packed_git *p)
785 if (p->index_data) {
786 munmap((void *)p->index_data, p->index_size);
787 p->index_data = NULL;
792 * This is used by git-repack in case a newly created pack happens to
793 * contain the same set of objects as an existing one. In that case
794 * the resulting file might be different even if its name would be the
795 * same. It is best to close any reference to the old pack before it is
796 * replaced on disk. Of course no index pointers or windows for given pack
797 * must subsist at this point. If ever objects from this pack are requested
798 * again, the new version of the pack will be reinitialized through
799 * reprepare_packed_git().
801 void free_pack_by_name(const char *pack_name)
803 struct packed_git *p, **pp = &packed_git;
805 while (*pp) {
806 p = *pp;
807 if (strcmp(pack_name, p->pack_name) == 0) {
808 clear_delta_base_cache();
809 close_pack_windows(p);
810 if (p->pack_fd != -1) {
811 close(p->pack_fd);
812 pack_open_fds--;
814 close_pack_index(p);
815 free(p->bad_object_sha1);
816 *pp = p->next;
817 if (last_found_pack == p)
818 last_found_pack = NULL;
819 free(p);
820 return;
822 pp = &p->next;
826 static unsigned int get_max_fd_limit(void)
828 #ifdef RLIMIT_NOFILE
830 struct rlimit lim;
832 if (!getrlimit(RLIMIT_NOFILE, &lim))
833 return lim.rlim_cur;
835 #endif
837 #ifdef _SC_OPEN_MAX
839 long open_max = sysconf(_SC_OPEN_MAX);
840 if (0 < open_max)
841 return open_max;
843 * Otherwise, we got -1 for one of the two
844 * reasons:
846 * (1) sysconf() did not understand _SC_OPEN_MAX
847 * and signaled an error with -1; or
848 * (2) sysconf() said there is no limit.
850 * We _could_ clear errno before calling sysconf() to
851 * tell these two cases apart and return a huge number
852 * in the latter case to let the caller cap it to a
853 * value that is not so selfish, but letting the
854 * fallback OPEN_MAX codepath take care of these cases
855 * is a lot simpler.
858 #endif
860 #ifdef OPEN_MAX
861 return OPEN_MAX;
862 #else
863 return 1; /* see the caller ;-) */
864 #endif
868 * Do not call this directly as this leaks p->pack_fd on error return;
869 * call open_packed_git() instead.
871 static int open_packed_git_1(struct packed_git *p)
873 struct stat st;
874 struct pack_header hdr;
875 unsigned char sha1[20];
876 unsigned char *idx_sha1;
877 long fd_flag;
879 if (!p->index_data && open_pack_index(p))
880 return error("packfile %s index unavailable", p->pack_name);
882 if (!pack_max_fds) {
883 unsigned int max_fds = get_max_fd_limit();
885 /* Save 3 for stdin/stdout/stderr, 22 for work */
886 if (25 < max_fds)
887 pack_max_fds = max_fds - 25;
888 else
889 pack_max_fds = 1;
892 while (pack_max_fds <= pack_open_fds && close_one_pack())
893 ; /* nothing */
895 p->pack_fd = git_open_noatime(p->pack_name);
896 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
897 return -1;
898 pack_open_fds++;
900 /* If we created the struct before we had the pack we lack size. */
901 if (!p->pack_size) {
902 if (!S_ISREG(st.st_mode))
903 return error("packfile %s not a regular file", p->pack_name);
904 p->pack_size = st.st_size;
905 } else if (p->pack_size != st.st_size)
906 return error("packfile %s size changed", p->pack_name);
908 /* We leave these file descriptors open with sliding mmap;
909 * there is no point keeping them open across exec(), though.
911 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
912 if (fd_flag < 0)
913 return error("cannot determine file descriptor flags");
914 fd_flag |= FD_CLOEXEC;
915 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
916 return error("cannot set FD_CLOEXEC");
918 /* Verify we recognize this pack file format. */
919 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
920 return error("file %s is far too short to be a packfile", p->pack_name);
921 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
922 return error("file %s is not a GIT packfile", p->pack_name);
923 if (!pack_version_ok(hdr.hdr_version))
924 return error("packfile %s is version %"PRIu32" and not"
925 " supported (try upgrading GIT to a newer version)",
926 p->pack_name, ntohl(hdr.hdr_version));
928 /* Verify the pack matches its index. */
929 if (p->num_objects != ntohl(hdr.hdr_entries))
930 return error("packfile %s claims to have %"PRIu32" objects"
931 " while index indicates %"PRIu32" objects",
932 p->pack_name, ntohl(hdr.hdr_entries),
933 p->num_objects);
934 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
935 return error("end of packfile %s is unavailable", p->pack_name);
936 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
937 return error("packfile %s signature is unavailable", p->pack_name);
938 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
939 if (hashcmp(sha1, idx_sha1))
940 return error("packfile %s does not match index", p->pack_name);
941 return 0;
944 static int open_packed_git(struct packed_git *p)
946 if (!open_packed_git_1(p))
947 return 0;
948 if (p->pack_fd != -1) {
949 close(p->pack_fd);
950 pack_open_fds--;
951 p->pack_fd = -1;
953 return -1;
956 static int in_window(struct pack_window *win, off_t offset)
958 /* We must promise at least 20 bytes (one hash) after the
959 * offset is available from this window, otherwise the offset
960 * is not actually in this window and a different window (which
961 * has that one hash excess) must be used. This is to support
962 * the object header and delta base parsing routines below.
964 off_t win_off = win->offset;
965 return win_off <= offset
966 && (offset + 20) <= (win_off + win->len);
969 unsigned char *use_pack(struct packed_git *p,
970 struct pack_window **w_cursor,
971 off_t offset,
972 unsigned long *left)
974 struct pack_window *win = *w_cursor;
976 /* Since packfiles end in a hash of their content and it's
977 * pointless to ask for an offset into the middle of that
978 * hash, and the in_window function above wouldn't match
979 * don't allow an offset too close to the end of the file.
981 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
982 die("packfile %s cannot be accessed", p->pack_name);
983 if (offset > (p->pack_size - 20))
984 die("offset beyond end of packfile (truncated pack?)");
986 if (!win || !in_window(win, offset)) {
987 if (win)
988 win->inuse_cnt--;
989 for (win = p->windows; win; win = win->next) {
990 if (in_window(win, offset))
991 break;
993 if (!win) {
994 size_t window_align = packed_git_window_size / 2;
995 off_t len;
997 if (p->pack_fd == -1 && open_packed_git(p))
998 die("packfile %s cannot be accessed", p->pack_name);
1000 win = xcalloc(1, sizeof(*win));
1001 win->offset = (offset / window_align) * window_align;
1002 len = p->pack_size - win->offset;
1003 if (len > packed_git_window_size)
1004 len = packed_git_window_size;
1005 win->len = (size_t)len;
1006 pack_mapped += win->len;
1007 while (packed_git_limit < pack_mapped
1008 && unuse_one_window(p))
1009 ; /* nothing */
1010 win->base = xmmap(NULL, win->len,
1011 PROT_READ, MAP_PRIVATE,
1012 p->pack_fd, win->offset);
1013 if (win->base == MAP_FAILED)
1014 die("packfile %s cannot be mapped: %s",
1015 p->pack_name,
1016 strerror(errno));
1017 if (!win->offset && win->len == p->pack_size
1018 && !p->do_not_close) {
1019 close(p->pack_fd);
1020 pack_open_fds--;
1021 p->pack_fd = -1;
1023 pack_mmap_calls++;
1024 pack_open_windows++;
1025 if (pack_mapped > peak_pack_mapped)
1026 peak_pack_mapped = pack_mapped;
1027 if (pack_open_windows > peak_pack_open_windows)
1028 peak_pack_open_windows = pack_open_windows;
1029 win->next = p->windows;
1030 p->windows = win;
1033 if (win != *w_cursor) {
1034 win->last_used = pack_used_ctr++;
1035 win->inuse_cnt++;
1036 *w_cursor = win;
1038 offset -= win->offset;
1039 if (left)
1040 *left = win->len - xsize_t(offset);
1041 return win->base + offset;
1044 static struct packed_git *alloc_packed_git(int extra)
1046 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1047 memset(p, 0, sizeof(*p));
1048 p->pack_fd = -1;
1049 return p;
1052 static void try_to_free_pack_memory(size_t size)
1054 release_pack_memory(size);
1057 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1059 static int have_set_try_to_free_routine;
1060 struct stat st;
1061 struct packed_git *p = alloc_packed_git(path_len + 2);
1063 if (!have_set_try_to_free_routine) {
1064 have_set_try_to_free_routine = 1;
1065 set_try_to_free_routine(try_to_free_pack_memory);
1069 * Make sure a corresponding .pack file exists and that
1070 * the index looks sane.
