lockfile.h: extract new header file for the functions in lockfile.c
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
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 "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
34
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38 const unsigned char null_sha1[20];
39
40 /*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
53
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59 };
60
61 /*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67 static struct packed_git *last_found_pack;
68
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 {
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81 }
82
83 int mkdir_in_gitdir(const char *path)
84 {
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110 }
111
112 enum scld_error safe_create_leading_directories(char *path)
113 {
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162 }
163
164 enum scld_error safe_create_leading_directories_const(const char *path)
165 {
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171 }
172
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 {
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183 }
184
185 const char *sha1_file_name(const unsigned char *sha1)
186 {
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203 }
204
205 /*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212 {
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235 }
236
237 char *sha1_pack_name(const unsigned char *sha1)
238 {
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242 }
243
244 char *sha1_pack_index_name(const unsigned char *sha1)
245 {
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249 }
250
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
253
254 /*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271 {
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338 }
339
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342 {
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373 }
374
375 void read_info_alternates(const char * relative_base, int depth)
376 {
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 const char alt_file_name[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path[PATH_MAX + 1 + sizeof alt_file_name];
385 int fd;
386
387 sprintf(path, "%s/%s", relative_base, alt_file_name);
388 fd = git_open_noatime(path);
389 if (fd < 0)
390 return;
391 if (fstat(fd, &st) || (st.st_size == 0)) {
392 close(fd);
393 return;
394 }
395 mapsz = xsize_t(st.st_size);
396 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
397 close(fd);
398
399 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
400
401 munmap(map, mapsz);
402 }
403
404 void add_to_alternates_file(const char *reference)
405 {
406 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
407 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
408 char *alt = mkpath("%s\n", reference);
409 write_or_die(fd, alt, strlen(alt));
410 if (commit_lock_file(lock))
411 die("could not close alternates file");
412 if (alt_odb_tail)
413 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
414 }
415
416 void foreach_alt_odb(alt_odb_fn fn, void *cb)
417 {
418 struct alternate_object_database *ent;
419
420 prepare_alt_odb();
421 for (ent = alt_odb_list; ent; ent = ent->next)
422 if (fn(ent, cb))
423 return;
424 }
425
426 void prepare_alt_odb(void)
427 {
428 const char *alt;
429
430 if (alt_odb_tail)
431 return;
432
433 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
434 if (!alt) alt = "";
435
436 alt_odb_tail = &alt_odb_list;
437 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
438
439 read_info_alternates(get_object_directory(), 0);
440 }
441
442 static int has_loose_object_local(const unsigned char *sha1)
443 {
444 return !access(sha1_file_name(sha1), F_OK);
445 }
446
447 int has_loose_object_nonlocal(const unsigned char *sha1)
448 {
449 struct alternate_object_database *alt;
450 prepare_alt_odb();
451 for (alt = alt_odb_list; alt; alt = alt->next) {
452 fill_sha1_path(alt->name, sha1);
453 if (!access(alt->base, F_OK))
454 return 1;
455 }
456 return 0;
457 }
458
459 static int has_loose_object(const unsigned char *sha1)
460 {
461 return has_loose_object_local(sha1) ||
462 has_loose_object_nonlocal(sha1);
463 }
464
465 static unsigned int pack_used_ctr;
466 static unsigned int pack_mmap_calls;
467 static unsigned int peak_pack_open_windows;
468 static unsigned int pack_open_windows;
469 static unsigned int pack_open_fds;
470 static unsigned int pack_max_fds;
471 static size_t peak_pack_mapped;
472 static size_t pack_mapped;
473 struct packed_git *packed_git;
474
475 void pack_report(void)
476 {
477 fprintf(stderr,
478 "pack_report: getpagesize() = %10" SZ_FMT "\n"
479 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
480 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
481 sz_fmt(getpagesize()),
482 sz_fmt(packed_git_window_size),
483 sz_fmt(packed_git_limit));
484 fprintf(stderr,
485 "pack_report: pack_used_ctr = %10u\n"
486 "pack_report: pack_mmap_calls = %10u\n"
487 "pack_report: pack_open_windows = %10u / %10u\n"
488 "pack_report: pack_mapped = "
489 "%10" SZ_FMT " / %10" SZ_FMT "\n",
490 pack_used_ctr,
491 pack_mmap_calls,
492 pack_open_windows, peak_pack_open_windows,
493 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
494 }
495
496 /*
497 * Open and mmap the index file at path, perform a couple of
498 * consistency checks, then record its information to p. Return 0 on
499 * success.
