Merge branch 'maint'
[git.git] / sha1_file.c
bloba38854ce553c1e59294d5542a37a5404ccd66dc5
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];
39 static const char *no_log_pack_access = "no_log_pack_access";
40 static const char *log_pack_access;
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
48 static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53 } *cached_objects;
54 static int cached_object_nr, cached_object_alloc;
56 static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
64 * A pointer to the last packed_git in which an object was found.
65 * When an object is sought, we look in this packfile first, because
66 * objects that are looked up at similar times are often in the same
67 * packfile as one another.
69 static struct packed_git *last_found_pack;
71 static struct cached_object *find_cached_object(const unsigned char *sha1)
73 int i;
74 struct cached_object *co = cached_objects;
76 for (i = 0; i < cached_object_nr; i++, co++) {
77 if (!hashcmp(co->sha1, sha1))
78 return co;
80 if (!hashcmp(sha1, empty_tree.sha1))
81 return &empty_tree;
82 return NULL;
85 int mkdir_in_gitdir(const char *path)
87 if (mkdir(path, 0777)) {
88 int saved_errno = errno;
89 struct stat st;
90 struct strbuf sb = STRBUF_INIT;
92 if (errno != EEXIST)
93 return -1;
95 * Are we looking at a path in a symlinked worktree
96 * whose original repository does not yet have it?
97 * e.g. .git/rr-cache pointing at its original
98 * repository in which the user hasn't performed any
99 * conflict resolution yet?
101 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
102 strbuf_readlink(&sb, path, st.st_size) ||
103 !is_absolute_path(sb.buf) ||
104 mkdir(sb.buf, 0777)) {
105 strbuf_release(&sb);
106 errno = saved_errno;
107 return -1;
109 strbuf_release(&sb);
111 return adjust_shared_perm(path);
114 enum scld_error safe_create_leading_directories(char *path)
116 char *next_component = path + offset_1st_component(path);
117 enum scld_error ret = SCLD_OK;
119 while (ret == SCLD_OK && next_component) {
120 struct stat st;
121 char *slash = next_component, slash_character;
123 while (*slash && !is_dir_sep(*slash))
124 slash++;
126 if (!*slash)
127 break;
129 next_component = slash + 1;
130 while (is_dir_sep(*next_component))
131 next_component++;
132 if (!*next_component)
133 break;
135 slash_character = *slash;
136 *slash = '\0';
137 if (!stat(path, &st)) {
138 /* path exists */
139 if (!S_ISDIR(st.st_mode))
140 ret = SCLD_EXISTS;
141 } else if (mkdir(path, 0777)) {
142 if (errno == EEXIST &&
143 !stat(path, &st) && S_ISDIR(st.st_mode))
144 ; /* somebody created it since we checked */
145 else if (errno == ENOENT)
147 * Either mkdir() failed because
148 * somebody just pruned the containing
149 * directory, or stat() failed because
150 * the file that was in our way was
151 * just removed. Either way, inform
152 * the caller that it might be worth
153 * trying again:
155 ret = SCLD_VANISHED;
156 else
157 ret = SCLD_FAILED;
158 } else if (adjust_shared_perm(path)) {
159 ret = SCLD_PERMS;
161 *slash = slash_character;
163 return ret;
166 enum scld_error safe_create_leading_directories_const(const char *path)
168 /* path points to cache entries, so xstrdup before messing with it */
169 char *buf = xstrdup(path);
170 enum scld_error result = safe_create_leading_directories(buf);
171 free(buf);
172 return result;
175 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
177 int i;
178 for (i = 0; i < 20; i++) {
179 static char hex[] = "0123456789abcdef";
180 unsigned int val = sha1[i];
181 char *pos = pathbuf + i*2 + (i > 0);
182 *pos++ = hex[val >> 4];
183 *pos = hex[val & 0xf];
187 const char *sha1_file_name(const unsigned char *sha1)
189 static char buf[PATH_MAX];
190 const char *objdir;
191 int len;
193 objdir = get_object_directory();
194 len = strlen(objdir);
196 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
197 if (len + 43 > PATH_MAX)
198 die("insanely long object directory %s", objdir);
199 memcpy(buf, objdir, len);
200 buf[len] = '/';
201 buf[len+3] = '/';
202 buf[len+42] = '\0';
203 fill_sha1_path(buf + len + 1, sha1);
204 return buf;
208 * Return the name of the pack or index file with the specified sha1
209 * in its filename. *base and *name are scratch space that must be
210 * provided by the caller. which should be "pack" or "idx".
212 static char *sha1_get_pack_name(const unsigned char *sha1,
213 char **name, char **base, const char *which)
215 static const char hex[] = "0123456789abcdef";
216 char *buf;
217 int i;
219 if (!*base) {
220 const char *sha1_file_directory = get_object_directory();
221 int len = strlen(sha1_file_directory);
222 *base = xmalloc(len + 60);
223 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
224 sha1_file_directory, which);
225 *name = *base + len + 11;
228 buf = *name;
230 for (i = 0; i < 20; i++) {
231 unsigned int val = *sha1++;
232 *buf++ = hex[val >> 4];
233 *buf++ = hex[val & 0xf];
236 return *base;
239 char *sha1_pack_name(const unsigned char *sha1)
241 static char *name, *base;
243 return sha1_get_pack_name(sha1, &name, &base, "pack");
246 char *sha1_pack_index_name(const unsigned char *sha1)
248 static char *name, *base;
250 return sha1_get_pack_name(sha1, &name, &base, "idx");
253 struct alternate_object_database *alt_odb_list;
254 static struct alternate_object_database **alt_odb_tail;
257 * Prepare alternate object database registry.
259 * The variable alt_odb_list points at the list of struct
260 * alternate_object_database. The elements on this list come from
261 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
262 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
263 * whose contents is similar to that environment variable but can be
264 * LF separated. Its base points at a statically allocated buffer that
265 * contains "/the/directory/corresponding/to/.git/objects/...", while
266 * its name points just after the slash at the end of ".git/objects/"
267 * in the example above, and has enough space to hold 40-byte hex
268 * SHA1, an extra slash for the first level indirection, and the
269 * terminating NUL.
271 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
273 const char *objdir = get_object_directory();
274 struct alternate_object_database *ent;
275 struct alternate_object_database *alt;
276 int pfxlen, entlen;
277 struct strbuf pathbuf = STRBUF_INIT;
279 if (!is_absolute_path(entry) && relative_base) {
280 strbuf_addstr(&pathbuf, real_path(relative_base));
281 strbuf_addch(&pathbuf, '/');
283 strbuf_addstr(&pathbuf, entry);
285 normalize_path_copy(pathbuf.buf, pathbuf.buf);
287 pfxlen = strlen(pathbuf.buf);
290 * The trailing slash after the directory name is given by
291 * this function at the end. Remove duplicates.
293 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
294 pfxlen -= 1;
296 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
297 ent = xmalloc(sizeof(*ent) + entlen);
298 memcpy(ent->base, pathbuf.buf, pfxlen);
299 strbuf_release(&pathbuf);
301 ent->name = ent->base + pfxlen + 1;
302 ent->base[pfxlen + 3] = '/';
303 ent->base[pfxlen] = ent->base[entlen-1] = 0;
305 /* Detect cases where alternate disappeared */
306 if (!is_directory(ent->base)) {
307 error("object directory %s does not exist; "
308 "check .git/objects/info/alternates.",
309 ent->base);
310 free(ent);
311 return -1;
314 /* Prevent the common mistake of listing the same
315 * thing twice, or object directory itself.
317 for (alt = alt_odb_list; alt; alt = alt->next) {
318 if (pfxlen == alt->name - alt->base - 1 &&
319 !memcmp(ent->base, alt->base, pfxlen)) {
320 free(ent);
321 return -1;
324 if (!strcmp(ent->base, objdir)) {
325 free(ent);
326 return -1;
329 /* add the alternate entry */
330 *alt_odb_tail = ent;
331 alt_odb_tail = &(ent->next);
332 ent->next = NULL;
334 /* recursively add alternates */
335 read_info_alternates(ent->base, depth + 1);
337 ent->base[pfxlen] = '/';
339 return 0;
342 static void link_alt_odb_entries(const char *alt, int len, int sep,
343 const char *relative_base, int depth)
345 struct string_list entries = STRING_LIST_INIT_NODUP;
346 char *alt_copy;
347 int i;
349 if (depth > 5) {
350 error("%s: ignoring alternate object stores, nesting too deep.",
351 relative_base);
352 return;
355 alt_copy = xmemdupz(alt, len);
356 string_list_split_in_place(&entries, alt_copy, sep, -1);
357 for (i = 0; i < entries.nr; i++) {
358 const char *entry = entries.items[i].string;
359 if (entry[0] == '\0' || entry[0] == '#')
360 continue;
361 if (!is_absolute_path(entry) && depth) {
362 error("%s: ignoring relative alternate object store %s",
363 relative_base, entry);
364 } else {
365 link_alt_odb_entry(entry, relative_base, depth);
368 string_list_clear(&entries, 0);
369 free(alt_copy);
372 void read_info_alternates(const char * relative_base, int depth)
374 char *map;
375 size_t mapsz;
376 struct stat st;
377 const char alt_file_name[] = "info/alternates";
378 /* Given that relative_base is no longer than PATH_MAX,
379 ensure that "path" has enough space to append "/", the
380 file name, "info/alternates", and a trailing NUL. */
381 char path[PATH_MAX + 1 + sizeof alt_file_name];
382 int fd;
384 sprintf(path, "%s/%s", relative_base, alt_file_name);
385 fd = git_open_noatime(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
398 munmap(map, mapsz);
401 void add_to_alternates_file(const char *reference)
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
405 char *alt = mkpath("%s\n", reference);
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
413 void foreach_alt_odb(alt_odb_fn fn, void *cb)
415 struct alternate_object_database *ent;
417 prepare_alt_odb();
418 for (ent = alt_odb_list; ent; ent = ent->next)
419 if (fn(ent, cb))
420 return;
423 void prepare_alt_odb(void)
425 const char *alt;
427 if (alt_odb_tail)
428 return;
430 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
431 if (!alt) alt = "";
433 alt_odb_tail = &alt_odb_list;
434 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
436 read_info_alternates(get_object_directory(), 0);
439 static int has_loose_object_local(const unsigned char *sha1)
441 return !access(sha1_file_name(sha1), F_OK);
444 int has_loose_object_nonlocal(const unsigned char *sha1)
446 struct alternate_object_database *alt;
447 prepare_alt_odb();
448 for (alt = alt_odb_list; alt; alt = alt->next) {
449 fill_sha1_path(alt->name, sha1);
450 if (!access(alt->base, F_OK))
451 return 1;
453 return 0;
456 static int has_loose_object(const unsigned char *sha1)
458 return has_loose_object_local(sha1) ||
459 has_loose_object_nonlocal(sha1);
462 static unsigned int pack_used_ctr;
463 static unsigned int pack_mmap_calls;
464 static unsigned int peak_pack_open_windows;
465 static unsigned int pack_open_windows;
466 static unsigned int pack_open_fds;
467 static unsigned int pack_max_fds;
468 static size_t peak_pack_mapped;
469 static size_t pack_mapped;
470 struct packed_git *packed_git;
472 void pack_report(void)
474 fprintf(stderr,
475 "pack_report: getpagesize() = %10" SZ_FMT "\n"
476 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
477 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
478 sz_fmt(getpagesize()),
479 sz_fmt(packed_git_window_size),
480 sz_fmt(packed_git_limit));
481 fprintf(stderr,
482 "pack_report: pack_used_ctr = %10u\n"
483 "pack_report: pack_mmap_calls = %10u\n"
484 "pack_report: pack_open_windows = %10u / %10u\n"
485 "pack_report: pack_mapped = "
486 "%10" SZ_FMT " / %10" SZ_FMT "\n",
487 pack_used_ctr,
488 pack_mmap_calls,
489 pack_open_windows, peak_pack_open_windows,
490 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
494 * Open and mmap the index file at path, perform a couple of
495 * consistency checks, then record its information to p. Return 0 on
496 * success.