1072 path_len -= strlen(".idx");
1073 if (path_len < 1) {
1074 free(p);
1075 return NULL;
1077 memcpy(p->pack_name, path, path_len);
1079 strcpy(p->pack_name + path_len, ".keep");
1080 if (!access(p->pack_name, F_OK))
1081 p->pack_keep = 1;
1083 strcpy(p->pack_name + path_len, ".pack");
1084 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1085 free(p);
1086 return NULL;
1089 /* ok, it looks sane as far as we can check without
1090 * actually mapping the pack file.
1092 p->pack_size = st.st_size;
1093 p->pack_local = local;
1094 p->mtime = st.st_mtime;
1095 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1096 hashclr(p->sha1);
1097 return p;
1100 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1102 const char *path = sha1_pack_name(sha1);
1103 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1105 strcpy(p->pack_name, path);
1106 hashcpy(p->sha1, sha1);
1107 if (check_packed_git_idx(idx_path, p)) {
1108 free(p);
1109 return NULL;
1112 return p;
1115 void install_packed_git(struct packed_git *pack)
1117 if (pack->pack_fd != -1)
1118 pack_open_fds++;
1120 pack->next = packed_git;
1121 packed_git = pack;
1124 void (*report_garbage)(const char *desc, const char *path);
1126 static void report_helper(const struct string_list *list,
1127 int seen_bits, int first, int last)
1129 const char *msg;
1130 switch (seen_bits) {
1131 case 0:
1132 msg = "no corresponding .idx or .pack";
1133 break;
1134 case 1:
1135 msg = "no corresponding .idx";
1136 break;
1137 case 2:
1138 msg = "no corresponding .pack";
1139 break;
1140 default:
1141 return;
1143 for (; first < last; first++)
1144 report_garbage(msg, list->items[first].string);
1147 static void report_pack_garbage(struct string_list *list)
1149 int i, baselen = -1, first = 0, seen_bits = 0;
1151 if (!report_garbage)
1152 return;
1154 sort_string_list(list);
1156 for (i = 0; i < list->nr; i++) {
1157 const char *path = list->items[i].string;
1158 if (baselen != -1 &&
1159 strncmp(path, list->items[first].string, baselen)) {
1160 report_helper(list, seen_bits, first, i);
1161 baselen = -1;
1162 seen_bits = 0;
1164 if (baselen == -1) {
1165 const char *dot = strrchr(path, '.');
1166 if (!dot) {
1167 report_garbage("garbage found", path);
1168 continue;
1170 baselen = dot - path + 1;
1171 first = i;
1173 if (!strcmp(path + baselen, "pack"))
1174 seen_bits |= 1;
1175 else if (!strcmp(path + baselen, "idx"))
1176 seen_bits |= 2;
1178 report_helper(list, seen_bits, first, list->nr);
1181 static void prepare_packed_git_one(char *objdir, int local)
1183 struct strbuf path = STRBUF_INIT;
1184 size_t dirnamelen;
1185 DIR *dir;
1186 struct dirent *de;
1187 struct string_list garbage = STRING_LIST_INIT_DUP;
1189 strbuf_addstr(&path, objdir);
1190 strbuf_addstr(&path, "/pack");
1191 dir = opendir(path.buf);
1192 if (!dir) {
1193 if (errno != ENOENT)
1194 error("unable to open object pack directory: %s: %s",
1195 path.buf, strerror(errno));
1196 strbuf_release(&path);
1197 return;
1199 strbuf_addch(&path, '/');
1200 dirnamelen = path.len;
1201 while ((de = readdir(dir)) != NULL) {
1202 struct packed_git *p;
1203 size_t base_len;
1205 if (is_dot_or_dotdot(de->d_name))
1206 continue;
1208 strbuf_setlen(&path, dirnamelen);
1209 strbuf_addstr(&path, de->d_name);
1211 base_len = path.len;
1212 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1213 /* Don't reopen a pack we already have. */
1214 for (p = packed_git; p; p = p->next) {
1215 size_t len;
1216 if (strip_suffix(p->pack_name, ".pack", &len) &&
1217 len == base_len &&
1218 !memcmp(p->pack_name, path.buf, len))
1219 break;
1221 if (p == NULL &&
1223 * See if it really is a valid .idx file with
1224 * corresponding .pack file that we can map.
1226 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1227 install_packed_git(p);
1230 if (!report_garbage)
1231 continue;
1233 if (ends_with(de->d_name, ".idx") ||
1234 ends_with(de->d_name, ".pack") ||
1235 ends_with(de->d_name, ".bitmap") ||
1236 ends_with(de->d_name, ".keep"))
1237 string_list_append(&garbage, path.buf);
1238 else
1239 report_garbage("garbage found", path.buf);
1241 closedir(dir);
1242 report_pack_garbage(&garbage);
1243 string_list_clear(&garbage, 0);
1244 strbuf_release(&path);
1247 static int sort_pack(const void *a_, const void *b_)
1249 struct packed_git *a = *((struct packed_git **)a_);
1250 struct packed_git *b = *((struct packed_git **)b_);
1251 int st;
1254 * Local packs tend to contain objects specific to our
1255 * variant of the project than remote ones. In addition,
1256 * remote ones could be on a network mounted filesystem.
1257 * Favor local ones for these reasons.
1259 st = a->pack_local - b->pack_local;
1260 if (st)
1261 return -st;
1264 * Younger packs tend to contain more recent objects,
1265 * and more recent objects tend to get accessed more
1266 * often.
1268 if (a->mtime < b->mtime)
1269 return 1;
1270 else if (a->mtime == b->mtime)
1271 return 0;
1272 return -1;
1275 static void rearrange_packed_git(void)
1277 struct packed_git **ary, *p;
1278 int i, n;
1280 for (n = 0, p = packed_git; p; p = p->next)
1281 n++;
1282 if (n < 2)
1283 return;
1285 /* prepare an array of packed_git for easier sorting */
1286 ary = xcalloc(n, sizeof(struct packed_git *));
1287 for (n = 0, p = packed_git; p; p = p->next)
1288 ary[n++] = p;
1290 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1292 /* link them back again */
1293 for (i = 0; i < n - 1; i++)
1294 ary[i]->next = ary[i + 1];
1295 ary[n - 1]->next = NULL;
1296 packed_git = ary[0];
1298 free(ary);
1301 static int prepare_packed_git_run_once = 0;
1302 void prepare_packed_git(void)
1304 struct alternate_object_database *alt;
1306 if (prepare_packed_git_run_once)
1307 return;
1308 prepare_packed_git_one(get_object_directory(), 1);
1309 prepare_alt_odb();
1310 for (alt = alt_odb_list; alt; alt = alt->next) {
1311 alt->name[-1] = 0;
1312 prepare_packed_git_one(alt->base, 0);
1313 alt->name[-1] = '/';
1315 rearrange_packed_git();
1316 prepare_packed_git_run_once = 1;
1319 void reprepare_packed_git(void)
1321 prepare_packed_git_run_once = 0;
1322 prepare_packed_git();
1325 static void mark_bad_packed_object(struct packed_git *p,
1326 const unsigned char *sha1)
1328 unsigned i;
1329 for (i = 0; i < p->num_bad_objects; i++)
1330 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1331 return;
1332 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1333 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1334 p->num_bad_objects++;
1337 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1339 struct packed_git *p;
1340 unsigned i;
1342 for (p = packed_git; p; p = p->next)
1343 for (i = 0; i < p->num_bad_objects; i++)
1344 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1345 return p;
1346 return NULL;
1350 * With an in-core object data in "map", rehash it to make sure the
1351 * object name actually matches "sha1" to detect object corruption.
1352 * With "map" == NULL, try reading the object named with "sha1" using
1353 * the streaming interface and rehash it to do the same.