500 */
501 static int check_packed_git_idx(const char *path, struct packed_git *p)
502 {
503 void *idx_map;
504 struct pack_idx_header *hdr;
505 size_t idx_size;
506 uint32_t version, nr, i, *index;
507 int fd = git_open_noatime(path);
508 struct stat st;
509
510 if (fd < 0)
511 return -1;
512 if (fstat(fd, &st)) {
513 close(fd);
514 return -1;
515 }
516 idx_size = xsize_t(st.st_size);
517 if (idx_size < 4 * 256 + 20 + 20) {
518 close(fd);
519 return error("index file %s is too small", path);
520 }
521 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
522 close(fd);
523
524 hdr = idx_map;
525 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
526 version = ntohl(hdr->idx_version);
527 if (version < 2 || version > 2) {
528 munmap(idx_map, idx_size);
529 return error("index file %s is version %"PRIu32
530 " and is not supported by this binary"
531 " (try upgrading GIT to a newer version)",
532 path, version);
533 }
534 } else
535 version = 1;
536
537 nr = 0;
538 index = idx_map;
539 if (version > 1)
540 index += 2; /* skip index header */
541 for (i = 0; i < 256; i++) {
542 uint32_t n = ntohl(index[i]);
543 if (n < nr) {
544 munmap(idx_map, idx_size);
545 return error("non-monotonic index %s", path);
546 }
547 nr = n;
548 }
549
550 if (version == 1) {
551 /*
552 * Total size:
553 * - 256 index entries 4 bytes each
554 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
555 * - 20-byte SHA1 of the packfile
556 * - 20-byte SHA1 file checksum
557 */
558 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
559 munmap(idx_map, idx_size);
560 return error("wrong index v1 file size in %s", path);
561 }
562 } else if (version == 2) {
563 /*
564 * Minimum size:
565 * - 8 bytes of header
566 * - 256 index entries 4 bytes each
567 * - 20-byte sha1 entry * nr
568 * - 4-byte crc entry * nr
569 * - 4-byte offset entry * nr
570 * - 20-byte SHA1 of the packfile
571 * - 20-byte SHA1 file checksum
572 * And after the 4-byte offset table might be a
573 * variable sized table containing 8-byte entries
574 * for offsets larger than 2^31.
575 */
576 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
577 unsigned long max_size = min_size;
578 if (nr)
579 max_size += (nr - 1)*8;
580 if (idx_size < min_size || idx_size > max_size) {
581 munmap(idx_map, idx_size);
582 return error("wrong index v2 file size in %s", path);
583 }
584 if (idx_size != min_size &&
585 /*
586 * make sure we can deal with large pack offsets.
587 * 31-bit signed offset won't be enough, neither
588 * 32-bit unsigned one will be.
589 */
590 (sizeof(off_t) <= 4)) {
591 munmap(idx_map, idx_size);
592 return error("pack too large for current definition of off_t in %s", path);
593 }
594 }
595
596 p->index_version = version;
597 p->index_data = idx_map;
598 p->index_size = idx_size;
599 p->num_objects = nr;
600 return 0;
601 }
602
603 int open_pack_index(struct packed_git *p)
604 {
605 char *idx_name;
606 int ret;
607
608 if (p->index_data)
609 return 0;
610
611 idx_name = xstrdup(p->pack_name);
612 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
613 ret = check_packed_git_idx(idx_name, p);
614 free(idx_name);
615 return ret;
616 }
617
618 static void scan_windows(struct packed_git *p,
619 struct packed_git **lru_p,
620 struct pack_window **lru_w,
621 struct pack_window **lru_l)
622 {
623 struct pack_window *w, *w_l;
624
625 for (w_l = NULL, w = p->windows; w; w = w->next) {
626 if (!w->inuse_cnt) {
627 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
628 *lru_p = p;
629 *lru_w = w;
630 *lru_l = w_l;
631 }
632 }
633 w_l = w;
634 }
635 }
636
637 static int unuse_one_window(struct packed_git *current)
638 {
639 struct packed_git *p, *lru_p = NULL;
640 struct pack_window *lru_w = NULL, *lru_l = NULL;
641
642 if (current)
643 scan_windows(current, &lru_p, &lru_w, &lru_l);
644 for (p = packed_git; p; p = p->next)
645 scan_windows(p, &lru_p, &lru_w, &lru_l);
646 if (lru_p) {
647 munmap(lru_w->base, lru_w->len);
648 pack_mapped -= lru_w->len;
649 if (lru_l)
650 lru_l->next = lru_w->next;
651 else
652 lru_p->windows = lru_w->next;
653 free(lru_w);
654 pack_open_windows--;
655 return 1;
656 }
657 return 0;
658 }
659
660 void release_pack_memory(size_t need)
661 {
662 size_t cur = pack_mapped;
663 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
664 ; /* nothing */
665 }
666
667 void *xmmap(void *start, size_t length,
668 int prot, int flags, int fd, off_t offset)
669 {
670 void *ret = mmap(start, length, prot, flags, fd, offset);
671 if (ret == MAP_FAILED) {
672 if (!length)
673 return NULL;
674 release_pack_memory(length);
675 ret = mmap(start, length, prot, flags, fd, offset);
676 if (ret == MAP_FAILED)
677 die_errno("Out of memory? mmap failed");
678 }
679 return ret;
680 }
681
682 void close_pack_windows(struct packed_git *p)
683 {
684 while (p->windows) {
685 struct pack_window *w = p->windows;
686
687 if (w->inuse_cnt)
688 die("pack '%s' still has open windows to it",
689 p->pack_name);
690 munmap(w->base, w->len);
691 pack_mapped -= w->len;
692 pack_open_windows--;
693 p->windows = w->next;
694 free(w);
695 }
696 }
697
698 /*
699 * The LRU pack is the one with the oldest MRU window, preferring packs
700 * with no used windows, or the oldest mtime if it has no windows allocated.