498 static int check_packed_git_idx(const char *path, struct packed_git *p)
500 void *idx_map;
501 struct pack_idx_header *hdr;
502 size_t idx_size;
503 uint32_t version, nr, i, *index;
504 int fd = git_open_noatime(path);
505 struct stat st;
507 if (fd < 0)
508 return -1;
509 if (fstat(fd, &st)) {
510 close(fd);
511 return -1;
513 idx_size = xsize_t(st.st_size);
514 if (idx_size < 4 * 256 + 20 + 20) {
515 close(fd);
516 return error("index file %s is too small", path);
518 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
519 close(fd);
521 hdr = idx_map;
522 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
523 version = ntohl(hdr->idx_version);
524 if (version < 2 || version > 2) {
525 munmap(idx_map, idx_size);
526 return error("index file %s is version %"PRIu32
527 " and is not supported by this binary"
528 " (try upgrading GIT to a newer version)",
529 path, version);
531 } else
532 version = 1;
534 nr = 0;
535 index = idx_map;
536 if (version > 1)
537 index += 2; /* skip index header */
538 for (i = 0; i < 256; i++) {
539 uint32_t n = ntohl(index[i]);
540 if (n < nr) {
541 munmap(idx_map, idx_size);
542 return error("non-monotonic index %s", path);
544 nr = n;
547 if (version == 1) {
549 * Total size:
550 * - 256 index entries 4 bytes each
551 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
552 * - 20-byte SHA1 of the packfile
553 * - 20-byte SHA1 file checksum
555 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
556 munmap(idx_map, idx_size);
557 return error("wrong index v1 file size in %s", path);
559 } else if (version == 2) {
561 * Minimum size:
562 * - 8 bytes of header
563 * - 256 index entries 4 bytes each
564 * - 20-byte sha1 entry * nr
565 * - 4-byte crc entry * nr
566 * - 4-byte offset entry * nr
567 * - 20-byte SHA1 of the packfile
568 * - 20-byte SHA1 file checksum
569 * And after the 4-byte offset table might be a
570 * variable sized table containing 8-byte entries
571 * for offsets larger than 2^31.
573 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
574 unsigned long max_size = min_size;
575 if (nr)
576 max_size += (nr - 1)*8;
577 if (idx_size < min_size || idx_size > max_size) {
578 munmap(idx_map, idx_size);
579 return error("wrong index v2 file size in %s", path);
581 if (idx_size != min_size &&
583 * make sure we can deal with large pack offsets.
584 * 31-bit signed offset won't be enough, neither
585 * 32-bit unsigned one will be.
587 (sizeof(off_t) <= 4)) {
588 munmap(idx_map, idx_size);
589 return error("pack too large for current definition of off_t in %s", path);
593 p->index_version = version;
594 p->index_data = idx_map;
595 p->index_size = idx_size;
596 p->num_objects = nr;
597 return 0;
600 int open_pack_index(struct packed_git *p)
602 char *idx_name;
603 int ret;
605 if (p->index_data)
606 return 0;
608 idx_name = xstrdup(p->pack_name);
609 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
610 ret = check_packed_git_idx(idx_name, p);
611 free(idx_name);
612 return ret;
615 static void scan_windows(struct packed_git *p,
616 struct packed_git **lru_p,
617 struct pack_window **lru_w,
618 struct pack_window **lru_l)
620 struct pack_window *w, *w_l;
622 for (w_l = NULL, w = p->windows; w; w = w->next) {
623 if (!w->inuse_cnt) {
624 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
625 *lru_p = p;
626 *lru_w = w;
627 *lru_l = w_l;
630 w_l = w;
634 static int unuse_one_window(struct packed_git *current)
636 struct packed_git *p, *lru_p = NULL;
637 struct pack_window *lru_w = NULL, *lru_l = NULL;
639 if (current)
640 scan_windows(current, &lru_p, &lru_w, &lru_l);
641 for (p = packed_git; p; p = p->next)
642 scan_windows(p, &lru_p, &lru_w, &lru_l);
643 if (lru_p) {
644 munmap(lru_w->base, lru_w->len);
645 pack_mapped -= lru_w->len;
646 if (lru_l)
647 lru_l->next = lru_w->next;
648 else
649 lru_p->windows = lru_w->next;
650 free(lru_w);
651 pack_open_windows--;
652 return 1;
654 return 0;
657 void release_pack_memory(size_t need)
659 size_t cur = pack_mapped;
660 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
661 ; /* nothing */
664 void *xmmap(void *start, size_t length,
665 int prot, int flags, int fd, off_t offset)
667 void *ret = mmap(start, length, prot, flags, fd, offset);
668 if (ret == MAP_FAILED) {
669 if (!length)
670 return NULL;
671 release_pack_memory(length);
672 ret = mmap(start, length, prot, flags, fd, offset);
673 if (ret == MAP_FAILED)
674 die_errno("Out of memory? mmap failed");
676 return ret;
679 void close_pack_windows(struct packed_git *p)
681 while (p->windows) {
682 struct pack_window *w = p->windows;
684 if (w->inuse_cnt)
685 die("pack '%s' still has open windows to it",
686 p->pack_name);
687 munmap(w->base, w->len);
688 pack_mapped -= w->len;
689 pack_open_windows--;
690 p->windows = w->next;
691 free(w);
696 * The LRU pack is the one with the oldest MRU window, preferring packs
697 * with no used windows, or the oldest mtime if it has no windows allocated.
699 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
701 struct pack_window *w, *this_mru_w;
702 int has_windows_inuse = 0;
705 * Reject this pack if it has windows and the previously selected
706 * one does not. If this pack does not have windows, reject
707 * it if the pack file is newer than the previously selected one.
709 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
710 return;
712 for (w = this_mru_w = p->windows; w; w = w->next) {
714 * Reject this pack if any of its windows are in use,
715 * but the previously selected pack did not have any
716 * inuse windows. Otherwise, record that this pack
717 * has windows in use.
719 if (w->inuse_cnt) {
720 if (*accept_windows_inuse)
721 has_windows_inuse = 1;
722 else
723 return;
726 if (w->last_used > this_mru_w->last_used)
727 this_mru_w = w;
730 * Reject this pack if it has windows that have been
731 * used more recently than the previously selected pack.
732 * If the previously selected pack had windows inuse and
733 * we have not encountered a window in this pack that is
734 * inuse, skip this check since we prefer a pack with no
735 * inuse windows to one that has inuse windows.
737 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
738 this_mru_w->last_used > (*mru_w)->last_used)
739 return;
743 * Select this pack.
745 *mru_w = this_mru_w;
746 *lru_p = p;
747 *accept_windows_inuse = has_windows_inuse;
750 static int close_one_pack(void)
752 struct packed_git *p, *lru_p = NULL;
753 struct pack_window *mru_w = NULL;
754 int accept_windows_inuse = 1;
756 for (p = packed_git; p; p = p->next) {
757 if (p->pack_fd == -1)
758 continue;
759 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
762 if (lru_p) {
763 close(lru_p->pack_fd);
764 pack_open_fds--;
765 lru_p->pack_fd = -1;
766 return 1;
769 return 0;
772 void unuse_pack(struct pack_window **w_cursor)
774 struct pack_window *w = *w_cursor;
775 if (w) {
776 w->inuse_cnt--;
777 *w_cursor = NULL;
781 void close_pack_index(struct packed_git *p)
783 if (p->index_data) {
784 munmap((void *)p->index_data, p->index_size);
785 p->index_data = NULL;
790 * This is used by git-repack in case a newly created pack happens to
791 * contain the same set of objects as an existing one. In that case
792 * the resulting file might be different even if its name would be the
793 * same. It is best to close any reference to the old pack before it is
794 * replaced on disk. Of course no index pointers or windows for given pack
795 * must subsist at this point. If ever objects from this pack are requested
796 * again, the new version of the pack will be reinitialized through
797 * reprepare_packed_git().
799 void free_pack_by_name(const char *pack_name)
801 struct packed_git *p, **pp = &packed_git;
803 while (*pp) {
804 p = *pp;
805 if (strcmp(pack_name, p->pack_name) == 0) {
806 clear_delta_base_cache();
807 close_pack_windows(p);
808 if (p->pack_fd != -1) {
809 close(p->pack_fd);
810 pack_open_fds--;
812 close_pack_index(p);
813 free(p->bad_object_sha1);
814 *pp = p->next;
815 if (last_found_pack == p)
816 last_found_pack = NULL;
817 free(p);
818 return;
820 pp = &p->next;
824 static unsigned int get_max_fd_limit(void)
826 #ifdef RLIMIT_NOFILE
828 struct rlimit lim;
830 if (!getrlimit(RLIMIT_NOFILE, &lim))
831 return lim.rlim_cur;
833 #endif
835 #ifdef _SC_OPEN_MAX
837 long open_max = sysconf(_SC_OPEN_MAX);
838 if (0 < open_max)
839 return open_max;
841 * Otherwise, we got -1 for one of the two
842 * reasons:
844 * (1) sysconf() did not understand _SC_OPEN_MAX
845 * and signaled an error with -1; or
846 * (2) sysconf() said there is no limit.
848 * We _could_ clear errno before calling sysconf() to
849 * tell these two cases apart and return a huge number
850 * in the latter case to let the caller cap it to a
851 * value that is not so selfish, but letting the
852 * fallback OPEN_MAX codepath take care of these cases
853 * is a lot simpler.
856 #endif
858 #ifdef OPEN_MAX
859 return OPEN_MAX;
860 #else
861 return 1; /* see the caller ;-) */
862 #endif
866 * Do not call this directly as this leaks p->pack_fd on error return;
867 * call open_packed_git() instead.