1355 int check_sha1_signature(const unsigned char *sha1, void *map,
1356 unsigned long size, const char *type)
1358 unsigned char real_sha1[20];
1359 enum object_type obj_type;
1360 struct git_istream *st;
1361 git_SHA_CTX c;
1362 char hdr[32];
1363 int hdrlen;
1365 if (map) {
1366 hash_sha1_file(map, size, type, real_sha1);
1367 return hashcmp(sha1, real_sha1) ? -1 : 0;
1370 st = open_istream(sha1, &obj_type, &size, NULL);
1371 if (!st)
1372 return -1;
1374 /* Generate the header */
1375 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1377 /* Sha1.. */
1378 git_SHA1_Init(&c);
1379 git_SHA1_Update(&c, hdr, hdrlen);
1380 for (;;) {
1381 char buf[1024 * 16];
1382 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1384 if (readlen < 0) {
1385 close_istream(st);
1386 return -1;
1388 if (!readlen)
1389 break;
1390 git_SHA1_Update(&c, buf, readlen);
1392 git_SHA1_Final(real_sha1, &c);
1393 close_istream(st);
1394 return hashcmp(sha1, real_sha1) ? -1 : 0;
1397 int git_open_noatime(const char *name)
1399 static int sha1_file_open_flag = O_NOATIME;
1401 for (;;) {
1402 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1403 if (fd >= 0)
1404 return fd;
1406 /* Might the failure be due to O_NOATIME? */
1407 if (errno != ENOENT && sha1_file_open_flag) {
1408 sha1_file_open_flag = 0;
1409 continue;
1412 return -1;
1416 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1418 struct alternate_object_database *alt;
1420 if (!lstat(sha1_file_name(sha1), st))
1421 return 0;
1423 prepare_alt_odb();
1424 errno = ENOENT;
1425 for (alt = alt_odb_list; alt; alt = alt->next) {
1426 fill_sha1_path(alt->name, sha1);
1427 if (!lstat(alt->base, st))
1428 return 0;
1431 return -1;
1434 static int open_sha1_file(const unsigned char *sha1)
1436 int fd;
1437 struct alternate_object_database *alt;
1438 int most_interesting_errno;
1440 fd = git_open_noatime(sha1_file_name(sha1));
1441 if (fd >= 0)
1442 return fd;
1443 most_interesting_errno = errno;
1445 prepare_alt_odb();
1446 for (alt = alt_odb_list; alt; alt = alt->next) {
1447 fill_sha1_path(alt->name, sha1);
1448 fd = git_open_noatime(alt->base);
1449 if (fd >= 0)
1450 return fd;
1451 if (most_interesting_errno == ENOENT)
1452 most_interesting_errno = errno;
1454 errno = most_interesting_errno;
1455 return -1;
1458 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1460 void *map;
1461 int fd;
1463 fd = open_sha1_file(sha1);
1464 map = NULL;
1465 if (fd >= 0) {
1466 struct stat st;
1468 if (!fstat(fd, &st)) {
1469 *size = xsize_t(st.st_size);
1470 if (!*size) {
1471 /* mmap() is forbidden on empty files */
1472 error("object file %s is empty", sha1_file_name(sha1));
1473 return NULL;
1475 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1477 close(fd);
1479 return map;
1482 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1483 unsigned long len, enum object_type *type, unsigned long *sizep)
1485 unsigned shift;
1486 unsigned long size, c;
1487 unsigned long used = 0;
1489 c = buf[used++];
1490 *type = (c >> 4) & 7;
1491 size = c & 15;
1492 shift = 4;
1493 while (c & 0x80) {
1494 if (len <= used || bitsizeof(long) <= shift) {
1495 error("bad object header");
1496 size = used = 0;
1497 break;
1499 c = buf[used++];
1500 size += (c & 0x7f) << shift;
1501 shift += 7;
1503 *sizep = size;
1504 return used;
1507 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1509 /* Get the data stream */
1510 memset(stream, 0, sizeof(*stream));
1511 stream->next_in = map;
1512 stream->avail_in = mapsize;
1513 stream->next_out = buffer;
1514 stream->avail_out = bufsiz;
1516 git_inflate_init(stream);
1517 return git_inflate(stream, 0);
1520 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1522 int bytes = strlen(buffer) + 1;
1523 unsigned char *buf = xmallocz(size);
1524 unsigned long n;
1525 int status = Z_OK;
1527 n = stream->total_out - bytes;
1528 if (n > size)
1529 n = size;
1530 memcpy(buf, (char *) buffer + bytes, n);
1531 bytes = n;
1532 if (bytes <= size) {
1534 * The above condition must be (bytes <= size), not
1535 * (bytes < size). In other words, even though we
1536 * expect no more output and set avail_out to zero,
1537 * the input zlib stream may have bytes that express
1538 * "this concludes the stream", and we *do* want to
1539 * eat that input.
1541 * Otherwise we would not be able to test that we
1542 * consumed all the input to reach the expected size;
1543 * we also want to check that zlib tells us that all
1544 * went well with status == Z_STREAM_END at the end.
1546 stream->next_out = buf + bytes;
1547 stream->avail_out = size - bytes;
1548 while (status == Z_OK)
1549 status = git_inflate(stream, Z_FINISH);
1551 if (status == Z_STREAM_END && !stream->avail_in) {
1552 git_inflate_end(stream);
1553 return buf;
1556 if (status < 0)
1557 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1558 else if (stream->avail_in)
1559 error("garbage at end of loose object '%s'",
1560 sha1_to_hex(sha1));
1561 free(buf);
1562 return NULL;
1566 * We used to just use "sscanf()", but that's actually way
1567 * too permissive for what we want to check. So do an anal
1568 * object header parse by hand.
1570 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1572 char type[10];
1573 int i;
1574 unsigned long size;
1577 * The type can be at most ten bytes (including the
1578 * terminating '\0' that we add), and is followed by
1579 * a space.
1581 i = 0;
1582 for (;;) {
1583 char c = *hdr++;
1584 if (c == ' ')
1585 break;
1586 type[i++] = c;
1587 if (i >= sizeof(type))
1588 return -1;
1590 type[i] = 0;
1593 * The length must follow immediately, and be in canonical
1594 * decimal format (ie "010" is not valid).
1596 size = *hdr++ - '0';
1597 if (size > 9)
1598 return -1;
1599 if (size) {
1600 for (;;) {
1601 unsigned long c = *hdr - '0';
1602 if (c > 9)
1603 break;
1604 hdr++;
1605 size = size * 10 + c;
1608 *sizep = size;
1611 * The length must be followed by a zero byte
1613 return *hdr ? -1 : type_from_string(type);
1616 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1618 int ret;
1619 git_zstream stream;
1620 char hdr[8192];
1622 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1623 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1624 return NULL;
1626 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1629 unsigned long get_size_from_delta(struct packed_git *p,
1630 struct pack_window **w_curs,
1631 off_t curpos)
1633 const unsigned char *data;
1634 unsigned char delta_head[20], *in;
1635 git_zstream stream;
1636 int st;
1638 memset(&stream, 0, sizeof(stream));
1639 stream.next_out = delta_head;
1640 stream.avail_out = sizeof(delta_head);
1642 git_inflate_init(&stream);
1643 do {
1644 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1645 stream.next_in = in;
1646 st = git_inflate(&stream, Z_FINISH);
1647 curpos += stream.next_in - in;
1648 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1649 stream.total_out < sizeof(delta_head));
1650 git_inflate_end(&stream);
1651 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1652 error("delta data unpack-initial failed");
1653 return 0;
1656 /* Examine the initial part of the delta to figure out
1657 * the result size.
1659 data = delta_head;
1661 /* ignore base size */
1662 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1664 /* Read the result size */
1665 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1668 static off_t get_delta_base(struct packed_git *p,
1669 struct pack_window **w_curs,
1670 off_t *curpos,
1671 enum object_type type,
1672 off_t delta_obj_offset)
1674 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1675 off_t base_offset;
1677 /* use_pack() assured us we have [base_info, base_info + 20)
1678 * as a range that we can look at without walking off the
1679 * end of the mapped window. Its actually the hash size
1680 * that is assured. An OFS_DELTA longer than the hash size
1681 * is stupid, as then a REF_DELTA would be smaller to store.