701 */
702 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
703 {
704 struct pack_window *w, *this_mru_w;
705 int has_windows_inuse = 0;
706
707 /*
708 * Reject this pack if it has windows and the previously selected
709 * one does not. If this pack does not have windows, reject
710 * it if the pack file is newer than the previously selected one.
711 */
712 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
713 return;
714
715 for (w = this_mru_w = p->windows; w; w = w->next) {
716 /*
717 * Reject this pack if any of its windows are in use,
718 * but the previously selected pack did not have any
719 * inuse windows. Otherwise, record that this pack
720 * has windows in use.
721 */
722 if (w->inuse_cnt) {
723 if (*accept_windows_inuse)
724 has_windows_inuse = 1;
725 else
726 return;
727 }
728
729 if (w->last_used > this_mru_w->last_used)
730 this_mru_w = w;
731
732 /*
733 * Reject this pack if it has windows that have been
734 * used more recently than the previously selected pack.
735 * If the previously selected pack had windows inuse and
736 * we have not encountered a window in this pack that is
737 * inuse, skip this check since we prefer a pack with no
738 * inuse windows to one that has inuse windows.
739 */
740 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
741 this_mru_w->last_used > (*mru_w)->last_used)
742 return;
743 }
744
745 /*
746 * Select this pack.
747 */
748 *mru_w = this_mru_w;
749 *lru_p = p;
750 *accept_windows_inuse = has_windows_inuse;
751 }
752
753 static int close_one_pack(void)
754 {
755 struct packed_git *p, *lru_p = NULL;
756 struct pack_window *mru_w = NULL;
757 int accept_windows_inuse = 1;
758
759 for (p = packed_git; p; p = p->next) {
760 if (p->pack_fd == -1)
761 continue;
762 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
763 }
764
765 if (lru_p) {
766 close(lru_p->pack_fd);
767 pack_open_fds--;
768 lru_p->pack_fd = -1;
769 return 1;
770 }
771
772 return 0;
773 }
774
775 void unuse_pack(struct pack_window **w_cursor)
776 {
777 struct pack_window *w = *w_cursor;
778 if (w) {
779 w->inuse_cnt--;
780 *w_cursor = NULL;
781 }
782 }
783
784 void close_pack_index(struct packed_git *p)
785 {
786 if (p->index_data) {
787 munmap((void *)p->index_data, p->index_size);
788 p->index_data = NULL;
789 }
790 }
791
792 /*
793 * This is used by git-repack in case a newly created pack happens to
794 * contain the same set of objects as an existing one. In that case
795 * the resulting file might be different even if its name would be the
796 * same. It is best to close any reference to the old pack before it is
797 * replaced on disk. Of course no index pointers or windows for given pack
798 * must subsist at this point. If ever objects from this pack are requested
799 * again, the new version of the pack will be reinitialized through
800 * reprepare_packed_git().
801 */
802 void free_pack_by_name(const char *pack_name)
803 {
804 struct packed_git *p, **pp = &packed_git;
805
806 while (*pp) {
807 p = *pp;
808 if (strcmp(pack_name, p->pack_name) == 0) {
809 clear_delta_base_cache();
810 close_pack_windows(p);
811 if (p->pack_fd != -1) {
812 close(p->pack_fd);
813 pack_open_fds--;
814 }
815 close_pack_index(p);
816 free(p->bad_object_sha1);
817 *pp = p->next;
818 if (last_found_pack == p)
819 last_found_pack = NULL;
820 free(p);
821 return;
822 }
823 pp = &p->next;
824 }
825 }
826
827 static unsigned int get_max_fd_limit(void)
828 {
829 #ifdef RLIMIT_NOFILE
830 {
831 struct rlimit lim;
832
833 if (!getrlimit(RLIMIT_NOFILE, &lim))
834 return lim.rlim_cur;
835 }
836 #endif
837
838 #ifdef _SC_OPEN_MAX
839 {
840 long open_max = sysconf(_SC_OPEN_MAX);
841 if (0 < open_max)
842 return open_max;
843 /*
844 * Otherwise, we got -1 for one of the two
845 * reasons:
846 *
847 * (1) sysconf() did not understand _SC_OPEN_MAX
848 * and signaled an error with -1; or
849 * (2) sysconf() said there is no limit.