869 static int open_packed_git_1(struct packed_git *p)
871 struct stat st;
872 struct pack_header hdr;
873 unsigned char sha1[20];
874 unsigned char *idx_sha1;
875 long fd_flag;
877 if (!p->index_data && open_pack_index(p))
878 return error("packfile %s index unavailable", p->pack_name);
880 if (!pack_max_fds) {
881 unsigned int max_fds = get_max_fd_limit();
883 /* Save 3 for stdin/stdout/stderr, 22 for work */
884 if (25 < max_fds)
885 pack_max_fds = max_fds - 25;
886 else
887 pack_max_fds = 1;
890 while (pack_max_fds <= pack_open_fds && close_one_pack())
891 ; /* nothing */
893 p->pack_fd = git_open_noatime(p->pack_name);
894 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
895 return -1;
896 pack_open_fds++;
898 /* If we created the struct before we had the pack we lack size. */
899 if (!p->pack_size) {
900 if (!S_ISREG(st.st_mode))
901 return error("packfile %s not a regular file", p->pack_name);
902 p->pack_size = st.st_size;
903 } else if (p->pack_size != st.st_size)
904 return error("packfile %s size changed", p->pack_name);
906 /* We leave these file descriptors open with sliding mmap;
907 * there is no point keeping them open across exec(), though.
909 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
910 if (fd_flag < 0)
911 return error("cannot determine file descriptor flags");
912 fd_flag |= FD_CLOEXEC;
913 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
914 return error("cannot set FD_CLOEXEC");
916 /* Verify we recognize this pack file format. */
917 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
918 return error("file %s is far too short to be a packfile", p->pack_name);
919 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
920 return error("file %s is not a GIT packfile", p->pack_name);
921 if (!pack_version_ok(hdr.hdr_version))
922 return error("packfile %s is version %"PRIu32" and not"
923 " supported (try upgrading GIT to a newer version)",
924 p->pack_name, ntohl(hdr.hdr_version));
926 /* Verify the pack matches its index. */
927 if (p->num_objects != ntohl(hdr.hdr_entries))
928 return error("packfile %s claims to have %"PRIu32" objects"
929 " while index indicates %"PRIu32" objects",
930 p->pack_name, ntohl(hdr.hdr_entries),
931 p->num_objects);
932 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
933 return error("end of packfile %s is unavailable", p->pack_name);
934 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
935 return error("packfile %s signature is unavailable", p->pack_name);
936 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
937 if (hashcmp(sha1, idx_sha1))
938 return error("packfile %s does not match index", p->pack_name);
939 return 0;
942 static int open_packed_git(struct packed_git *p)
944 if (!open_packed_git_1(p))
945 return 0;
946 if (p->pack_fd != -1) {
947 close(p->pack_fd);
948 pack_open_fds--;
949 p->pack_fd = -1;
951 return -1;
954 static int in_window(struct pack_window *win, off_t offset)
956 /* We must promise at least 20 bytes (one hash) after the
957 * offset is available from this window, otherwise the offset
958 * is not actually in this window and a different window (which
959 * has that one hash excess) must be used. This is to support
960 * the object header and delta base parsing routines below.
962 off_t win_off = win->offset;
963 return win_off <= offset
964 && (offset + 20) <= (win_off + win->len);
967 unsigned char *use_pack(struct packed_git *p,
968 struct pack_window **w_cursor,
969 off_t offset,
970 unsigned long *left)
972 struct pack_window *win = *w_cursor;
974 /* Since packfiles end in a hash of their content and it's
975 * pointless to ask for an offset into the middle of that
976 * hash, and the in_window function above wouldn't match
977 * don't allow an offset too close to the end of the file.
979 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
980 die("packfile %s cannot be accessed", p->pack_name);
981 if (offset > (p->pack_size - 20))
982 die("offset beyond end of packfile (truncated pack?)");
984 if (!win || !in_window(win, offset)) {
985 if (win)
986 win->inuse_cnt--;
987 for (win = p->windows; win; win = win->next) {
988 if (in_window(win, offset))
989 break;
991 if (!win) {
992 size_t window_align = packed_git_window_size / 2;
993 off_t len;
995 if (p->pack_fd == -1 && open_packed_git(p))
996 die("packfile %s cannot be accessed", p->pack_name);
998 win = xcalloc(1, sizeof(*win));
999 win->offset = (offset / window_align) * window_align;
1000 len = p->pack_size - win->offset;
1001 if (len > packed_git_window_size)
1002 len = packed_git_window_size;
1003 win->len = (size_t)len;
1004 pack_mapped += win->len;
1005 while (packed_git_limit < pack_mapped
1006 && unuse_one_window(p))
1007 ; /* nothing */
1008 win->base = xmmap(NULL, win->len,
1009 PROT_READ, MAP_PRIVATE,
1010 p->pack_fd, win->offset);
1011 if (win->base == MAP_FAILED)
1012 die("packfile %s cannot be mapped: %s",
1013 p->pack_name,
1014 strerror(errno));
1015 if (!win->offset && win->len == p->pack_size
1016 && !p->do_not_close) {
1017 close(p->pack_fd);
1018 pack_open_fds--;
1019 p->pack_fd = -1;
1021 pack_mmap_calls++;
1022 pack_open_windows++;
1023 if (pack_mapped > peak_pack_mapped)
1024 peak_pack_mapped = pack_mapped;
1025 if (pack_open_windows > peak_pack_open_windows)
1026 peak_pack_open_windows = pack_open_windows;
1027 win->next = p->windows;
1028 p->windows = win;
1031 if (win != *w_cursor) {
1032 win->last_used = pack_used_ctr++;
1033 win->inuse_cnt++;
1034 *w_cursor = win;
1036 offset -= win->offset;
1037 if (left)
1038 *left = win->len - xsize_t(offset);
1039 return win->base + offset;
1042 static struct packed_git *alloc_packed_git(int extra)
1044 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1045 memset(p, 0, sizeof(*p));
1046 p->pack_fd = -1;
1047 return p;
1050 static void try_to_free_pack_memory(size_t size)
1052 release_pack_memory(size);
1055 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1057 static int have_set_try_to_free_routine;
1058 struct stat st;
1059 struct packed_git *p = alloc_packed_git(path_len + 2);
1061 if (!have_set_try_to_free_routine) {
1062 have_set_try_to_free_routine = 1;
1063 set_try_to_free_routine(try_to_free_pack_memory);
1067 * Make sure a corresponding .pack file exists and that
1068 * the index looks sane.
1070 path_len -= strlen(".idx");
1071 if (path_len < 1) {
1072 free(p);
1073 return NULL;
1075 memcpy(p->pack_name, path, path_len);
1077 strcpy(p->pack_name + path_len, ".keep");
1078 if (!access(p->pack_name, F_OK))
1079 p->pack_keep = 1;
1081 strcpy(p->pack_name + path_len, ".pack");
1082 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1083 free(p);
1084 return NULL;
1087 /* ok, it looks sane as far as we can check without
1088 * actually mapping the pack file.
1090 p->pack_size = st.st_size;
1091 p->pack_local = local;
1092 p->mtime = st.st_mtime;
1093 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1094 hashclr(p->sha1);
1095 return p;
1098 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1100 const char *path = sha1_pack_name(sha1);
1101 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1103 strcpy(p->pack_name, path);
1104 hashcpy(p->sha1, sha1);
1105 if (check_packed_git_idx(idx_path, p)) {
1106 free(p);
1107 return NULL;
1110 return p;
1113 void install_packed_git(struct packed_git *pack)
1115 if (pack->pack_fd != -1)
1116 pack_open_fds++;
1118 pack->next = packed_git;
1119 packed_git = pack;
1122 void (*report_garbage)(const char *desc, const char *path);
1124 static void report_helper(const struct string_list *list,
1125 int seen_bits, int first, int last)
1127 const char *msg;
1128 switch (seen_bits) {
1129 case 0:
1130 msg = "no corresponding .idx or .pack";
1131 break;
1132 case 1:
1133 msg = "no corresponding .idx";
1134 break;
1135 case 2:
1136 msg = "no corresponding .pack";
1137 break;
1138 default:
1139 return;
1141 for (; first < last; first++)
1142 report_garbage(msg, list->items[first].string);
1145 static void report_pack_garbage(struct string_list *list)
1147 int i, baselen = -1, first = 0, seen_bits = 0;
1149 if (!report_garbage)
1150 return;
1152 sort_string_list(list);
1154 for (i = 0; i < list->nr; i++) {
1155 const char *path = list->items[i].string;
1156 if (baselen != -1 &&
1157 strncmp(path, list->items[first].string, baselen)) {
1158 report_helper(list, seen_bits, first, i);
1159 baselen = -1;
1160 seen_bits = 0;
1162 if (baselen == -1) {
1163 const char *dot = strrchr(path, '.');
1164 if (!dot) {
1165 report_garbage("garbage found", path);
1166 continue;
1168 baselen = dot - path + 1;
1169 first = i;
1171 if (!strcmp(path + baselen, "pack"))
1172 seen_bits |= 1;
1173 else if (!strcmp(path + baselen, "idx"))
1174 seen_bits |= 2;
1176 report_helper(list, seen_bits, first, list->nr);
1179 static void prepare_packed_git_one(char *objdir, int local)
1181 /* Ensure that this buffer is large enough so that we can
1182 append "/pack/" without clobbering the stack even if
1183 strlen(objdir) were PATH_MAX. */
1184 char path[PATH_MAX + 1 + 4 + 1 + 1];
1185 int len;
1186 DIR *dir;
1187 struct dirent *de;
1188 struct string_list garbage = STRING_LIST_INIT_DUP;
1190 sprintf(path, "%s/pack", objdir);
1191 len = strlen(path);
1192 dir = opendir(path);
1193 if (!dir) {
1194 if (errno != ENOENT)
1195 error("unable to open object pack directory: %s: %s",
1196 path, strerror(errno));
1197 return;
1199 path[len++] = '/';
1200 while ((de = readdir(dir)) != NULL) {
1201 int namelen = strlen(de->d_name);
1202 struct packed_git *p;
1204 if (len + namelen + 1 > sizeof(path)) {
1205 if (report_garbage) {
1206 struct strbuf sb = STRBUF_INIT;
1207 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1208 report_garbage("path too long", sb.buf);
1209 strbuf_release(&sb);
1211 continue;
1214 if (is_dot_or_dotdot(de->d_name))
1215 continue;
1217 strcpy(path + len, de->d_name);
1219 if (has_extension(de->d_name, ".idx")) {
1220 /* Don't reopen a pack we already have. */
1221 for (p = packed_git; p; p = p->next) {
1222 if (!memcmp(path, p->pack_name, len + namelen - 4))
1223 break;
1225 if (p == NULL &&
1227 * See if it really is a valid .idx file with
1228 * corresponding .pack file that we can map.