1683 if (type == OBJ_OFS_DELTA) {
1684 unsigned used = 0;
1685 unsigned char c = base_info[used++];
1686 base_offset = c & 127;
1687 while (c & 128) {
1688 base_offset += 1;
1689 if (!base_offset || MSB(base_offset, 7))
1690 return 0; /* overflow */
1691 c = base_info[used++];
1692 base_offset = (base_offset << 7) + (c & 127);
1694 base_offset = delta_obj_offset - base_offset;
1695 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1696 return 0; /* out of bound */
1697 *curpos += used;
1698 } else if (type == OBJ_REF_DELTA) {
1699 /* The base entry _must_ be in the same pack */
1700 base_offset = find_pack_entry_one(base_info, p);
1701 *curpos += 20;
1702 } else
1703 die("I am totally screwed");
1704 return base_offset;
1708 * Like get_delta_base above, but we return the sha1 instead of the pack
1709 * offset. This means it is cheaper for REF deltas (we do not have to do
1710 * the final object lookup), but more expensive for OFS deltas (we
1711 * have to load the revidx to convert the offset back into a sha1).
1713 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1714 struct pack_window **w_curs,
1715 off_t curpos,
1716 enum object_type type,
1717 off_t delta_obj_offset)
1719 if (type == OBJ_REF_DELTA) {
1720 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1721 return base;
1722 } else if (type == OBJ_OFS_DELTA) {
1723 struct revindex_entry *revidx;
1724 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1725 type, delta_obj_offset);
1727 if (!base_offset)
1728 return NULL;
1730 revidx = find_pack_revindex(p, base_offset);
1731 if (!revidx)
1732 return NULL;
1734 return nth_packed_object_sha1(p, revidx->nr);
1735 } else
1736 return NULL;
1739 int unpack_object_header(struct packed_git *p,
1740 struct pack_window **w_curs,
1741 off_t *curpos,
1742 unsigned long *sizep)
1744 unsigned char *base;
1745 unsigned long left;
1746 unsigned long used;
1747 enum object_type type;
1749 /* use_pack() assures us we have [base, base + 20) available
1750 * as a range that we can look at. (Its actually the hash
1751 * size that is assured.) With our object header encoding
1752 * the maximum deflated object size is 2^137, which is just
1753 * insane, so we know won't exceed what we have been given.
1755 base = use_pack(p, w_curs, *curpos, &left);
1756 used = unpack_object_header_buffer(base, left, &type, sizep);
1757 if (!used) {
1758 type = OBJ_BAD;
1759 } else
1760 *curpos += used;
1762 return type;
1765 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1767 int type;
1768 struct revindex_entry *revidx;
1769 const unsigned char *sha1;
1770 revidx = find_pack_revindex(p, obj_offset);
1771 if (!revidx)
1772 return OBJ_BAD;
1773 sha1 = nth_packed_object_sha1(p, revidx->nr);
1774 mark_bad_packed_object(p, sha1);
1775 type = sha1_object_info(sha1, NULL);
1776 if (type <= OBJ_NONE)
1777 return OBJ_BAD;
1778 return type;
1781 #define POI_STACK_PREALLOC 64
1783 static enum object_type packed_to_object_type(struct packed_git *p,
1784 off_t obj_offset,
1785 enum object_type type,
1786 struct pack_window **w_curs,
1787 off_t curpos)
1789 off_t small_poi_stack[POI_STACK_PREALLOC];
1790 off_t *poi_stack = small_poi_stack;
1791 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1793 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1794 off_t base_offset;
1795 unsigned long size;
1796 /* Push the object we're going to leave behind */
1797 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1798 poi_stack_alloc = alloc_nr(poi_stack_nr);
1799 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1800 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1801 } else {
1802 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1804 poi_stack[poi_stack_nr++] = obj_offset;
1805 /* If parsing the base offset fails, just unwind */
1806 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1807 if (!base_offset)
1808 goto unwind;
1809 curpos = obj_offset = base_offset;
1810 type = unpack_object_header(p, w_curs, &curpos, &size);
1811 if (type <= OBJ_NONE) {
1812 /* If getting the base itself fails, we first
1813 * retry the base, otherwise unwind */
1814 type = retry_bad_packed_offset(p, base_offset);
1815 if (type > OBJ_NONE)
1816 goto out;
1817 goto unwind;
1821 switch (type) {
1822 case OBJ_BAD:
1823 case OBJ_COMMIT:
1824 case OBJ_TREE:
1825 case OBJ_BLOB:
1826 case OBJ_TAG:
1827 break;
1828 default:
1829 error("unknown object type %i at offset %"PRIuMAX" in %s",
1830 type, (uintmax_t)obj_offset, p->pack_name);
1831 type = OBJ_BAD;
1834 out:
1835 if (poi_stack != small_poi_stack)
1836 free(poi_stack);
1837 return type;
1839 unwind:
1840 while (poi_stack_nr) {
1841 obj_offset = poi_stack[--poi_stack_nr];
1842 type = retry_bad_packed_offset(p, obj_offset);
1843 if (type > OBJ_NONE)
1844 goto out;
1846 type = OBJ_BAD;
1847 goto out;
1850 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1851 struct object_info *oi)
1853 struct pack_window *w_curs = NULL;
1854 unsigned long size;
1855 off_t curpos = obj_offset;
1856 enum object_type type;
1859 * We always get the representation type, but only convert it to
1860 * a "real" type later if the caller is interested.
1862 type = unpack_object_header(p, &w_curs, &curpos, &size);
1864 if (oi->sizep) {
1865 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1866 off_t tmp_pos = curpos;
1867 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1868 type, obj_offset);
1869 if (!base_offset) {
1870 type = OBJ_BAD;
1871 goto out;
1873 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1874 if (*oi->sizep == 0) {
1875 type = OBJ_BAD;
1876 goto out;
1878 } else {
1879 *oi->sizep = size;
1883 if (oi->disk_sizep) {
1884 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1885 *oi->disk_sizep = revidx[1].offset - obj_offset;
1888 if (oi->typep) {
1889 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1890 if (*oi->typep < 0) {
1891 type = OBJ_BAD;
1892 goto out;
1896 if (oi->delta_base_sha1) {
1897 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1898 const unsigned char *base;
1900 base = get_delta_base_sha1(p, &w_curs, curpos,
1901 type, obj_offset);
1902 if (!base) {
1903 type = OBJ_BAD;
1904 goto out;
1907 hashcpy(oi->delta_base_sha1, base);
1908 } else
1909 hashclr(oi->delta_base_sha1);
1912 out:
1913 unuse_pack(&w_curs);
1914 return type;
1917 static void *unpack_compressed_entry(struct packed_git *p,
1918 struct pack_window **w_curs,
1919 off_t curpos,
1920 unsigned long size)
1922 int st;
1923 git_zstream stream;
1924 unsigned char *buffer, *in;
1926 buffer = xmallocz(size);
1927 memset(&stream, 0, sizeof(stream));
1928 stream.next_out = buffer;
1929 stream.avail_out = size + 1;
1931 git_inflate_init(&stream);
1932 do {
1933 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1934 stream.next_in = in;
1935 st = git_inflate(&stream, Z_FINISH);
1936 if (!stream.avail_out)
1937 break; /* the payload is larger than it should be */
1938 curpos += stream.next_in - in;
1939 } while (st == Z_OK || st == Z_BUF_ERROR);
1940 git_inflate_end(&stream);
1941 if ((st != Z_STREAM_END) || stream.total_out != size) {
1942 free(buffer);
1943 return NULL;
1946 return buffer;
1949 #define MAX_DELTA_CACHE (256)
1951 static size_t delta_base_cached;
1953 static struct delta_base_cache_lru_list {
1954 struct delta_base_cache_lru_list *prev;
1955 struct delta_base_cache_lru_list *next;
1956 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1958 static struct delta_base_cache_entry {
1959 struct delta_base_cache_lru_list lru;
1960 void *data;
1961 struct packed_git *p;
1962 off_t base_offset;
1963 unsigned long size;
1964 enum object_type type;
1965 } delta_base_cache[MAX_DELTA_CACHE];
1967 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1969 unsigned long hash;
1971 hash = (unsigned long)p + (unsigned long)base_offset;
1972 hash += (hash >> 8) + (hash >> 16);
1973 return hash % MAX_DELTA_CACHE;
1976 static struct delta_base_cache_entry *
1977 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1979 unsigned long hash = pack_entry_hash(p, base_offset);
1980 return delta_base_cache + hash;