850 *
851 * We _could_ clear errno before calling sysconf() to
852 * tell these two cases apart and return a huge number
853 * in the latter case to let the caller cap it to a
854 * value that is not so selfish, but letting the
855 * fallback OPEN_MAX codepath take care of these cases
856 * is a lot simpler.
857 */
858 }
859 #endif
860
861 #ifdef OPEN_MAX
862 return OPEN_MAX;
863 #else
864 return 1; /* see the caller ;-) */
865 #endif
866 }
867
868 /*
869 * Do not call this directly as this leaks p->pack_fd on error return;
870 * call open_packed_git() instead.
871 */
872 static int open_packed_git_1(struct packed_git *p)
873 {
874 struct stat st;
875 struct pack_header hdr;
876 unsigned char sha1[20];
877 unsigned char *idx_sha1;
878 long fd_flag;
879
880 if (!p->index_data && open_pack_index(p))
881 return error("packfile %s index unavailable", p->pack_name);
882
883 if (!pack_max_fds) {
884 unsigned int max_fds = get_max_fd_limit();
885
886 /* Save 3 for stdin/stdout/stderr, 22 for work */
887 if (25 < max_fds)
888 pack_max_fds = max_fds - 25;
889 else
890 pack_max_fds = 1;
891 }
892
893 while (pack_max_fds <= pack_open_fds && close_one_pack())
894 ; /* nothing */
895
896 p->pack_fd = git_open_noatime(p->pack_name);
897 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
898 return -1;
899 pack_open_fds++;
900
901 /* If we created the struct before we had the pack we lack size. */
902 if (!p->pack_size) {
903 if (!S_ISREG(st.st_mode))
904 return error("packfile %s not a regular file", p->pack_name);
905 p->pack_size = st.st_size;
906 } else if (p->pack_size != st.st_size)
907 return error("packfile %s size changed", p->pack_name);
908
909 /* We leave these file descriptors open with sliding mmap;
910 * there is no point keeping them open across exec(), though.
911 */
912 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
913 if (fd_flag < 0)
914 return error("cannot determine file descriptor flags");
915 fd_flag |= FD_CLOEXEC;
916 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
917 return error("cannot set FD_CLOEXEC");
918
919 /* Verify we recognize this pack file format. */
920 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
921 return error("file %s is far too short to be a packfile", p->pack_name);
922 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
923 return error("file %s is not a GIT packfile", p->pack_name);
924 if (!pack_version_ok(hdr.hdr_version))
925 return error("packfile %s is version %"PRIu32" and not"
926 " supported (try upgrading GIT to a newer version)",
927 p->pack_name, ntohl(hdr.hdr_version));
928
929 /* Verify the pack matches its index. */
930 if (p->num_objects != ntohl(hdr.hdr_entries))
931 return error("packfile %s claims to have %"PRIu32" objects"
932 " while index indicates %"PRIu32" objects",
933 p->pack_name, ntohl(hdr.hdr_entries),
934 p->num_objects);
935 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
936 return error("end of packfile %s is unavailable", p->pack_name);
937 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
938 return error("packfile %s signature is unavailable", p->pack_name);
939 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
940 if (hashcmp(sha1, idx_sha1))
941 return error("packfile %s does not match index", p->pack_name);
942 return 0;
943 }
944
945 static int open_packed_git(struct packed_git *p)
946 {
947 if (!open_packed_git_1(p))
948 return 0;
949 if (p->pack_fd != -1) {
950 close(p->pack_fd);
951 pack_open_fds--;
952 p->pack_fd = -1;
953 }
954 return -1;
955 }
956
957 static int in_window(struct pack_window *win, off_t offset)
958 {
959 /* We must promise at least 20 bytes (one hash) after the
960 * offset is available from this window, otherwise the offset
961 * is not actually in this window and a different window (which
962 * has that one hash excess) must be used. This is to support
963 * the object header and delta base parsing routines below.
964 */
965 off_t win_off = win->offset;
966 return win_off <= offset
967 && (offset + 20) <= (win_off + win->len);
968 }
969
970 unsigned char *use_pack(struct packed_git *p,
971 struct pack_window **w_cursor,
972 off_t offset,
973 unsigned long *left)
974 {
975 struct pack_window *win = *w_cursor;
976
977 /* Since packfiles end in a hash of their content and it's
978 * pointless to ask for an offset into the middle of that
979 * hash, and the in_window function above wouldn't match
980 * don't allow an offset too close to the end of the file.