1230 (p = add_packed_git(path, len + namelen, local)) != NULL)
1231 install_packed_git(p);
1234 if (!report_garbage)
1235 continue;
1237 if (has_extension(de->d_name, ".idx") ||
1238 has_extension(de->d_name, ".pack") ||
1239 has_extension(de->d_name, ".bitmap") ||
1240 has_extension(de->d_name, ".keep"))
1241 string_list_append(&garbage, path);
1242 else
1243 report_garbage("garbage found", path);
1245 closedir(dir);
1246 report_pack_garbage(&garbage);
1247 string_list_clear(&garbage, 0);
1250 static int sort_pack(const void *a_, const void *b_)
1252 struct packed_git *a = *((struct packed_git **)a_);
1253 struct packed_git *b = *((struct packed_git **)b_);
1254 int st;
1257 * Local packs tend to contain objects specific to our
1258 * variant of the project than remote ones. In addition,
1259 * remote ones could be on a network mounted filesystem.
1260 * Favor local ones for these reasons.
1262 st = a->pack_local - b->pack_local;
1263 if (st)
1264 return -st;
1267 * Younger packs tend to contain more recent objects,
1268 * and more recent objects tend to get accessed more
1269 * often.
1271 if (a->mtime < b->mtime)
1272 return 1;
1273 else if (a->mtime == b->mtime)
1274 return 0;
1275 return -1;
1278 static void rearrange_packed_git(void)
1280 struct packed_git **ary, *p;
1281 int i, n;
1283 for (n = 0, p = packed_git; p; p = p->next)
1284 n++;
1285 if (n < 2)
1286 return;
1288 /* prepare an array of packed_git for easier sorting */
1289 ary = xcalloc(n, sizeof(struct packed_git *));
1290 for (n = 0, p = packed_git; p; p = p->next)
1291 ary[n++] = p;
1293 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1295 /* link them back again */
1296 for (i = 0; i < n - 1; i++)
1297 ary[i]->next = ary[i + 1];
1298 ary[n - 1]->next = NULL;
1299 packed_git = ary[0];
1301 free(ary);
1304 static int prepare_packed_git_run_once = 0;
1305 void prepare_packed_git(void)
1307 struct alternate_object_database *alt;
1309 if (prepare_packed_git_run_once)
1310 return;
1311 prepare_packed_git_one(get_object_directory(), 1);
1312 prepare_alt_odb();
1313 for (alt = alt_odb_list; alt; alt = alt->next) {
1314 alt->name[-1] = 0;
1315 prepare_packed_git_one(alt->base, 0);
1316 alt->name[-1] = '/';
1318 rearrange_packed_git();
1319 prepare_packed_git_run_once = 1;
1322 void reprepare_packed_git(void)
1324 prepare_packed_git_run_once = 0;
1325 prepare_packed_git();
1328 static void mark_bad_packed_object(struct packed_git *p,
1329 const unsigned char *sha1)
1331 unsigned i;
1332 for (i = 0; i < p->num_bad_objects; i++)
1333 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1334 return;
1335 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1336 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1337 p->num_bad_objects++;
1340 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1342 struct packed_git *p;
1343 unsigned i;
1345 for (p = packed_git; p; p = p->next)
1346 for (i = 0; i < p->num_bad_objects; i++)
1347 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1348 return p;
1349 return NULL;
1353 * With an in-core object data in "map", rehash it to make sure the
1354 * object name actually matches "sha1" to detect object corruption.
1355 * With "map" == NULL, try reading the object named with "sha1" using
1356 * the streaming interface and rehash it to do the same.
1358 int check_sha1_signature(const unsigned char *sha1, void *map,
1359 unsigned long size, const char *type)
1361 unsigned char real_sha1[20];
1362 enum object_type obj_type;
1363 struct git_istream *st;
1364 git_SHA_CTX c;
1365 char hdr[32];
1366 int hdrlen;
1368 if (map) {
1369 hash_sha1_file(map, size, type, real_sha1);
1370 return hashcmp(sha1, real_sha1) ? -1 : 0;
1373 st = open_istream(sha1, &obj_type, &size, NULL);
1374 if (!st)
1375 return -1;
1377 /* Generate the header */
1378 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1380 /* Sha1.. */
1381 git_SHA1_Init(&c);
1382 git_SHA1_Update(&c, hdr, hdrlen);
1383 for (;;) {
1384 char buf[1024 * 16];
1385 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1387 if (readlen < 0) {
1388 close_istream(st);
1389 return -1;
1391 if (!readlen)
1392 break;
1393 git_SHA1_Update(&c, buf, readlen);
1395 git_SHA1_Final(real_sha1, &c);
1396 close_istream(st);
1397 return hashcmp(sha1, real_sha1) ? -1 : 0;
1400 int git_open_noatime(const char *name)
1402 static int sha1_file_open_flag = O_NOATIME;
1404 for (;;) {
1405 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1406 if (fd >= 0)
1407 return fd;
1409 /* Might the failure be due to O_NOATIME? */
1410 if (errno != ENOENT && sha1_file_open_flag) {
1411 sha1_file_open_flag = 0;
1412 continue;
1415 return -1;
1419 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1421 struct alternate_object_database *alt;
1423 if (!lstat(sha1_file_name(sha1), st))
1424 return 0;
1426 prepare_alt_odb();
1427 errno = ENOENT;
1428 for (alt = alt_odb_list; alt; alt = alt->next) {
1429 fill_sha1_path(alt->name, sha1);
1430 if (!lstat(alt->base, st))
1431 return 0;
1434 return -1;
1437 static int open_sha1_file(const unsigned char *sha1)
1439 int fd;
1440 struct alternate_object_database *alt;
1441 int most_interesting_errno;
1443 fd = git_open_noatime(sha1_file_name(sha1));
1444 if (fd >= 0)
1445 return fd;
1446 most_interesting_errno = errno;
1448 prepare_alt_odb();
1449 for (alt = alt_odb_list; alt; alt = alt->next) {
1450 fill_sha1_path(alt->name, sha1);
1451 fd = git_open_noatime(alt->base);
1452 if (fd >= 0)
1453 return fd;
1454 if (most_interesting_errno == ENOENT)
1455 most_interesting_errno = errno;
1457 errno = most_interesting_errno;
1458 return -1;
1461 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1463 void *map;
1464 int fd;
1466 fd = open_sha1_file(sha1);
1467 map = NULL;
1468 if (fd >= 0) {
1469 struct stat st;
1471 if (!fstat(fd, &st)) {
1472 *size = xsize_t(st.st_size);
1473 if (!*size) {
1474 /* mmap() is forbidden on empty files */
1475 error("object file %s is empty", sha1_file_name(sha1));
1476 return NULL;
1478 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1480 close(fd);
1482 return map;
1485 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1486 unsigned long len, enum object_type *type, unsigned long *sizep)
1488 unsigned shift;
1489 unsigned long size, c;
1490 unsigned long used = 0;
1492 c = buf[used++];
1493 *type = (c >> 4) & 7;
1494 size = c & 15;
1495 shift = 4;
1496 while (c & 0x80) {
1497 if (len <= used || bitsizeof(long) <= shift) {
1498 error("bad object header");
1499 size = used = 0;
1500 break;
1502 c = buf[used++];
1503 size += (c & 0x7f) << shift;
1504 shift += 7;
1506 *sizep = size;
1507 return used;
1510 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1512 /* Get the data stream */
1513 memset(stream, 0, sizeof(*stream));
1514 stream->next_in = map;
1515 stream->avail_in = mapsize;
1516 stream->next_out = buffer;
1517 stream->avail_out = bufsiz;
1519 git_inflate_init(stream);
1520 return git_inflate(stream, 0);
1523 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1525 int bytes = strlen(buffer) + 1;
1526 unsigned char *buf = xmallocz(size);
1527 unsigned long n;
1528 int status = Z_OK;
1530 n = stream->total_out - bytes;
1531 if (n > size)
1532 n = size;
1533 memcpy(buf, (char *) buffer + bytes, n);
1534 bytes = n;
1535 if (bytes <= size) {
1537 * The above condition must be (bytes <= size), not
1538 * (bytes < size). In other words, even though we
1539 * expect no more output and set avail_out to zero,
1540 * the input zlib stream may have bytes that express
1541 * "this concludes the stream", and we *do* want to
1542 * eat that input.
1544 * Otherwise we would not be able to test that we
1545 * consumed all the input to reach the expected size;
1546 * we also want to check that zlib tells us that all
1547 * went well with status == Z_STREAM_END at the end.
1549 stream->next_out = buf + bytes;
1550 stream->avail_out = size - bytes;
1551 while (status == Z_OK)
1552 status = git_inflate(stream, Z_FINISH);
1554 if (status == Z_STREAM_END && !stream->avail_in) {
1555 git_inflate_end(stream);
1556 return buf;
1559 if (status < 0)
1560 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1561 else if (stream->avail_in)
1562 error("garbage at end of loose object '%s'",
1563 sha1_to_hex(sha1));
1564 free(buf);
1565 return NULL;
1569 * We used to just use "sscanf()", but that's actually way
1570 * too permissive for what we want to check. So do an anal
1571 * object header parse by hand.
1573 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1575 char type[10];
1576 int i;
1577 unsigned long size;
1580 * The type can be at most ten bytes (including the
1581 * terminating '\0' that we add), and is followed by
1582 * a space.
1584 i = 0;
1585 for (;;) {
1586 char c = *hdr++;
1587 if (c == ' ')
1588 break;
1589 type[i++] = c;
1590 if (i >= sizeof(type))
1591 return -1;
1593 type[i] = 0;
1596 * The length must follow immediately, and be in canonical
1597 * decimal format (ie "010" is not valid).
1599 size = *hdr++ - '0';
1600 if (size > 9)
1601 return -1;
1602 if (size) {
1603 for (;;) {
1604 unsigned long c = *hdr - '0';
1605 if (c > 9)
1606 break;
1607 hdr++;
1608 size = size * 10 + c;
1611 *sizep = size;
1614 * The length must be followed by a zero byte
1616 return *hdr ? -1 : type_from_string(type);
1619 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1621 int ret;
1622 git_zstream stream;
1623 char hdr[8192];
1625 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1626 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1627 return NULL;
1629 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1632 unsigned long get_size_from_delta(struct packed_git *p,
1633 struct pack_window **w_curs,
1634 off_t curpos)
1636 const unsigned char *data;
1637 unsigned char delta_head[20], *in;
1638 git_zstream stream;
1639 int st;
1641 memset(&stream, 0, sizeof(stream));
1642 stream.next_out = delta_head;
1643 stream.avail_out = sizeof(delta_head);
1645 git_inflate_init(&stream);
1646 do {
1647 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1648 stream.next_in = in;
1649 st = git_inflate(&stream, Z_FINISH);
1650 curpos += stream.next_in - in;
1651 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1652 stream.total_out < sizeof(delta_head));
1653 git_inflate_end(&stream);
1654 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1655 error("delta data unpack-initial failed");
1656 return 0;
1659 /* Examine the initial part of the delta to figure out
1660 * the result size.