1983 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1984 struct packed_git *p, off_t base_offset)
1986 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1989 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1991 struct delta_base_cache_entry *ent;
1992 ent = get_delta_base_cache_entry(p, base_offset);
1993 return eq_delta_base_cache_entry(ent, p, base_offset);
1996 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1998 ent->data = NULL;
1999 ent->lru.next->prev = ent->lru.prev;
2000 ent->lru.prev->next = ent->lru.next;
2001 delta_base_cached -= ent->size;
2004 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2005 unsigned long *base_size, enum object_type *type, int keep_cache)
2007 struct delta_base_cache_entry *ent;
2008 void *ret;
2010 ent = get_delta_base_cache_entry(p, base_offset);
2012 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2013 return unpack_entry(p, base_offset, type, base_size);
2015 ret = ent->data;
2017 if (!keep_cache)
2018 clear_delta_base_cache_entry(ent);
2019 else
2020 ret = xmemdupz(ent->data, ent->size);
2021 *type = ent->type;
2022 *base_size = ent->size;
2023 return ret;
2026 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2028 if (ent->data) {
2029 free(ent->data);
2030 ent->data = NULL;
2031 ent->lru.next->prev = ent->lru.prev;
2032 ent->lru.prev->next = ent->lru.next;
2033 delta_base_cached -= ent->size;
2037 void clear_delta_base_cache(void)
2039 unsigned long p;
2040 for (p = 0; p < MAX_DELTA_CACHE; p++)
2041 release_delta_base_cache(&delta_base_cache[p]);
2044 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2045 void *base, unsigned long base_size, enum object_type type)
2047 unsigned long hash = pack_entry_hash(p, base_offset);
2048 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2049 struct delta_base_cache_lru_list *lru;
2051 release_delta_base_cache(ent);
2052 delta_base_cached += base_size;
2054 for (lru = delta_base_cache_lru.next;
2055 delta_base_cached > delta_base_cache_limit
2056 && lru != &delta_base_cache_lru;
2057 lru = lru->next) {
2058 struct delta_base_cache_entry *f = (void *)lru;
2059 if (f->type == OBJ_BLOB)
2060 release_delta_base_cache(f);
2062 for (lru = delta_base_cache_lru.next;
2063 delta_base_cached > delta_base_cache_limit
2064 && lru != &delta_base_cache_lru;
2065 lru = lru->next) {
2066 struct delta_base_cache_entry *f = (void *)lru;
2067 release_delta_base_cache(f);
2070 ent->p = p;
2071 ent->base_offset = base_offset;
2072 ent->type = type;
2073 ent->data = base;
2074 ent->size = base_size;
2075 ent->lru.next = &delta_base_cache_lru;
2076 ent->lru.prev = delta_base_cache_lru.prev;
2077 delta_base_cache_lru.prev->next = &ent->lru;
2078 delta_base_cache_lru.prev = &ent->lru;
2081 static void *read_object(const unsigned char *sha1, enum object_type *type,
2082 unsigned long *size);
2084 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2086 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2087 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2088 p->pack_name, (uintmax_t)obj_offset);
2091 int do_check_packed_object_crc;
2093 #define UNPACK_ENTRY_STACK_PREALLOC 64
2094 struct unpack_entry_stack_ent {
2095 off_t obj_offset;
2096 off_t curpos;
2097 unsigned long size;
2100 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2101 enum object_type *final_type, unsigned long *final_size)
2103 struct pack_window *w_curs = NULL;
2104 off_t curpos = obj_offset;
2105 void *data = NULL;
2106 unsigned long size;
2107 enum object_type type;
2108 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2109 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2110 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2111 int base_from_cache = 0;
2113 write_pack_access_log(p, obj_offset);
2115 /* PHASE 1: drill down to the innermost base object */
2116 for (;;) {
2117 off_t base_offset;
2118 int i;
2119 struct delta_base_cache_entry *ent;
2121 ent = get_delta_base_cache_entry(p, curpos);
2122 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2123 type = ent->type;
2124 data = ent->data;
2125 size = ent->size;
2126 clear_delta_base_cache_entry(ent);
2127 base_from_cache = 1;
2128 break;
2131 if (do_check_packed_object_crc && p->index_version > 1) {
2132 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2133 unsigned long len = revidx[1].offset - obj_offset;
2134 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2135 const unsigned char *sha1 =
2136 nth_packed_object_sha1(p, revidx->nr);
2137 error("bad packed object CRC for %s",
2138 sha1_to_hex(sha1));
2139 mark_bad_packed_object(p, sha1);
2140 unuse_pack(&w_curs);
2141 return NULL;
2145 type = unpack_object_header(p, &w_curs, &curpos, &size);
2146 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2147 break;
2149 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2150 if (!base_offset) {
2151 error("failed to validate delta base reference "
2152 "at offset %"PRIuMAX" from %s",
2153 (uintmax_t)curpos, p->pack_name);
2154 /* bail to phase 2, in hopes of recovery */
2155 data = NULL;
2156 break;
2159 /* push object, proceed to base */
2160 if (delta_stack_nr >= delta_stack_alloc
2161 && delta_stack == small_delta_stack) {
2162 delta_stack_alloc = alloc_nr(delta_stack_nr);
2163 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2164 memcpy(delta_stack, small_delta_stack,
2165 sizeof(*delta_stack)*delta_stack_nr);
2166 } else {
2167 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2169 i = delta_stack_nr++;
2170 delta_stack[i].obj_offset = obj_offset;
2171 delta_stack[i].curpos = curpos;
2172 delta_stack[i].size = size;
2174 curpos = obj_offset = base_offset;
2177 /* PHASE 2: handle the base */
2178 switch (type) {
2179 case OBJ_OFS_DELTA:
2180 case OBJ_REF_DELTA:
2181 if (data)
2182 die("BUG in unpack_entry: left loop at a valid delta");
2183 break;
2184 case OBJ_COMMIT:
2185 case OBJ_TREE:
2186 case OBJ_BLOB:
2187 case OBJ_TAG:
2188 if (!base_from_cache)
2189 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2190 break;
2191 default:
2192 data = NULL;
2193 error("unknown object type %i at offset %"PRIuMAX" in %s",
2194 type, (uintmax_t)obj_offset, p->pack_name);
2197 /* PHASE 3: apply deltas in order */
2199 /* invariants:
2200 * 'data' holds the base data, or NULL if there was corruption
2202 while (delta_stack_nr) {
2203 void *delta_data;
2204 void *base = data;
2205 unsigned long delta_size, base_size = size;
2206 int i;
2208 data = NULL;
2210 if (base)
2211 add_delta_base_cache(p, obj_offset, base, base_size, type);
2213 if (!base) {
2215 * We're probably in deep shit, but let's try to fetch
2216 * the required base anyway from another pack or loose.
2217 * This is costly but should happen only in the presence
2218 * of a corrupted pack, and is better than failing outright.
2220 struct revindex_entry *revidx;
2221 const unsigned char *base_sha1;
2222 revidx = find_pack_revindex(p, obj_offset);
2223 if (revidx) {
2224 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2225 error("failed to read delta base object %s"
2226 " at offset %"PRIuMAX" from %s",
2227 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2228 p->pack_name);
2229 mark_bad_packed_object(p, base_sha1);
2230 base = read_object(base_sha1, &type, &base_size);
2234 i = --delta_stack_nr;
2235 obj_offset = delta_stack[i].obj_offset;
2236 curpos = delta_stack[i].curpos;
2237 delta_size = delta_stack[i].size;
2239 if (!base)
2240 continue;
2242 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2244 if (!delta_data) {
2245 error("failed to unpack compressed delta "
2246 "at offset %"PRIuMAX" from %s",
2247 (uintmax_t)curpos, p->pack_name);
2248 data = NULL;
2249 continue;
2252 data = patch_delta(base, base_size,
2253 delta_data, delta_size,
2254 &size);
2257 * We could not apply the delta; warn the user, but keep going.
2258 * Our failure will be noticed either in the next iteration of
2259 * the loop, or if this is the final delta, in the caller when
2260 * we return NULL. Those code paths will take care of making
2261 * a more explicit warning and retrying with another copy of
2262 * the object.