981 */
982 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
983 die("packfile %s cannot be accessed", p->pack_name);
984 if (offset > (p->pack_size - 20))
985 die("offset beyond end of packfile (truncated pack?)");
986
987 if (!win || !in_window(win, offset)) {
988 if (win)
989 win->inuse_cnt--;
990 for (win = p->windows; win; win = win->next) {
991 if (in_window(win, offset))
992 break;
993 }
994 if (!win) {
995 size_t window_align = packed_git_window_size / 2;
996 off_t len;
997
998 if (p->pack_fd == -1 && open_packed_git(p))
999 die("packfile %s cannot be accessed", p->pack_name);
1000
1001 win = xcalloc(1, sizeof(*win));
1002 win->offset = (offset / window_align) * window_align;
1003 len = p->pack_size - win->offset;
1004 if (len > packed_git_window_size)
1005 len = packed_git_window_size;
1006 win->len = (size_t)len;
1007 pack_mapped += win->len;
1008 while (packed_git_limit < pack_mapped
1009 && unuse_one_window(p))
1010 ; /* nothing */
1011 win->base = xmmap(NULL, win->len,
1012 PROT_READ, MAP_PRIVATE,
1013 p->pack_fd, win->offset);
1014 if (win->base == MAP_FAILED)
1015 die("packfile %s cannot be mapped: %s",
1016 p->pack_name,
1017 strerror(errno));
1018 if (!win->offset && win->len == p->pack_size
1019 && !p->do_not_close) {
1020 close(p->pack_fd);
1021 pack_open_fds--;
1022 p->pack_fd = -1;
1023 }
1024 pack_mmap_calls++;
1025 pack_open_windows++;
1026 if (pack_mapped > peak_pack_mapped)
1027 peak_pack_mapped = pack_mapped;
1028 if (pack_open_windows > peak_pack_open_windows)
1029 peak_pack_open_windows = pack_open_windows;
1030 win->next = p->windows;
1031 p->windows = win;
1032 }
1033 }
1034 if (win != *w_cursor) {
1035 win->last_used = pack_used_ctr++;
1036 win->inuse_cnt++;
1037 *w_cursor = win;
1038 }
1039 offset -= win->offset;
1040 if (left)
1041 *left = win->len - xsize_t(offset);
1042 return win->base + offset;
1043 }
1044
1045 static struct packed_git *alloc_packed_git(int extra)
1046 {
1047 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1048 memset(p, 0, sizeof(*p));
1049 p->pack_fd = -1;
1050 return p;
1051 }
1052
1053 static void try_to_free_pack_memory(size_t size)
1054 {
1055 release_pack_memory(size);
1056 }
1057
1058 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1059 {
1060 static int have_set_try_to_free_routine;
1061 struct stat st;
1062 struct packed_git *p = alloc_packed_git(path_len + 2);
1063
1064 if (!have_set_try_to_free_routine) {
1065 have_set_try_to_free_routine = 1;
1066 set_try_to_free_routine(try_to_free_pack_memory);
1067 }
1068
1069 /*
1070 * Make sure a corresponding .pack file exists and that
1071 * the index looks sane.
1072 */
1073 path_len -= strlen(".idx");
1074 if (path_len < 1) {
1075 free(p);
1076 return NULL;
1077 }
1078 memcpy(p->pack_name, path, path_len);
1079
1080 strcpy(p->pack_name + path_len, ".keep");
1081 if (!access(p->pack_name, F_OK))
1082 p->pack_keep = 1;
1083
1084 strcpy(p->pack_name + path_len, ".pack");
1085 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1086 free(p);
1087 return NULL;
1088 }
1089
1090 /* ok, it looks sane as far as we can check without
1091 * actually mapping the pack file.
1092 */
1093 p->pack_size = st.st_size;
1094 p->pack_local = local;
1095 p->mtime = st.st_mtime;
1096 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1097 hashclr(p->sha1);
1098 return p;
1099 }
1100
1101 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1102 {
1103 const char *path = sha1_pack_name(sha1);
1104 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1105
1106 strcpy(p->pack_name, path);
1107 hashcpy(p->sha1, sha1);
1108 if (check_packed_git_idx(idx_path, p)) {
1109 free(p);
1110 return NULL;
1111 }
1112
1113 return p;
1114 }
1115
1116 void install_packed_git(struct packed_git *pack)
1117 {
1118 if (pack->pack_fd != -1)
1119 pack_open_fds++;
1120
1121 pack->next = packed_git;
1122 packed_git = pack;
1123 }
1124
1125 void (*report_garbage)(const char *desc, const char *path);
1126
1127 static void report_helper(const struct string_list *list,
1128 int seen_bits, int first, int last)
1129 {
1130 const char *msg;
1131 switch (seen_bits) {
1132 case 0:
1133 msg = "no corresponding .idx or .pack";
1134 break;
1135 case 1:
1136 msg = "no corresponding .idx";
1137 break;
1138 case 2:
1139 msg = "no corresponding .pack";
1140 break;
1141 default:
1142 return;
1143 }
1144 for (; first < last; first++)
1145 report_garbage(msg, list->items[first].string);
1146 }
1147
1148 static void report_pack_garbage(struct string_list *list)
1149 {
1150 int i, baselen = -1, first = 0, seen_bits = 0;
1151
1152 if (!report_garbage)
1153 return;
1154
1155 sort_string_list(list);
1156
1157 for (i = 0; i < list->nr; i++) {
1158 const char *path = list->items[i].string;
1159 if (baselen != -1 &&
1160 strncmp(path, list->items[first].string, baselen)) {
1161 report_helper(list, seen_bits, first, i);
1162 baselen = -1;
1163 seen_bits = 0;
1164 }
1165 if (baselen == -1) {
1166 const char *dot = strrchr(path, '.');
1167 if (!dot) {
1168 report_garbage("garbage found", path);