1662 data = delta_head;
1664 /* ignore base size */
1665 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1667 /* Read the result size */
1668 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1671 static off_t get_delta_base(struct packed_git *p,
1672 struct pack_window **w_curs,
1673 off_t *curpos,
1674 enum object_type type,
1675 off_t delta_obj_offset)
1677 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1678 off_t base_offset;
1680 /* use_pack() assured us we have [base_info, base_info + 20)
1681 * as a range that we can look at without walking off the
1682 * end of the mapped window. Its actually the hash size
1683 * that is assured. An OFS_DELTA longer than the hash size
1684 * is stupid, as then a REF_DELTA would be smaller to store.
1686 if (type == OBJ_OFS_DELTA) {
1687 unsigned used = 0;
1688 unsigned char c = base_info[used++];
1689 base_offset = c & 127;
1690 while (c & 128) {
1691 base_offset += 1;
1692 if (!base_offset || MSB(base_offset, 7))
1693 return 0; /* overflow */
1694 c = base_info[used++];
1695 base_offset = (base_offset << 7) + (c & 127);
1697 base_offset = delta_obj_offset - base_offset;
1698 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1699 return 0; /* out of bound */
1700 *curpos += used;
1701 } else if (type == OBJ_REF_DELTA) {
1702 /* The base entry _must_ be in the same pack */
1703 base_offset = find_pack_entry_one(base_info, p);
1704 *curpos += 20;
1705 } else
1706 die("I am totally screwed");
1707 return base_offset;
1711 * Like get_delta_base above, but we return the sha1 instead of the pack
1712 * offset. This means it is cheaper for REF deltas (we do not have to do
1713 * the final object lookup), but more expensive for OFS deltas (we
1714 * have to load the revidx to convert the offset back into a sha1).
1716 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1717 struct pack_window **w_curs,
1718 off_t curpos,
1719 enum object_type type,
1720 off_t delta_obj_offset)
1722 if (type == OBJ_REF_DELTA) {
1723 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1724 return base;
1725 } else if (type == OBJ_OFS_DELTA) {
1726 struct revindex_entry *revidx;
1727 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1728 type, delta_obj_offset);
1730 if (!base_offset)
1731 return NULL;
1733 revidx = find_pack_revindex(p, base_offset);
1734 if (!revidx)
1735 return NULL;
1737 return nth_packed_object_sha1(p, revidx->nr);
1738 } else
1739 return NULL;
1742 int unpack_object_header(struct packed_git *p,
1743 struct pack_window **w_curs,
1744 off_t *curpos,
1745 unsigned long *sizep)
1747 unsigned char *base;
1748 unsigned long left;
1749 unsigned long used;
1750 enum object_type type;
1752 /* use_pack() assures us we have [base, base + 20) available
1753 * as a range that we can look at. (Its actually the hash
1754 * size that is assured.) With our object header encoding
1755 * the maximum deflated object size is 2^137, which is just
1756 * insane, so we know won't exceed what we have been given.
1758 base = use_pack(p, w_curs, *curpos, &left);
1759 used = unpack_object_header_buffer(base, left, &type, sizep);
1760 if (!used) {
1761 type = OBJ_BAD;
1762 } else
1763 *curpos += used;
1765 return type;
1768 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1770 int type;
1771 struct revindex_entry *revidx;
1772 const unsigned char *sha1;
1773 revidx = find_pack_revindex(p, obj_offset);
1774 if (!revidx)
1775 return OBJ_BAD;
1776 sha1 = nth_packed_object_sha1(p, revidx->nr);
1777 mark_bad_packed_object(p, sha1);
1778 type = sha1_object_info(sha1, NULL);
1779 if (type <= OBJ_NONE)
1780 return OBJ_BAD;
1781 return type;
1784 #define POI_STACK_PREALLOC 64
1786 static enum object_type packed_to_object_type(struct packed_git *p,
1787 off_t obj_offset,
1788 enum object_type type,
1789 struct pack_window **w_curs,
1790 off_t curpos)
1792 off_t small_poi_stack[POI_STACK_PREALLOC];
1793 off_t *poi_stack = small_poi_stack;
1794 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1796 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1797 off_t base_offset;
1798 unsigned long size;
1799 /* Push the object we're going to leave behind */
1800 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1801 poi_stack_alloc = alloc_nr(poi_stack_nr);
1802 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1803 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1804 } else {
1805 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1807 poi_stack[poi_stack_nr++] = obj_offset;
1808 /* If parsing the base offset fails, just unwind */
1809 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1810 if (!base_offset)
1811 goto unwind;
1812 curpos = obj_offset = base_offset;
1813 type = unpack_object_header(p, w_curs, &curpos, &size);
1814 if (type <= OBJ_NONE) {
1815 /* If getting the base itself fails, we first
1816 * retry the base, otherwise unwind */
1817 type = retry_bad_packed_offset(p, base_offset);
1818 if (type > OBJ_NONE)
1819 goto out;
1820 goto unwind;
1824 switch (type) {
1825 case OBJ_BAD:
1826 case OBJ_COMMIT:
1827 case OBJ_TREE:
1828 case OBJ_BLOB:
1829 case OBJ_TAG:
1830 break;
1831 default:
1832 error("unknown object type %i at offset %"PRIuMAX" in %s",
1833 type, (uintmax_t)obj_offset, p->pack_name);
1834 type = OBJ_BAD;
1837 out:
1838 if (poi_stack != small_poi_stack)
1839 free(poi_stack);
1840 return type;
1842 unwind:
1843 while (poi_stack_nr) {
1844 obj_offset = poi_stack[--poi_stack_nr];
1845 type = retry_bad_packed_offset(p, obj_offset);
1846 if (type > OBJ_NONE)
1847 goto out;
1849 type = OBJ_BAD;
1850 goto out;
1853 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1854 struct object_info *oi)
1856 struct pack_window *w_curs = NULL;
1857 unsigned long size;
1858 off_t curpos = obj_offset;
1859 enum object_type type;
1862 * We always get the representation type, but only convert it to
1863 * a "real" type later if the caller is interested.
1865 type = unpack_object_header(p, &w_curs, &curpos, &size);
1867 if (oi->sizep) {
1868 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1869 off_t tmp_pos = curpos;
1870 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1871 type, obj_offset);
1872 if (!base_offset) {
1873 type = OBJ_BAD;
1874 goto out;
1876 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1877 if (*oi->sizep == 0) {
1878 type = OBJ_BAD;
1879 goto out;
1881 } else {
1882 *oi->sizep = size;
1886 if (oi->disk_sizep) {
1887 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1888 *oi->disk_sizep = revidx[1].offset - obj_offset;
1891 if (oi->typep) {
1892 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1893 if (*oi->typep < 0) {
1894 type = OBJ_BAD;
1895 goto out;
1899 if (oi->delta_base_sha1) {
1900 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1901 const unsigned char *base;
1903 base = get_delta_base_sha1(p, &w_curs, curpos,
1904 type, obj_offset);
1905 if (!base) {
1906 type = OBJ_BAD;
1907 goto out;
1910 hashcpy(oi->delta_base_sha1, base);
1911 } else
1912 hashclr(oi->delta_base_sha1);
1915 out:
1916 unuse_pack(&w_curs);
1917 return type;
1920 static void *unpack_compressed_entry(struct packed_git *p,
1921 struct pack_window **w_curs,
1922 off_t curpos,
1923 unsigned long size)
1925 int st;
1926 git_zstream stream;
1927 unsigned char *buffer, *in;
1929 buffer = xmallocz(size);
1930 memset(&stream, 0, sizeof(stream));
1931 stream.next_out = buffer;
1932 stream.avail_out = size + 1;
1934 git_inflate_init(&stream);
1935 do {
1936 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1937 stream.next_in = in;
1938 st = git_inflate(&stream, Z_FINISH);
1939 if (!stream.avail_out)
1940 break; /* the payload is larger than it should be */
1941 curpos += stream.next_in - in;
1942 } while (st == Z_OK || st == Z_BUF_ERROR);
1943 git_inflate_end(&stream);
1944 if ((st != Z_STREAM_END) || stream.total_out != size) {
1945 free(buffer);
1946 return NULL;
1949 return buffer;
1952 #define MAX_DELTA_CACHE (256)
1954 static size_t delta_base_cached;
1956 static struct delta_base_cache_lru_list {
1957 struct delta_base_cache_lru_list *prev;
1958 struct delta_base_cache_lru_list *next;
1959 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1961 static struct delta_base_cache_entry {
1962 struct delta_base_cache_lru_list lru;
1963 void *data;
1964 struct packed_git *p;
1965 off_t base_offset;
1966 unsigned long size;
1967 enum object_type type;
1968 } delta_base_cache[MAX_DELTA_CACHE];
1970 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1972 unsigned long hash;
1974 hash = (unsigned long)p + (unsigned long)base_offset;
1975 hash += (hash >> 8) + (hash >> 16);
1976 return hash % MAX_DELTA_CACHE;
1979 static struct delta_base_cache_entry *
1980 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1982 unsigned long hash = pack_entry_hash(p, base_offset);
1983 return delta_base_cache + hash;
1986 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1987 struct packed_git *p, off_t base_offset)
1989 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1992 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1994 struct delta_base_cache_entry *ent;
1995 ent = get_delta_base_cache_entry(p, base_offset);
1996 return eq_delta_base_cache_entry(ent, p, base_offset);
1999 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2001 ent->data = NULL;
2002 ent->lru.next->prev = ent->lru.prev;
2003 ent->lru.prev->next = ent->lru.next;
2004 delta_base_cached -= ent->size;
2007 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2008 unsigned long *base_size, enum object_type *type, int keep_cache)
2010 struct delta_base_cache_entry *ent;
2011 void *ret;
2013 ent = get_delta_base_cache_entry(p, base_offset);
2015 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2016 return unpack_entry(p, base_offset, type, base_size);
2018 ret = ent->data;
2020 if (!keep_cache)
2021 clear_delta_base_cache_entry(ent);
2022 else
2023 ret = xmemdupz(ent->data, ent->size);
2024 *type = ent->type;
2025 *base_size = ent->size;
2026 return ret;
2029 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2031 if (ent->data) {
2032 free(ent->data);
2033 ent->data = NULL;
2034 ent->lru.next->prev = ent->lru.prev;
2035 ent->lru.prev->next = ent->lru.next;
2036 delta_base_cached -= ent->size;
2040 void clear_delta_base_cache(void)
2042 unsigned long p;
2043 for (p = 0; p < MAX_DELTA_CACHE; p++)
2044 release_delta_base_cache(&delta_base_cache[p]);
2047 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2048 void *base, unsigned long base_size, enum object_type type)
2050 unsigned long hash = pack_entry_hash(p, base_offset);
2051 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2052 struct delta_base_cache_lru_list *lru;