2264 if (!data)
2265 error("failed to apply delta");
2267 free(delta_data);
2270 *final_type = type;
2271 *final_size = size;
2273 unuse_pack(&w_curs);
2275 if (delta_stack != small_delta_stack)
2276 free(delta_stack);
2278 return data;
2281 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2282 uint32_t n)
2284 const unsigned char *index = p->index_data;
2285 if (!index) {
2286 if (open_pack_index(p))
2287 return NULL;
2288 index = p->index_data;
2290 if (n >= p->num_objects)
2291 return NULL;
2292 index += 4 * 256;
2293 if (p->index_version == 1) {
2294 return index + 24 * n + 4;
2295 } else {
2296 index += 8;
2297 return index + 20 * n;
2301 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2303 const unsigned char *index = p->index_data;
2304 index += 4 * 256;
2305 if (p->index_version == 1) {
2306 return ntohl(*((uint32_t *)(index + 24 * n)));
2307 } else {
2308 uint32_t off;
2309 index += 8 + p->num_objects * (20 + 4);
2310 off = ntohl(*((uint32_t *)(index + 4 * n)));
2311 if (!(off & 0x80000000))
2312 return off;
2313 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2314 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2315 ntohl(*((uint32_t *)(index + 4)));
2319 off_t find_pack_entry_one(const unsigned char *sha1,
2320 struct packed_git *p)
2322 const uint32_t *level1_ofs = p->index_data;
2323 const unsigned char *index = p->index_data;
2324 unsigned hi, lo, stride;
2325 static int use_lookup = -1;
2326 static int debug_lookup = -1;
2328 if (debug_lookup < 0)
2329 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2331 if (!index) {
2332 if (open_pack_index(p))
2333 return 0;
2334 level1_ofs = p->index_data;
2335 index = p->index_data;
2337 if (p->index_version > 1) {
2338 level1_ofs += 2;
2339 index += 8;
2341 index += 4 * 256;
2342 hi = ntohl(level1_ofs[*sha1]);
2343 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2344 if (p->index_version > 1) {
2345 stride = 20;
2346 } else {
2347 stride = 24;
2348 index += 4;
2351 if (debug_lookup)
2352 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2353 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2355 if (use_lookup < 0)
2356 use_lookup = !!getenv("GIT_USE_LOOKUP");
2357 if (use_lookup) {
2358 int pos = sha1_entry_pos(index, stride, 0,
2359 lo, hi, p->num_objects, sha1);
2360 if (pos < 0)
2361 return 0;
2362 return nth_packed_object_offset(p, pos);
2365 do {
2366 unsigned mi = (lo + hi) / 2;
2367 int cmp = hashcmp(index + mi * stride, sha1);
2369 if (debug_lookup)
2370 printf("lo %u hi %u rg %u mi %u\n",
2371 lo, hi, hi - lo, mi);
2372 if (!cmp)
2373 return nth_packed_object_offset(p, mi);
2374 if (cmp > 0)
2375 hi = mi;
2376 else
2377 lo = mi+1;
2378 } while (lo < hi);
2379 return 0;
2382 int is_pack_valid(struct packed_git *p)
2384 /* An already open pack is known to be valid. */
2385 if (p->pack_fd != -1)
2386 return 1;
2388 /* If the pack has one window completely covering the
2389 * file size, the pack is known to be valid even if
2390 * the descriptor is not currently open.
2392 if (p->windows) {
2393 struct pack_window *w = p->windows;
2395 if (!w->offset && w->len == p->pack_size)
2396 return 1;
2399 /* Force the pack to open to prove its valid. */
2400 return !open_packed_git(p);
2403 static int fill_pack_entry(const unsigned char *sha1,
2404 struct pack_entry *e,
2405 struct packed_git *p)
2407 off_t offset;
2409 if (p->num_bad_objects) {
2410 unsigned i;
2411 for (i = 0; i < p->num_bad_objects; i++)
2412 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2413 return 0;
2416 offset = find_pack_entry_one(sha1, p);
2417 if (!offset)
2418 return 0;
2421 * We are about to tell the caller where they can locate the
2422 * requested object. We better make sure the packfile is
2423 * still here and can be accessed before supplying that
2424 * answer, as it may have been deleted since the index was
2425 * loaded!
2427 if (!is_pack_valid(p)) {
2428 warning("packfile %s cannot be accessed", p->pack_name);
2429 return 0;
2431 e->offset = offset;
2432 e->p = p;
2433 hashcpy(e->sha1, sha1);
2434 return 1;
2438 * Iff a pack file contains the object named by sha1, return true and
2439 * store its location to e.
2441 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2443 struct packed_git *p;
2445 prepare_packed_git();
2446 if (!packed_git)
2447 return 0;
2449 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2450 return 1;
2452 for (p = packed_git; p; p = p->next) {
2453 if (p == last_found_pack)
2454 continue; /* we already checked this one */
2456 if (fill_pack_entry(sha1, e, p)) {
2457 last_found_pack = p;
2458 return 1;
2461 return 0;
2464 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2465 struct packed_git *packs)
2467 struct packed_git *p;
2469 for (p = packs; p; p = p->next) {
2470 if (find_pack_entry_one(sha1, p))
2471 return p;
2473 return NULL;
2477 static int sha1_loose_object_info(const unsigned char *sha1,
2478 struct object_info *oi)
2480 int status;
2481 unsigned long mapsize, size;
2482 void *map;
2483 git_zstream stream;
2484 char hdr[32];
2486 if (oi->delta_base_sha1)
2487 hashclr(oi->delta_base_sha1);
2490 * If we don't care about type or size, then we don't
2491 * need to look inside the object at all. Note that we
2492 * do not optimize out the stat call, even if the
2493 * caller doesn't care about the disk-size, since our
2494 * return value implicitly indicates whether the
2495 * object even exists.
2497 if (!oi->typep && !oi->sizep) {
2498 struct stat st;
2499 if (stat_sha1_file(sha1, &st) < 0)
2500 return -1;
2501 if (oi->disk_sizep)
2502 *oi->disk_sizep = st.st_size;
2503 return 0;
2506 map = map_sha1_file(sha1, &mapsize);
2507 if (!map)
2508 return -1;
2509 if (oi->disk_sizep)
2510 *oi->disk_sizep = mapsize;
2511 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2512 status = error("unable to unpack %s header",
2513 sha1_to_hex(sha1));
2514 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2515 status = error("unable to parse %s header", sha1_to_hex(sha1));
2516 else if (oi->sizep)
2517 *oi->sizep = size;
2518 git_inflate_end(&stream);
2519 munmap(map, mapsize);
2520 if (oi->typep)
2521 *oi->typep = status;
2522 return 0;
2525 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2527 struct cached_object *co;
2528 struct pack_entry e;
2529 int rtype;
2530 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2532 co = find_cached_object(real);
2533 if (co) {
2534 if (oi->typep)
2535 *(oi->typep) = co->type;
2536 if (oi->sizep)
2537 *(oi->sizep) = co->size;
2538 if (oi->disk_sizep)
2539 *(oi->disk_sizep) = 0;
2540 if (oi->delta_base_sha1)
2541 hashclr(oi->delta_base_sha1);
2542 oi->whence = OI_CACHED;
2543 return 0;
2546 if (!find_pack_entry(real, &e)) {
2547 /* Most likely it's a loose object. */
2548 if (!sha1_loose_object_info(real, oi)) {
2549 oi->whence = OI_LOOSE;
2550 return 0;
2553 /* Not a loose object; someone else may have just packed it. */
2554 reprepare_packed_git();
2555 if (!find_pack_entry(real, &e))
2556 return -1;
2559 rtype = packed_object_info(e.p, e.offset, oi);
2560 if (rtype < 0) {
2561 mark_bad_packed_object(e.p, real);
2562 return sha1_object_info_extended(real, oi, 0);
2563 } else if (in_delta_base_cache(e.p, e.offset)) {
2564 oi->whence = OI_DBCACHED;
2565 } else {
2566 oi->whence = OI_PACKED;
2567 oi->u.packed.offset = e.offset;
2568 oi->u.packed.pack = e.p;
2569 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2570 rtype == OBJ_OFS_DELTA);
2573 return 0;
2576 /* returns enum object_type or negative */
2577 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2579 enum object_type type;
2580 struct object_info oi = {NULL};
2582 oi.typep = &type;
2583 oi.sizep = sizep;
2584 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2585 return -1;
2586 return type;
2589 static void *read_packed_sha1(const unsigned char *sha1,
2590 enum object_type *type, unsigned long *size)
2592 struct pack_entry e;
2593 void *data;
2595 if (!find_pack_entry(sha1, &e))
2596 return NULL;
2597 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2598 if (!data) {
2600 * We're probably in deep shit, but let's try to fetch
2601 * the required object anyway from another pack or loose.
2602 * This should happen only in the presence of a corrupted
2603 * pack, and is better than failing outright.