1169 continue;
1170 }
1171 baselen = dot - path + 1;
1172 first = i;
1173 }
1174 if (!strcmp(path + baselen, "pack"))
1175 seen_bits |= 1;
1176 else if (!strcmp(path + baselen, "idx"))
1177 seen_bits |= 2;
1178 }
1179 report_helper(list, seen_bits, first, list->nr);
1180 }
1181
1182 static void prepare_packed_git_one(char *objdir, int local)
1183 {
1184 struct strbuf path = STRBUF_INIT;
1185 size_t dirnamelen;
1186 DIR *dir;
1187 struct dirent *de;
1188 struct string_list garbage = STRING_LIST_INIT_DUP;
1189
1190 strbuf_addstr(&path, objdir);
1191 strbuf_addstr(&path, "/pack");
1192 dir = opendir(path.buf);
1193 if (!dir) {
1194 if (errno != ENOENT)
1195 error("unable to open object pack directory: %s: %s",
1196 path.buf, strerror(errno));
1197 strbuf_release(&path);
1198 return;
1199 }
1200 strbuf_addch(&path, '/');
1201 dirnamelen = path.len;
1202 while ((de = readdir(dir)) != NULL) {
1203 struct packed_git *p;
1204 size_t base_len;
1205
1206 if (is_dot_or_dotdot(de->d_name))
1207 continue;
1208
1209 strbuf_setlen(&path, dirnamelen);
1210 strbuf_addstr(&path, de->d_name);
1211
1212 base_len = path.len;
1213 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1214 /* Don't reopen a pack we already have. */
1215 for (p = packed_git; p; p = p->next) {
1216 size_t len;
1217 if (strip_suffix(p->pack_name, ".pack", &len) &&
1218 len == base_len &&
1219 !memcmp(p->pack_name, path.buf, len))
1220 break;
1221 }
1222 if (p == NULL &&
1223 /*
1224 * See if it really is a valid .idx file with
1225 * corresponding .pack file that we can map.
1226 */
1227 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1228 install_packed_git(p);
1229 }
1230
1231 if (!report_garbage)
1232 continue;
1233
1234 if (ends_with(de->d_name, ".idx") ||
1235 ends_with(de->d_name, ".pack") ||
1236 ends_with(de->d_name, ".bitmap") ||
1237 ends_with(de->d_name, ".keep"))
1238 string_list_append(&garbage, path.buf);
1239 else
1240 report_garbage("garbage found", path.buf);
1241 }
1242 closedir(dir);
1243 report_pack_garbage(&garbage);
1244 string_list_clear(&garbage, 0);
1245 strbuf_release(&path);
1246 }
1247
1248 static int sort_pack(const void *a_, const void *b_)
1249 {
1250 struct packed_git *a = *((struct packed_git **)a_);
1251 struct packed_git *b = *((struct packed_git **)b_);
1252 int st;
1253
1254 /*
1255 * Local packs tend to contain objects specific to our
1256 * variant of the project than remote ones. In addition,
1257 * remote ones could be on a network mounted filesystem.
1258 * Favor local ones for these reasons.
1259 */
1260 st = a->pack_local - b->pack_local;
1261 if (st)
1262 return -st;
1263
1264 /*
1265 * Younger packs tend to contain more recent objects,
1266 * and more recent objects tend to get accessed more
1267 * often.
1268 */
1269 if (a->mtime < b->mtime)
1270 return 1;
1271 else if (a->mtime == b->mtime)
1272 return 0;
1273 return -1;
1274 }
1275
1276 static void rearrange_packed_git(void)
1277 {
1278 struct packed_git **ary, *p;
1279 int i, n;
1280
1281 for (n = 0, p = packed_git; p; p = p->next)
1282 n++;
1283 if (n < 2)
1284 return;
1285
1286 /* prepare an array of packed_git for easier sorting */
1287 ary = xcalloc(n, sizeof(struct packed_git *));
1288 for (n = 0, p = packed_git; p; p = p->next)
1289 ary[n++] = p;
1290
1291 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1292
1293 /* link them back again */
1294 for (i = 0; i < n - 1; i++)
1295 ary[i]->next = ary[i + 1];
1296 ary[n - 1]->next = NULL;
1297 packed_git = ary[0];
1298
1299 free(ary);
1300 }
1301
1302 static int prepare_packed_git_run_once = 0;
1303 void prepare_packed_git(void)
1304 {
1305 struct alternate_object_database *alt;
1306
1307 if (prepare_packed_git_run_once)
1308 return;
1309 prepare_packed_git_one(get_object_directory(), 1);
1310 prepare_alt_odb();
1311 for (alt = alt_odb_list; alt; alt = alt->next) {
1312 alt->name[-1] = 0;
1313 prepare_packed_git_one(alt->base, 0);
1314 alt->name[-1] = '/';
1315 }
1316 rearrange_packed_git();
1317 prepare_packed_git_run_once = 1;
1318 }
1319
1320 void reprepare_packed_git(void)
1321 {
1322 prepare_packed_git_run_once = 0;
1323 prepare_packed_git();
1324 }
1325
1326 static void mark_bad_packed_object(struct packed_git *p,
1327 const unsigned char *sha1)
1328 {
1329 unsigned i;
1330 for (i = 0; i < p->num_bad_objects; i++)
1331 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1332 return;
1333 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1334 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1335 p->num_bad_objects++;
1336 }
1337
1338 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1339 {
1340 struct packed_git *p;
1341 unsigned i;
1342
1343 for (p = packed_git; p; p = p->next)
1344 for (i = 0; i < p->num_bad_objects; i++)
1345 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1346 return p;
1347 return NULL;
1348 }
1349
1350 /*
1351 * With an in-core object data in "map", rehash it to make sure the
1352 * object name actually matches "sha1" to detect object corruption.