2054 release_delta_base_cache(ent);
2055 delta_base_cached += base_size;
2057 for (lru = delta_base_cache_lru.next;
2058 delta_base_cached > delta_base_cache_limit
2059 && lru != &delta_base_cache_lru;
2060 lru = lru->next) {
2061 struct delta_base_cache_entry *f = (void *)lru;
2062 if (f->type == OBJ_BLOB)
2063 release_delta_base_cache(f);
2065 for (lru = delta_base_cache_lru.next;
2066 delta_base_cached > delta_base_cache_limit
2067 && lru != &delta_base_cache_lru;
2068 lru = lru->next) {
2069 struct delta_base_cache_entry *f = (void *)lru;
2070 release_delta_base_cache(f);
2073 ent->p = p;
2074 ent->base_offset = base_offset;
2075 ent->type = type;
2076 ent->data = base;
2077 ent->size = base_size;
2078 ent->lru.next = &delta_base_cache_lru;
2079 ent->lru.prev = delta_base_cache_lru.prev;
2080 delta_base_cache_lru.prev->next = &ent->lru;
2081 delta_base_cache_lru.prev = &ent->lru;
2084 static void *read_object(const unsigned char *sha1, enum object_type *type,
2085 unsigned long *size);
2087 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2089 static FILE *log_file;
2091 if (!log_pack_access)
2092 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2093 if (!log_pack_access)
2094 log_pack_access = no_log_pack_access;
2095 if (log_pack_access == no_log_pack_access)
2096 return;
2098 if (!log_file) {
2099 log_file = fopen(log_pack_access, "w");
2100 if (!log_file) {
2101 error("cannot open pack access log '%s' for writing: %s",
2102 log_pack_access, strerror(errno));
2103 log_pack_access = no_log_pack_access;
2104 return;
2107 fprintf(log_file, "%s %"PRIuMAX"\n",
2108 p->pack_name, (uintmax_t)obj_offset);
2109 fflush(log_file);
2112 int do_check_packed_object_crc;
2114 #define UNPACK_ENTRY_STACK_PREALLOC 64
2115 struct unpack_entry_stack_ent {
2116 off_t obj_offset;
2117 off_t curpos;
2118 unsigned long size;
2121 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2122 enum object_type *final_type, unsigned long *final_size)
2124 struct pack_window *w_curs = NULL;
2125 off_t curpos = obj_offset;
2126 void *data = NULL;
2127 unsigned long size;
2128 enum object_type type;
2129 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2130 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2131 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2132 int base_from_cache = 0;
2134 if (log_pack_access != no_log_pack_access)
2135 write_pack_access_log(p, obj_offset);
2137 /* PHASE 1: drill down to the innermost base object */
2138 for (;;) {
2139 off_t base_offset;
2140 int i;
2141 struct delta_base_cache_entry *ent;
2143 ent = get_delta_base_cache_entry(p, curpos);
2144 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2145 type = ent->type;
2146 data = ent->data;
2147 size = ent->size;
2148 clear_delta_base_cache_entry(ent);
2149 base_from_cache = 1;
2150 break;
2153 if (do_check_packed_object_crc && p->index_version > 1) {
2154 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2155 unsigned long len = revidx[1].offset - obj_offset;
2156 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2157 const unsigned char *sha1 =
2158 nth_packed_object_sha1(p, revidx->nr);
2159 error("bad packed object CRC for %s",
2160 sha1_to_hex(sha1));
2161 mark_bad_packed_object(p, sha1);
2162 unuse_pack(&w_curs);
2163 return NULL;
2167 type = unpack_object_header(p, &w_curs, &curpos, &size);
2168 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2169 break;
2171 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2172 if (!base_offset) {
2173 error("failed to validate delta base reference "
2174 "at offset %"PRIuMAX" from %s",
2175 (uintmax_t)curpos, p->pack_name);
2176 /* bail to phase 2, in hopes of recovery */
2177 data = NULL;
2178 break;
2181 /* push object, proceed to base */
2182 if (delta_stack_nr >= delta_stack_alloc
2183 && delta_stack == small_delta_stack) {
2184 delta_stack_alloc = alloc_nr(delta_stack_nr);
2185 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2186 memcpy(delta_stack, small_delta_stack,
2187 sizeof(*delta_stack)*delta_stack_nr);
2188 } else {
2189 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2191 i = delta_stack_nr++;
2192 delta_stack[i].obj_offset = obj_offset;
2193 delta_stack[i].curpos = curpos;
2194 delta_stack[i].size = size;
2196 curpos = obj_offset = base_offset;
2199 /* PHASE 2: handle the base */
2200 switch (type) {
2201 case OBJ_OFS_DELTA:
2202 case OBJ_REF_DELTA:
2203 if (data)
2204 die("BUG in unpack_entry: left loop at a valid delta");
2205 break;
2206 case OBJ_COMMIT:
2207 case OBJ_TREE:
2208 case OBJ_BLOB:
2209 case OBJ_TAG:
2210 if (!base_from_cache)
2211 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2212 break;
2213 default:
2214 data = NULL;
2215 error("unknown object type %i at offset %"PRIuMAX" in %s",
2216 type, (uintmax_t)obj_offset, p->pack_name);
2219 /* PHASE 3: apply deltas in order */
2221 /* invariants:
2222 * 'data' holds the base data, or NULL if there was corruption
2224 while (delta_stack_nr) {
2225 void *delta_data;
2226 void *base = data;
2227 unsigned long delta_size, base_size = size;
2228 int i;
2230 data = NULL;
2232 if (base)
2233 add_delta_base_cache(p, obj_offset, base, base_size, type);
2235 if (!base) {
2237 * We're probably in deep shit, but let's try to fetch
2238 * the required base anyway from another pack or loose.
2239 * This is costly but should happen only in the presence
2240 * of a corrupted pack, and is better than failing outright.
2242 struct revindex_entry *revidx;
2243 const unsigned char *base_sha1;
2244 revidx = find_pack_revindex(p, obj_offset);
2245 if (revidx) {
2246 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2247 error("failed to read delta base object %s"
2248 " at offset %"PRIuMAX" from %s",
2249 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2250 p->pack_name);
2251 mark_bad_packed_object(p, base_sha1);
2252 base = read_object(base_sha1, &type, &base_size);
2256 i = --delta_stack_nr;
2257 obj_offset = delta_stack[i].obj_offset;
2258 curpos = delta_stack[i].curpos;
2259 delta_size = delta_stack[i].size;
2261 if (!base)
2262 continue;
2264 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2266 if (!delta_data) {
2267 error("failed to unpack compressed delta "
2268 "at offset %"PRIuMAX" from %s",
2269 (uintmax_t)curpos, p->pack_name);
2270 data = NULL;
2271 continue;
2274 data = patch_delta(base, base_size,
2275 delta_data, delta_size,
2276 &size);
2279 * We could not apply the delta; warn the user, but keep going.
2280 * Our failure will be noticed either in the next iteration of
2281 * the loop, or if this is the final delta, in the caller when
2282 * we return NULL. Those code paths will take care of making
2283 * a more explicit warning and retrying with another copy of
2284 * the object.
2286 if (!data)
2287 error("failed to apply delta");
2289 free(delta_data);
2292 *final_type = type;
2293 *final_size = size;
2295 unuse_pack(&w_curs);
2297 if (delta_stack != small_delta_stack)
2298 free(delta_stack);
2300 return data;
2303 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2304 uint32_t n)
2306 const unsigned char *index = p->index_data;
2307 if (!index) {
2308 if (open_pack_index(p))
2309 return NULL;
2310 index = p->index_data;
2312 if (n >= p->num_objects)
2313 return NULL;
2314 index += 4 * 256;
2315 if (p->index_version == 1) {
2316 return index + 24 * n + 4;
2317 } else {
2318 index += 8;
2319 return index + 20 * n;
2323 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2325 const unsigned char *index = p->index_data;
2326 index += 4 * 256;
2327 if (p->index_version == 1) {
2328 return ntohl(*((uint32_t *)(index + 24 * n)));
2329 } else {
2330 uint32_t off;
2331 index += 8 + p->num_objects * (20 + 4);
2332 off = ntohl(*((uint32_t *)(index + 4 * n)));
2333 if (!(off & 0x80000000))
2334 return off;
2335 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2336 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2337 ntohl(*((uint32_t *)(index + 4)));
2341 off_t find_pack_entry_one(const unsigned char *sha1,
2342 struct packed_git *p)
2344 const uint32_t *level1_ofs = p->index_data;
2345 const unsigned char *index = p->index_data;
2346 unsigned hi, lo, stride;
2347 static int use_lookup = -1;
2348 static int debug_lookup = -1;
2350 if (debug_lookup < 0)
2351 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2353 if (!index) {
2354 if (open_pack_index(p))
2355 return 0;
2356 level1_ofs = p->index_data;
2357 index = p->index_data;
2359 if (p->index_version > 1) {
2360 level1_ofs += 2;
2361 index += 8;
2363 index += 4 * 256;
2364 hi = ntohl(level1_ofs[*sha1]);
2365 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2366 if (p->index_version > 1) {
2367 stride = 20;
2368 } else {
2369 stride = 24;
2370 index += 4;
2373 if (debug_lookup)
2374 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2375 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2377 if (use_lookup < 0)
2378 use_lookup = !!getenv("GIT_USE_LOOKUP");
2379 if (use_lookup) {
2380 int pos = sha1_entry_pos(index, stride, 0,
2381 lo, hi, p->num_objects, sha1);
2382 if (pos < 0)
2383 return 0;
2384 return nth_packed_object_offset(p, pos);
2387 do {
2388 unsigned mi = (lo + hi) / 2;
2389 int cmp = hashcmp(index + mi * stride, sha1);
2391 if (debug_lookup)
2392 printf("lo %u hi %u rg %u mi %u\n",
2393 lo, hi, hi - lo, mi);
2394 if (!cmp)
2395 return nth_packed_object_offset(p, mi);
2396 if (cmp > 0)
2397 hi = mi;
2398 else
2399 lo = mi+1;
2400 } while (lo < hi);
2401 return 0;
2404 int is_pack_valid(struct packed_git *p)
2406 /* An already open pack is known to be valid. */
2407 if (p->pack_fd != -1)
2408 return 1;
2410 /* If the pack has one window completely covering the
2411 * file size, the pack is known to be valid even if
2412 * the descriptor is not currently open.
2414 if (p->windows) {
2415 struct pack_window *w = p->windows;
2417 if (!w->offset && w->len == p->pack_size)
2418 return 1;
2421 /* Force the pack to open to prove its valid. */
2422 return !open_packed_git(p);
2425 static int fill_pack_entry(const unsigned char *sha1,
2426 struct pack_entry *e,
2427 struct packed_git *p)
2429 off_t offset;
2431 if (p->num_bad_objects) {
2432 unsigned i;
2433 for (i = 0; i < p->num_bad_objects; i++)
2434 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2435 return 0;
2438 offset = find_pack_entry_one(sha1, p);
2439 if (!offset)
2440 return 0;
2443 * We are about to tell the caller where they can locate the
2444 * requested object. We better make sure the packfile is
2445 * still here and can be accessed before supplying that
2446 * answer, as it may have been deleted since the index was
2447 * loaded!