2605 error("failed to read object %s at offset %"PRIuMAX" from %s",
2606 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2607 mark_bad_packed_object(e.p, sha1);
2608 data = read_object(sha1, type, size);
2610 return data;
2613 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2614 unsigned char *sha1)
2616 struct cached_object *co;
2618 hash_sha1_file(buf, len, typename(type), sha1);
2619 if (has_sha1_file(sha1) || find_cached_object(sha1))
2620 return 0;
2621 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2622 co = &cached_objects[cached_object_nr++];
2623 co->size = len;
2624 co->type = type;
2625 co->buf = xmalloc(len);
2626 memcpy(co->buf, buf, len);
2627 hashcpy(co->sha1, sha1);
2628 return 0;
2631 static void *read_object(const unsigned char *sha1, enum object_type *type,
2632 unsigned long *size)
2634 unsigned long mapsize;
2635 void *map, *buf;
2636 struct cached_object *co;
2638 co = find_cached_object(sha1);
2639 if (co) {
2640 *type = co->type;
2641 *size = co->size;
2642 return xmemdupz(co->buf, co->size);
2645 buf = read_packed_sha1(sha1, type, size);
2646 if (buf)
2647 return buf;
2648 map = map_sha1_file(sha1, &mapsize);
2649 if (map) {
2650 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2651 munmap(map, mapsize);
2652 return buf;
2654 reprepare_packed_git();
2655 return read_packed_sha1(sha1, type, size);
2659 * This function dies on corrupt objects; the callers who want to
2660 * deal with them should arrange to call read_object() and give error
2661 * messages themselves.
2663 void *read_sha1_file_extended(const unsigned char *sha1,
2664 enum object_type *type,
2665 unsigned long *size,
2666 unsigned flag)
2668 void *data;
2669 const struct packed_git *p;
2670 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2672 errno = 0;
2673 data = read_object(repl, type, size);
2674 if (data)
2675 return data;
2677 if (errno && errno != ENOENT)
2678 die_errno("failed to read object %s", sha1_to_hex(sha1));
2680 /* die if we replaced an object with one that does not exist */
2681 if (repl != sha1)
2682 die("replacement %s not found for %s",
2683 sha1_to_hex(repl), sha1_to_hex(sha1));
2685 if (has_loose_object(repl)) {
2686 const char *path = sha1_file_name(sha1);
2688 die("loose object %s (stored in %s) is corrupt",
2689 sha1_to_hex(repl), path);
2692 if ((p = has_packed_and_bad(repl)) != NULL)
2693 die("packed object %s (stored in %s) is corrupt",
2694 sha1_to_hex(repl), p->pack_name);
2696 return NULL;
2699 void *read_object_with_reference(const unsigned char *sha1,
2700 const char *required_type_name,
2701 unsigned long *size,
2702 unsigned char *actual_sha1_return)
2704 enum object_type type, required_type;
2705 void *buffer;
2706 unsigned long isize;
2707 unsigned char actual_sha1[20];
2709 required_type = type_from_string(required_type_name);
2710 hashcpy(actual_sha1, sha1);
2711 while (1) {
2712 int ref_length = -1;
2713 const char *ref_type = NULL;
2715 buffer = read_sha1_file(actual_sha1, &type, &isize);
2716 if (!buffer)
2717 return NULL;
2718 if (type == required_type) {
2719 *size = isize;
2720 if (actual_sha1_return)
2721 hashcpy(actual_sha1_return, actual_sha1);
2722 return buffer;
2724 /* Handle references */
2725 else if (type == OBJ_COMMIT)
2726 ref_type = "tree ";
2727 else if (type == OBJ_TAG)
2728 ref_type = "object ";
2729 else {
2730 free(buffer);
2731 return NULL;
2733 ref_length = strlen(ref_type);
2735 if (ref_length + 40 > isize ||
2736 memcmp(buffer, ref_type, ref_length) ||
2737 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2738 free(buffer);
2739 return NULL;
2741 free(buffer);
2742 /* Now we have the ID of the referred-to object in
2743 * actual_sha1. Check again. */
2747 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2748 const char *type, unsigned char *sha1,
2749 char *hdr, int *hdrlen)
2751 git_SHA_CTX c;
2753 /* Generate the header */
2754 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2756 /* Sha1.. */
2757 git_SHA1_Init(&c);
2758 git_SHA1_Update(&c, hdr, *hdrlen);
2759 git_SHA1_Update(&c, buf, len);
2760 git_SHA1_Final(sha1, &c);
2764 * Move the just written object into its final resting place.
2765 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2766 * "moving" is only a part of what it does, when no patch between
2767 * master to pu changes the call sites of this function.
2769 int move_temp_to_file(const char *tmpfile, const char *filename)
2771 int ret = 0;
2773 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2774 goto try_rename;
2775 else if (link(tmpfile, filename))
2776 ret = errno;
2779 * Coda hack - coda doesn't like cross-directory links,
2780 * so we fall back to a rename, which will mean that it
2781 * won't be able to check collisions, but that's not a
2782 * big deal.
2784 * The same holds for FAT formatted media.
2786 * When this succeeds, we just return. We have nothing
2787 * left to unlink.
2789 if (ret && ret != EEXIST) {
2790 try_rename:
2791 if (!rename(tmpfile, filename))
2792 goto out;
2793 ret = errno;
2795 unlink_or_warn(tmpfile);
2796 if (ret) {
2797 if (ret != EEXIST) {
2798 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2800 /* FIXME!!! Collision check here ? */
2803 out:
2804 if (adjust_shared_perm(filename))
2805 return error("unable to set permission to '%s'", filename);
2806 return 0;
2809 static int write_buffer(int fd, const void *buf, size_t len)
2811 if (write_in_full(fd, buf, len) < 0)
2812 return error("file write error (%s)", strerror(errno));
2813 return 0;
2816 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2817 unsigned char *sha1)
2819 char hdr[32];
2820 int hdrlen;
2821 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2822 return 0;
2825 /* Finalize a file on disk, and close it. */
2826 static void close_sha1_file(int fd)
2828 if (fsync_object_files)
2829 fsync_or_die(fd, "sha1 file");
2830 if (close(fd) != 0)
2831 die_errno("error when closing sha1 file");
2834 /* Size of directory component, including the ending '/' */
2835 static inline int directory_size(const char *filename)
2837 const char *s = strrchr(filename, '/');
2838 if (!s)
2839 return 0;
2840 return s - filename + 1;
2844 * This creates a temporary file in the same directory as the final
2845 * 'filename'
2847 * We want to avoid cross-directory filename renames, because those
2848 * can have problems on various filesystems (FAT, NFS, Coda).
2850 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2852 int fd, dirlen = directory_size(filename);
2854 if (dirlen + 20 > bufsiz) {
2855 errno = ENAMETOOLONG;
2856 return -1;
2858 memcpy(buffer, filename, dirlen);
2859 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2860 fd = git_mkstemp_mode(buffer, 0444);
2861 if (fd < 0 && dirlen && errno == ENOENT) {
2862 /* Make sure the directory exists */
2863 memcpy(buffer, filename, dirlen);
2864 buffer[dirlen-1] = 0;
2865 if (mkdir(buffer, 0777) && errno != EEXIST)
2866 return -1;
2867 if (adjust_shared_perm(buffer))
2868 return -1;
2870 /* Try again */
2871 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2872 fd = git_mkstemp_mode(buffer, 0444);
2874 return fd;
2877 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2878 const void *buf, unsigned long len, time_t mtime)
2880 int fd, ret;
2881 unsigned char compressed[4096];
2882 git_zstream stream;
2883 git_SHA_CTX c;
2884 unsigned char parano_sha1[20];
2885 static char tmp_file[PATH_MAX];
2886 const char *filename = sha1_file_name(sha1);
2888 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2889 if (fd < 0) {
2890 if (errno == EACCES)
2891 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2892 else
2893 return error("unable to create temporary file: %s", strerror(errno));
2896 /* Set it up */
2897 memset(&stream, 0, sizeof(stream));
2898 git_deflate_init(&stream, zlib_compression_level);
2899 stream.next_out = compressed;
2900 stream.avail_out = sizeof(compressed);
2901 git_SHA1_Init(&c);
2903 /* First header.. */
2904 stream.next_in = (unsigned char *)hdr;
2905 stream.avail_in = hdrlen;
2906 while (git_deflate(&stream, 0) == Z_OK)
2907 ; /* nothing */
2908 git_SHA1_Update(&c, hdr, hdrlen);
2910 /* Then the data itself.. */
2911 stream.next_in = (void *)buf;
2912 stream.avail_in = len;
2913 do {
2914 unsigned char *in0 = stream.next_in;
2915 ret = git_deflate(&stream, Z_FINISH);
2916 git_SHA1_Update(&c, in0, stream.next_in - in0);
2917 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2918 die("unable to write sha1 file");
2919 stream.next_out = compressed;
2920 stream.avail_out = sizeof(compressed);
2921 } while (ret == Z_OK);
2923 if (ret != Z_STREAM_END)
2924 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2925 ret = git_deflate_end_gently(&stream);
2926 if (ret != Z_OK)
2927 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2928 git_SHA1_Final(parano_sha1, &c);
2929 if (hashcmp(sha1, parano_sha1) != 0)
2930 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2932 close_sha1_file(fd);
2934 if (mtime) {
2935 struct utimbuf utb;
2936 utb.actime = mtime;
2937 utb.modtime = mtime;
2938 if (utime(tmp_file, &utb) < 0)
2939 warning("failed utime() on %s: %s",
2940 tmp_file, strerror(errno));
2943 return move_temp_to_file(tmp_file, filename);
2946 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2948 unsigned char sha1[20];
2949 char hdr[32];
2950 int hdrlen;
2952 /* Normally if we have it in the pack then we do not bother writing
2953 * it out into .git/objects/??/?{38} file.