1353 * With "map" == NULL, try reading the object named with "sha1" using
1354 * the streaming interface and rehash it to do the same.
1355 */
1356 int check_sha1_signature(const unsigned char *sha1, void *map,
1357 unsigned long size, const char *type)
1358 {
1359 unsigned char real_sha1[20];
1360 enum object_type obj_type;
1361 struct git_istream *st;
1362 git_SHA_CTX c;
1363 char hdr[32];
1364 int hdrlen;
1365
1366 if (map) {
1367 hash_sha1_file(map, size, type, real_sha1);
1368 return hashcmp(sha1, real_sha1) ? -1 : 0;
1369 }
1370
1371 st = open_istream(sha1, &obj_type, &size, NULL);
1372 if (!st)
1373 return -1;
1374
1375 /* Generate the header */
1376 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1377
1378 /* Sha1.. */
1379 git_SHA1_Init(&c);
1380 git_SHA1_Update(&c, hdr, hdrlen);
1381 for (;;) {
1382 char buf[1024 * 16];
1383 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1384
1385 if (readlen < 0) {
1386 close_istream(st);
1387 return -1;
1388 }
1389 if (!readlen)
1390 break;
1391 git_SHA1_Update(&c, buf, readlen);
1392 }
1393 git_SHA1_Final(real_sha1, &c);
1394 close_istream(st);
1395 return hashcmp(sha1, real_sha1) ? -1 : 0;
1396 }
1397
1398 int git_open_noatime(const char *name)
1399 {
1400 static int sha1_file_open_flag = O_NOATIME;
1401
1402 for (;;) {
1403 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1404 if (fd >= 0)
1405 return fd;
1406
1407 /* Might the failure be due to O_NOATIME? */
1408 if (errno != ENOENT && sha1_file_open_flag) {
1409 sha1_file_open_flag = 0;
1410 continue;
1411 }
1412
1413 return -1;
1414 }
1415 }
1416
1417 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1418 {
1419 struct alternate_object_database *alt;
1420
1421 if (!lstat(sha1_file_name(sha1), st))
1422 return 0;
1423
1424 prepare_alt_odb();
1425 errno = ENOENT;
1426 for (alt = alt_odb_list; alt; alt = alt->next) {
1427 fill_sha1_path(alt->name, sha1);
1428 if (!lstat(alt->base, st))
1429 return 0;
1430 }
1431
1432 return -1;
1433 }
1434
1435 static int open_sha1_file(const unsigned char *sha1)
1436 {
1437 int fd;
1438 struct alternate_object_database *alt;
1439 int most_interesting_errno;
1440
1441 fd = git_open_noatime(sha1_file_name(sha1));
1442 if (fd >= 0)
1443 return fd;
1444 most_interesting_errno = errno;
1445
1446 prepare_alt_odb();
1447 for (alt = alt_odb_list; alt; alt = alt->next) {
1448 fill_sha1_path(alt->name, sha1);
1449 fd = git_open_noatime(alt->base);
1450 if (fd >= 0)
1451 return fd;
1452 if (most_interesting_errno == ENOENT)
1453 most_interesting_errno = errno;
1454 }
1455 errno = most_interesting_errno;
1456 return -1;
1457 }
1458
1459 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1460 {
1461 void *map;
1462 int fd;
1463
1464 fd = open_sha1_file(sha1);
1465 map = NULL;
1466 if (fd >= 0) {
1467 struct stat st;
1468
1469 if (!fstat(fd, &st)) {
1470 *size = xsize_t(st.st_size);
1471 if (!*size) {
1472 /* mmap() is forbidden on empty files */
1473 error("object file %s is empty", sha1_file_name(sha1));
1474 return NULL;
1475 }
1476 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1477 }
1478 close(fd);
1479 }
1480 return map;
1481 }
1482
1483 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1484 unsigned long len, enum object_type *type, unsigned long *sizep)
1485 {
1486 unsigned shift;
1487 unsigned long size, c;
1488 unsigned long used = 0;
1489
1490 c = buf[used++];
1491 *type = (c >> 4) & 7;
1492 size = c & 15;
1493 shift = 4;
1494 while (c & 0x80) {
1495 if (len <= used || bitsizeof(long) <= shift) {
1496 error("bad object header");
1497 size = used = 0;
1498 break;
1499 }
1500 c = buf[used++];
1501 size += (c & 0x7f) << shift;
1502 shift += 7;
1503 }
1504 *sizep = size;
1505 return used;
1506 }
1507
1508 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1509 {
1510 /* Get the data stream */
1511 memset(stream, 0, sizeof(*stream));
1512 stream->next_in = map;