2449 if (!is_pack_valid(p)) {
2450 warning("packfile %s cannot be accessed", p->pack_name);
2451 return 0;
2453 e->offset = offset;
2454 e->p = p;
2455 hashcpy(e->sha1, sha1);
2456 return 1;
2460 * Iff a pack file contains the object named by sha1, return true and
2461 * store its location to e.
2463 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2465 struct packed_git *p;
2467 prepare_packed_git();
2468 if (!packed_git)
2469 return 0;
2471 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2472 return 1;
2474 for (p = packed_git; p; p = p->next) {
2475 if (p == last_found_pack)
2476 continue; /* we already checked this one */
2478 if (fill_pack_entry(sha1, e, p)) {
2479 last_found_pack = p;
2480 return 1;
2483 return 0;
2486 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2487 struct packed_git *packs)
2489 struct packed_git *p;
2491 for (p = packs; p; p = p->next) {
2492 if (find_pack_entry_one(sha1, p))
2493 return p;
2495 return NULL;
2499 static int sha1_loose_object_info(const unsigned char *sha1,
2500 struct object_info *oi)
2502 int status;
2503 unsigned long mapsize, size;
2504 void *map;
2505 git_zstream stream;
2506 char hdr[32];
2508 if (oi->delta_base_sha1)
2509 hashclr(oi->delta_base_sha1);
2512 * If we don't care about type or size, then we don't
2513 * need to look inside the object at all. Note that we
2514 * do not optimize out the stat call, even if the
2515 * caller doesn't care about the disk-size, since our
2516 * return value implicitly indicates whether the
2517 * object even exists.
2519 if (!oi->typep && !oi->sizep) {
2520 struct stat st;
2521 if (stat_sha1_file(sha1, &st) < 0)
2522 return -1;
2523 if (oi->disk_sizep)
2524 *oi->disk_sizep = st.st_size;
2525 return 0;
2528 map = map_sha1_file(sha1, &mapsize);
2529 if (!map)
2530 return -1;
2531 if (oi->disk_sizep)
2532 *oi->disk_sizep = mapsize;
2533 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2534 status = error("unable to unpack %s header",
2535 sha1_to_hex(sha1));
2536 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2537 status = error("unable to parse %s header", sha1_to_hex(sha1));
2538 else if (oi->sizep)
2539 *oi->sizep = size;
2540 git_inflate_end(&stream);
2541 munmap(map, mapsize);
2542 if (oi->typep)
2543 *oi->typep = status;
2544 return 0;
2547 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2549 struct cached_object *co;
2550 struct pack_entry e;
2551 int rtype;
2552 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2554 co = find_cached_object(real);
2555 if (co) {
2556 if (oi->typep)
2557 *(oi->typep) = co->type;
2558 if (oi->sizep)
2559 *(oi->sizep) = co->size;
2560 if (oi->disk_sizep)
2561 *(oi->disk_sizep) = 0;
2562 if (oi->delta_base_sha1)
2563 hashclr(oi->delta_base_sha1);
2564 oi->whence = OI_CACHED;
2565 return 0;
2568 if (!find_pack_entry(real, &e)) {
2569 /* Most likely it's a loose object. */
2570 if (!sha1_loose_object_info(real, oi)) {
2571 oi->whence = OI_LOOSE;
2572 return 0;
2575 /* Not a loose object; someone else may have just packed it. */
2576 reprepare_packed_git();
2577 if (!find_pack_entry(real, &e))
2578 return -1;
2581 rtype = packed_object_info(e.p, e.offset, oi);
2582 if (rtype < 0) {
2583 mark_bad_packed_object(e.p, real);
2584 return sha1_object_info_extended(real, oi, 0);
2585 } else if (in_delta_base_cache(e.p, e.offset)) {
2586 oi->whence = OI_DBCACHED;
2587 } else {
2588 oi->whence = OI_PACKED;
2589 oi->u.packed.offset = e.offset;
2590 oi->u.packed.pack = e.p;
2591 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2592 rtype == OBJ_OFS_DELTA);
2595 return 0;
2598 /* returns enum object_type or negative */
2599 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2601 enum object_type type;
2602 struct object_info oi = {NULL};
2604 oi.typep = &type;
2605 oi.sizep = sizep;
2606 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2607 return -1;
2608 return type;
2611 static void *read_packed_sha1(const unsigned char *sha1,
2612 enum object_type *type, unsigned long *size)
2614 struct pack_entry e;
2615 void *data;
2617 if (!find_pack_entry(sha1, &e))
2618 return NULL;
2619 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2620 if (!data) {
2622 * We're probably in deep shit, but let's try to fetch
2623 * the required object anyway from another pack or loose.
2624 * This should happen only in the presence of a corrupted
2625 * pack, and is better than failing outright.
2627 error("failed to read object %s at offset %"PRIuMAX" from %s",
2628 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2629 mark_bad_packed_object(e.p, sha1);
2630 data = read_object(sha1, type, size);
2632 return data;
2635 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2636 unsigned char *sha1)
2638 struct cached_object *co;
2640 hash_sha1_file(buf, len, typename(type), sha1);
2641 if (has_sha1_file(sha1) || find_cached_object(sha1))
2642 return 0;
2643 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2644 co = &cached_objects[cached_object_nr++];
2645 co->size = len;
2646 co->type = type;
2647 co->buf = xmalloc(len);
2648 memcpy(co->buf, buf, len);
2649 hashcpy(co->sha1, sha1);
2650 return 0;
2653 static void *read_object(const unsigned char *sha1, enum object_type *type,
2654 unsigned long *size)
2656 unsigned long mapsize;
2657 void *map, *buf;
2658 struct cached_object *co;
2660 co = find_cached_object(sha1);
2661 if (co) {
2662 *type = co->type;
2663 *size = co->size;
2664 return xmemdupz(co->buf, co->size);
2667 buf = read_packed_sha1(sha1, type, size);
2668 if (buf)
2669 return buf;
2670 map = map_sha1_file(sha1, &mapsize);
2671 if (map) {
2672 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2673 munmap(map, mapsize);
2674 return buf;
2676 reprepare_packed_git();
2677 return read_packed_sha1(sha1, type, size);
2681 * This function dies on corrupt objects; the callers who want to
2682 * deal with them should arrange to call read_object() and give error
2683 * messages themselves.
2685 void *read_sha1_file_extended(const unsigned char *sha1,
2686 enum object_type *type,
2687 unsigned long *size,
2688 unsigned flag)
2690 void *data;
2691 const struct packed_git *p;
2692 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2694 errno = 0;
2695 data = read_object(repl, type, size);
2696 if (data)
2697 return data;
2699 if (errno && errno != ENOENT)
2700 die_errno("failed to read object %s", sha1_to_hex(sha1));
2702 /* die if we replaced an object with one that does not exist */
2703 if (repl != sha1)
2704 die("replacement %s not found for %s",
2705 sha1_to_hex(repl), sha1_to_hex(sha1));
2707 if (has_loose_object(repl)) {
2708 const char *path = sha1_file_name(sha1);
2710 die("loose object %s (stored in %s) is corrupt",
2711 sha1_to_hex(repl), path);
2714 if ((p = has_packed_and_bad(repl)) != NULL)
2715 die("packed object %s (stored in %s) is corrupt",
2716 sha1_to_hex(repl), p->pack_name);
2718 return NULL;
2721 void *read_object_with_reference(const unsigned char *sha1,
2722 const char *required_type_name,
2723 unsigned long *size,
2724 unsigned char *actual_sha1_return)
2726 enum object_type type, required_type;
2727 void *buffer;
2728 unsigned long isize;
2729 unsigned char actual_sha1[20];
2731 required_type = type_from_string(required_type_name);
2732 hashcpy(actual_sha1, sha1);
2733 while (1) {
2734 int ref_length = -1;
2735 const char *ref_type = NULL;
2737 buffer = read_sha1_file(actual_sha1, &type, &isize);
2738 if (!buffer)
2739 return NULL;
2740 if (type == required_type) {
2741 *size = isize;
2742 if (actual_sha1_return)
2743 hashcpy(actual_sha1_return, actual_sha1);
2744 return buffer;
2746 /* Handle references */
2747 else if (type == OBJ_COMMIT)
2748 ref_type = "tree ";
2749 else if (type == OBJ_TAG)
2750 ref_type = "object ";
2751 else {
2752 free(buffer);
2753 return NULL;
2755 ref_length = strlen(ref_type);
2757 if (ref_length + 40 > isize ||
2758 memcmp(buffer, ref_type, ref_length) ||
2759 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2760 free(buffer);
2761 return NULL;
2763 free(buffer);
2764 /* Now we have the ID of the referred-to object in
2765 * actual_sha1. Check again. */
2769 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2770 const char *type, unsigned char *sha1,
2771 char *hdr, int *hdrlen)
2773 git_SHA_CTX c;
2775 /* Generate the header */
2776 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2778 /* Sha1.. */
2779 git_SHA1_Init(&c);
2780 git_SHA1_Update(&c, hdr, *hdrlen);
2781 git_SHA1_Update(&c, buf, len);
2782 git_SHA1_Final(sha1, &c);
2786 * Move the just written object into its final resting place.
2787 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2788 * "moving" is only a part of what it does, when no patch between
2789 * master to pu changes the call sites of this function.
2791 int move_temp_to_file(const char *tmpfile, const char *filename)
2793 int ret = 0;
2795 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2796 goto try_rename;
2797 else if (link(tmpfile, filename))
2798 ret = errno;
2801 * Coda hack - coda doesn't like cross-directory links,
2802 * so we fall back to a rename, which will mean that it
2803 * won't be able to check collisions, but that's not a
2804 * big deal.
2806 * The same holds for FAT formatted media.
2808 * When this succeeds, we just return. We have nothing
2809 * left to unlink.
2811 if (ret && ret != EEXIST) {
2812 try_rename:
2813 if (!rename(tmpfile, filename))
2814 goto out;
2815 ret = errno;
2817 unlink_or_warn(tmpfile);
2818 if (ret) {
2819 if (ret != EEXIST) {
2820 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2822 /* FIXME!!! Collision check here ? */
2825 out:
2826 if (adjust_shared_perm(filename))
2827 return error("unable to set permission to '%s'", filename);
2828 return 0;
2831 static int write_buffer(int fd, const void *buf, size_t len)
2833 if (write_in_full(fd, buf, len) < 0)
2834 return error("file write error (%s)", strerror(errno));
2835 return 0;
2838 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2839 unsigned char *sha1)
2841 char hdr[32];
2842 int hdrlen;
2843 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2844 return 0;
2847 /* Finalize a file on disk, and close it. */
2848 static void close_sha1_file(int fd)
2850 if (fsync_object_files)
2851 fsync_or_die(fd, "sha1 file");
2852 if (close(fd) != 0)
2853 die_errno("error when closing sha1 file");
2856 /* Size of directory component, including the ending '/' */
2857 static inline int directory_size(const char *filename)
2859 const char *s = strrchr(filename, '/');
2860 if (!s)
2861 return 0;
2862 return s - filename + 1;
2866 * This creates a temporary file in the same directory as the final
2867 * 'filename'
2869 * We want to avoid cross-directory filename renames, because those
2870 * can have problems on various filesystems (FAT, NFS, Coda).