2955 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2956 if (returnsha1)
2957 hashcpy(returnsha1, sha1);
2958 if (has_sha1_file(sha1))
2959 return 0;
2960 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2963 int force_object_loose(const unsigned char *sha1, time_t mtime)
2965 void *buf;
2966 unsigned long len;
2967 enum object_type type;
2968 char hdr[32];
2969 int hdrlen;
2970 int ret;
2972 if (has_loose_object(sha1))
2973 return 0;
2974 buf = read_packed_sha1(sha1, &type, &len);
2975 if (!buf)
2976 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2977 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2978 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2979 free(buf);
2981 return ret;
2984 int has_pack_index(const unsigned char *sha1)
2986 struct stat st;
2987 if (stat(sha1_pack_index_name(sha1), &st))
2988 return 0;
2989 return 1;
2992 int has_sha1_pack(const unsigned char *sha1)
2994 struct pack_entry e;
2995 return find_pack_entry(sha1, &e);
2998 int has_sha1_file(const unsigned char *sha1)
3000 struct pack_entry e;
3002 if (find_pack_entry(sha1, &e))
3003 return 1;
3004 if (has_loose_object(sha1))
3005 return 1;
3006 reprepare_packed_git();
3007 return find_pack_entry(sha1, &e);
3010 static void check_tree(const void *buf, size_t size)
3012 struct tree_desc desc;
3013 struct name_entry entry;
3015 init_tree_desc(&desc, buf, size);
3016 while (tree_entry(&desc, &entry))
3017 /* do nothing
3018 * tree_entry() will die() on malformed entries */
3022 static void check_commit(const void *buf, size_t size)
3024 struct commit c;
3025 memset(&c, 0, sizeof(c));
3026 if (parse_commit_buffer(&c, buf, size))
3027 die("corrupt commit");
3030 static void check_tag(const void *buf, size_t size)
3032 struct tag t;
3033 memset(&t, 0, sizeof(t));
3034 if (parse_tag_buffer(&t, buf, size))
3035 die("corrupt tag");
3038 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3039 enum object_type type,
3040 const char *path, unsigned flags)
3042 int ret, re_allocated = 0;
3043 int write_object = flags & HASH_WRITE_OBJECT;
3045 if (!type)
3046 type = OBJ_BLOB;
3049 * Convert blobs to git internal format
3051 if ((type == OBJ_BLOB) && path) {
3052 struct strbuf nbuf = STRBUF_INIT;
3053 if (convert_to_git(path, buf, size, &nbuf,
3054 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3055 buf = strbuf_detach(&nbuf, &size);
3056 re_allocated = 1;
3059 if (flags & HASH_FORMAT_CHECK) {
3060 if (type == OBJ_TREE)
3061 check_tree(buf, size);
3062 if (type == OBJ_COMMIT)
3063 check_commit(buf, size);
3064 if (type == OBJ_TAG)
3065 check_tag(buf, size);
3068 if (write_object)
3069 ret = write_sha1_file(buf, size, typename(type), sha1);
3070 else
3071 ret = hash_sha1_file(buf, size, typename(type), sha1);
3072 if (re_allocated)
3073 free(buf);
3074 return ret;
3077 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3078 const char *path, unsigned flags)
3080 struct strbuf sbuf = STRBUF_INIT;
3081 int ret;
3083 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3084 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3085 else
3086 ret = -1;
3087 strbuf_release(&sbuf);
3088 return ret;
3091 #define SMALL_FILE_SIZE (32*1024)
3093 static int index_core(unsigned char *sha1, int fd, size_t size,
3094 enum object_type type, const char *path,
3095 unsigned flags)
3097 int ret;
3099 if (!size) {
3100 ret = index_mem(sha1, NULL, size, type, path, flags);
3101 } else if (size <= SMALL_FILE_SIZE) {
3102 char *buf = xmalloc(size);
3103 if (size == read_in_full(fd, buf, size))
3104 ret = index_mem(sha1, buf, size, type, path, flags);
3105 else
3106 ret = error("short read %s", strerror(errno));
3107 free(buf);
3108 } else {
3109 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3110 ret = index_mem(sha1, buf, size, type, path, flags);
3111 munmap(buf, size);
3113 return ret;
3117 * This creates one packfile per large blob unless bulk-checkin
3118 * machinery is "plugged".
3120 * This also bypasses the usual "convert-to-git" dance, and that is on
3121 * purpose. We could write a streaming version of the converting
3122 * functions and insert that before feeding the data to fast-import
3123 * (or equivalent in-core API described above). However, that is
3124 * somewhat complicated, as we do not know the size of the filter
3125 * result, which we need to know beforehand when writing a git object.
3126 * Since the primary motivation for trying to stream from the working
3127 * tree file and to avoid mmaping it in core is to deal with large
3128 * binary blobs, they generally do not want to get any conversion, and
3129 * callers should avoid this code path when filters are requested.
3131 static int index_stream(unsigned char *sha1, int fd, size_t size,
3132 enum object_type type, const char *path,
3133 unsigned flags)
3135 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3138 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3139 enum object_type type, const char *path, unsigned flags)
3141 int ret;
3142 size_t size = xsize_t(st->st_size);
3144 if (!S_ISREG(st->st_mode))
3145 ret = index_pipe(sha1, fd, type, path, flags);
3146 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3147 (path && would_convert_to_git(path, NULL, 0, 0)))
3148 ret = index_core(sha1, fd, size, type, path, flags);
3149 else
3150 ret = index_stream(sha1, fd, size, type, path, flags);
3151 close(fd);
3152 return ret;
3155 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3157 int fd;
3158 struct strbuf sb = STRBUF_INIT;
3160 switch (st->st_mode & S_IFMT) {
3161 case S_IFREG:
3162 fd = open(path, O_RDONLY);
3163 if (fd < 0)
3164 return error("open(\"%s\"): %s", path,
3165 strerror(errno));
3166 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3167 return error("%s: failed to insert into database",
3168 path);
3169 break;
3170 case S_IFLNK:
3171 if (strbuf_readlink(&sb, path, st->st_size)) {
3172 char *errstr = strerror(errno);
3173 return error("readlink(\"%s\"): %s", path,
3174 errstr);
3176 if (!(flags & HASH_WRITE_OBJECT))
3177 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3178 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3179 return error("%s: failed to insert into database",
3180 path);
3181 strbuf_release(&sb);
3182 break;
3183 case S_IFDIR:
3184 return resolve_gitlink_ref(path, "HEAD", sha1);
3185 default:
3186 return error("%s: unsupported file type", path);
3188 return 0;
3191 int read_pack_header(int fd, struct pack_header *header)
3193 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3194 /* "eof before pack header was fully read" */
3195 return PH_ERROR_EOF;
3197 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3198 /* "protocol error (pack signature mismatch detected)" */
3199 return PH_ERROR_PACK_SIGNATURE;
3200 if (!pack_version_ok(header->hdr_version))
3201 /* "protocol error (pack version unsupported)" */
3202 return PH_ERROR_PROTOCOL;
3203 return 0;
3206 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3208 enum object_type type = sha1_object_info(sha1, NULL);
3209 if (type < 0)
3210 die("%s is not a valid object", sha1_to_hex(sha1));
3211 if (type != expect)
3212 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3213 typename(expect));