2872 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2874 int fd, dirlen = directory_size(filename);
2876 if (dirlen + 20 > bufsiz) {
2877 errno = ENAMETOOLONG;
2878 return -1;
2880 memcpy(buffer, filename, dirlen);
2881 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2882 fd = git_mkstemp_mode(buffer, 0444);
2883 if (fd < 0 && dirlen && errno == ENOENT) {
2884 /* Make sure the directory exists */
2885 memcpy(buffer, filename, dirlen);
2886 buffer[dirlen-1] = 0;
2887 if (mkdir(buffer, 0777) && errno != EEXIST)
2888 return -1;
2889 if (adjust_shared_perm(buffer))
2890 return -1;
2892 /* Try again */
2893 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2894 fd = git_mkstemp_mode(buffer, 0444);
2896 return fd;
2899 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2900 const void *buf, unsigned long len, time_t mtime)
2902 int fd, ret;
2903 unsigned char compressed[4096];
2904 git_zstream stream;
2905 git_SHA_CTX c;
2906 unsigned char parano_sha1[20];
2907 static char tmp_file[PATH_MAX];
2908 const char *filename = sha1_file_name(sha1);
2910 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2911 if (fd < 0) {
2912 if (errno == EACCES)
2913 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2914 else
2915 return error("unable to create temporary file: %s", strerror(errno));
2918 /* Set it up */
2919 memset(&stream, 0, sizeof(stream));
2920 git_deflate_init(&stream, zlib_compression_level);
2921 stream.next_out = compressed;
2922 stream.avail_out = sizeof(compressed);
2923 git_SHA1_Init(&c);
2925 /* First header.. */
2926 stream.next_in = (unsigned char *)hdr;
2927 stream.avail_in = hdrlen;
2928 while (git_deflate(&stream, 0) == Z_OK)
2929 ; /* nothing */
2930 git_SHA1_Update(&c, hdr, hdrlen);
2932 /* Then the data itself.. */
2933 stream.next_in = (void *)buf;
2934 stream.avail_in = len;
2935 do {
2936 unsigned char *in0 = stream.next_in;
2937 ret = git_deflate(&stream, Z_FINISH);
2938 git_SHA1_Update(&c, in0, stream.next_in - in0);
2939 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2940 die("unable to write sha1 file");
2941 stream.next_out = compressed;
2942 stream.avail_out = sizeof(compressed);
2943 } while (ret == Z_OK);
2945 if (ret != Z_STREAM_END)
2946 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2947 ret = git_deflate_end_gently(&stream);
2948 if (ret != Z_OK)
2949 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2950 git_SHA1_Final(parano_sha1, &c);
2951 if (hashcmp(sha1, parano_sha1) != 0)
2952 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2954 close_sha1_file(fd);
2956 if (mtime) {
2957 struct utimbuf utb;
2958 utb.actime = mtime;
2959 utb.modtime = mtime;
2960 if (utime(tmp_file, &utb) < 0)
2961 warning("failed utime() on %s: %s",
2962 tmp_file, strerror(errno));
2965 return move_temp_to_file(tmp_file, filename);
2968 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2970 unsigned char sha1[20];
2971 char hdr[32];
2972 int hdrlen;
2974 /* Normally if we have it in the pack then we do not bother writing
2975 * it out into .git/objects/??/?{38} file.
2977 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2978 if (returnsha1)
2979 hashcpy(returnsha1, sha1);
2980 if (has_sha1_file(sha1))
2981 return 0;
2982 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2985 int force_object_loose(const unsigned char *sha1, time_t mtime)
2987 void *buf;
2988 unsigned long len;
2989 enum object_type type;
2990 char hdr[32];
2991 int hdrlen;
2992 int ret;
2994 if (has_loose_object(sha1))
2995 return 0;
2996 buf = read_packed_sha1(sha1, &type, &len);
2997 if (!buf)
2998 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2999 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3000 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3001 free(buf);
3003 return ret;
3006 int has_pack_index(const unsigned char *sha1)
3008 struct stat st;
3009 if (stat(sha1_pack_index_name(sha1), &st))
3010 return 0;
3011 return 1;
3014 int has_sha1_pack(const unsigned char *sha1)
3016 struct pack_entry e;
3017 return find_pack_entry(sha1, &e);
3020 int has_sha1_file(const unsigned char *sha1)
3022 struct pack_entry e;
3024 if (find_pack_entry(sha1, &e))
3025 return 1;
3026 if (has_loose_object(sha1))
3027 return 1;
3028 reprepare_packed_git();
3029 return find_pack_entry(sha1, &e);
3032 static void check_tree(const void *buf, size_t size)
3034 struct tree_desc desc;
3035 struct name_entry entry;
3037 init_tree_desc(&desc, buf, size);
3038 while (tree_entry(&desc, &entry))
3039 /* do nothing
3040 * tree_entry() will die() on malformed entries */
3044 static void check_commit(const void *buf, size_t size)
3046 struct commit c;
3047 memset(&c, 0, sizeof(c));
3048 if (parse_commit_buffer(&c, buf, size))
3049 die("corrupt commit");
3052 static void check_tag(const void *buf, size_t size)
3054 struct tag t;
3055 memset(&t, 0, sizeof(t));
3056 if (parse_tag_buffer(&t, buf, size))
3057 die("corrupt tag");
3060 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3061 enum object_type type,
3062 const char *path, unsigned flags)
3064 int ret, re_allocated = 0;
3065 int write_object = flags & HASH_WRITE_OBJECT;
3067 if (!type)
3068 type = OBJ_BLOB;
3071 * Convert blobs to git internal format
3073 if ((type == OBJ_BLOB) && path) {
3074 struct strbuf nbuf = STRBUF_INIT;
3075 if (convert_to_git(path, buf, size, &nbuf,
3076 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3077 buf = strbuf_detach(&nbuf, &size);
3078 re_allocated = 1;
3081 if (flags & HASH_FORMAT_CHECK) {
3082 if (type == OBJ_TREE)
3083 check_tree(buf, size);
3084 if (type == OBJ_COMMIT)
3085 check_commit(buf, size);
3086 if (type == OBJ_TAG)
3087 check_tag(buf, size);
3090 if (write_object)
3091 ret = write_sha1_file(buf, size, typename(type), sha1);
3092 else
3093 ret = hash_sha1_file(buf, size, typename(type), sha1);
3094 if (re_allocated)
3095 free(buf);
3096 return ret;
3099 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3100 const char *path, unsigned flags)
3102 struct strbuf sbuf = STRBUF_INIT;
3103 int ret;
3105 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3106 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3107 else
3108 ret = -1;
3109 strbuf_release(&sbuf);
3110 return ret;
3113 #define SMALL_FILE_SIZE (32*1024)
3115 static int index_core(unsigned char *sha1, int fd, size_t size,
3116 enum object_type type, const char *path,
3117 unsigned flags)
3119 int ret;
3121 if (!size) {
3122 ret = index_mem(sha1, NULL, size, type, path, flags);
3123 } else if (size <= SMALL_FILE_SIZE) {
3124 char *buf = xmalloc(size);
3125 if (size == read_in_full(fd, buf, size))
3126 ret = index_mem(sha1, buf, size, type, path, flags);
3127 else
3128 ret = error("short read %s", strerror(errno));
3129 free(buf);
3130 } else {
3131 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3132 ret = index_mem(sha1, buf, size, type, path, flags);
3133 munmap(buf, size);
3135 return ret;
3139 * This creates one packfile per large blob unless bulk-checkin
3140 * machinery is "plugged".
3142 * This also bypasses the usual "convert-to-git" dance, and that is on
3143 * purpose. We could write a streaming version of the converting
3144 * functions and insert that before feeding the data to fast-import
3145 * (or equivalent in-core API described above). However, that is
3146 * somewhat complicated, as we do not know the size of the filter
3147 * result, which we need to know beforehand when writing a git object.
3148 * Since the primary motivation for trying to stream from the working
3149 * tree file and to avoid mmaping it in core is to deal with large
3150 * binary blobs, they generally do not want to get any conversion, and
3151 * callers should avoid this code path when filters are requested.
3153 static int index_stream(unsigned char *sha1, int fd, size_t size,
3154 enum object_type type, const char *path,
3155 unsigned flags)
3157 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3160 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3161 enum object_type type, const char *path, unsigned flags)
3163 int ret;
3164 size_t size = xsize_t(st->st_size);
3166 if (!S_ISREG(st->st_mode))
3167 ret = index_pipe(sha1, fd, type, path, flags);
3168 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3169 (path && would_convert_to_git(path, NULL, 0, 0)))
3170 ret = index_core(sha1, fd, size, type, path, flags);
3171 else
3172 ret = index_stream(sha1, fd, size, type, path, flags);
3173 close(fd);
3174 return ret;
3177 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3179 int fd;
3180 struct strbuf sb = STRBUF_INIT;
3182 switch (st->st_mode & S_IFMT) {
3183 case S_IFREG:
3184 fd = open(path, O_RDONLY);
3185 if (fd < 0)
3186 return error("open(\"%s\"): %s", path,
3187 strerror(errno));
3188 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3189 return error("%s: failed to insert into database",
3190 path);
3191 break;
3192 case S_IFLNK:
3193 if (strbuf_readlink(&sb, path, st->st_size)) {
3194 char *errstr = strerror(errno);
3195 return error("readlink(\"%s\"): %s", path,
3196 errstr);
3198 if (!(flags & HASH_WRITE_OBJECT))
3199 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3200 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3201 return error("%s: failed to insert into database",
3202 path);
3203 strbuf_release(&sb);
3204 break;
3205 case S_IFDIR:
3206 return resolve_gitlink_ref(path, "HEAD", sha1);
3207 default:
3208 return error("%s: unsupported file type", path);
3210 return 0;
3213 int read_pack_header(int fd, struct pack_header *header)
3215 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3216 /* "eof before pack header was fully read" */
3217 return PH_ERROR_EOF;
3219 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3220 /* "protocol error (pack signature mismatch detected)" */
3221 return PH_ERROR_PACK_SIGNATURE;
3222 if (!pack_version_ok(header->hdr_version))
3223 /* "protocol error (pack version unsupported)" */
3224 return PH_ERROR_PROTOCOL;
3225 return 0;
3228 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3230 enum object_type type = sha1_object_info(sha1, NULL);
3231 if (type < 0)
3232 die("%s is not a valid object", sha1_to_hex(sha1));
3233 if (type != expect)
3234 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3235 typename(expect));