1513 stream->avail_in = mapsize;
1514 stream->next_out = buffer;
1515 stream->avail_out = bufsiz;
1516
1517 git_inflate_init(stream);
1518 return git_inflate(stream, 0);
1519 }
1520
1521 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1522 {
1523 int bytes = strlen(buffer) + 1;
1524 unsigned char *buf = xmallocz(size);
1525 unsigned long n;
1526 int status = Z_OK;
1527
1528 n = stream->total_out - bytes;
1529 if (n > size)
1530 n = size;
1531 memcpy(buf, (char *) buffer + bytes, n);
1532 bytes = n;
1533 if (bytes <= size) {
1534 /*
1535 * The above condition must be (bytes <= size), not
1536 * (bytes < size). In other words, even though we
1537 * expect no more output and set avail_out to zero,
1538 * the input zlib stream may have bytes that express
1539 * "this concludes the stream", and we *do* want to
1540 * eat that input.
1541 *
1542 * Otherwise we would not be able to test that we
1543 * consumed all the input to reach the expected size;
1544 * we also want to check that zlib tells us that all
1545 * went well with status == Z_STREAM_END at the end.
1546 */
1547 stream->next_out = buf + bytes;
1548 stream->avail_out = size - bytes;
1549 while (status == Z_OK)
1550 status = git_inflate(stream, Z_FINISH);
1551 }
1552 if (status == Z_STREAM_END && !stream->avail_in) {
1553 git_inflate_end(stream);
1554 return buf;
1555 }
1556
1557 if (status < 0)
1558 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1559 else if (stream->avail_in)
1560 error("garbage at end of loose object '%s'",
1561 sha1_to_hex(sha1));
1562 free(buf);
1563 return NULL;
1564 }
1565
1566 /*
1567 * We used to just use "sscanf()", but that's actually way
1568 * too permissive for what we want to check. So do an anal
1569 * object header parse by hand.
1570 */
1571 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1572 {
1573 char type[10];
1574 int i;
1575 unsigned long size;
1576
1577 /*
1578 * The type can be at most ten bytes (including the
1579 * terminating '\0' that we add), and is followed by
1580 * a space.
1581 */
1582 i = 0;
1583 for (;;) {
1584 char c = *hdr++;
1585 if (c == ' ')
1586 break;
1587 type[i++] = c;
1588 if (i >= sizeof(type))
1589 return -1;
1590 }
1591 type[i] = 0;
1592
1593 /*
1594 * The length must follow immediately, and be in canonical
1595 * decimal format (ie "010" is not valid).
1596 */
1597 size = *hdr++ - '0';
1598 if (size > 9)
1599 return -1;
1600 if (size) {
1601 for (;;) {
1602 unsigned long c = *hdr - '0';
1603 if (c > 9)
1604 break;
1605 hdr++;
1606 size = size * 10 + c;
1607 }
1608 }
1609 *sizep = size;
1610
1611 /*
1612 * The length must be followed by a zero byte
1613 */
1614 return *hdr ? -1 : type_from_string(type);
1615 }
1616
1617 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1618 {
1619 int ret;
1620 git_zstream stream;
1621 char hdr[8192];
1622
1623 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1624 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1625 return NULL;
1626
1627 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1628 }
1629
1630 unsigned long get_size_from_delta(struct packed_git *p,
1631 struct pack_window **w_curs,
1632 off_t curpos)
1633 {
1634 const unsigned char *data;
1635 unsigned char delta_head[20], *in;
1636 git_zstream stream;
1637 int st;
1638
1639 memset(&stream, 0, sizeof(stream));
1640 stream.next_out = delta_head;
1641 stream.avail_out = sizeof(delta_head);
1642
1643 git_inflate_init(&stream);
1644 do {
1645 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1646 stream.next_in = in;
1647 st = git_inflate(&stream, Z_FINISH);
1648 curpos += stream.next_in - in;
1649 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1650 stream.total_out < sizeof(delta_head));
1651 git_inflate_end(&stream);
1652 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1653 error("delta data unpack-initial failed");
1654 return 0;
1655 }
1656
1657 /* Examine the initial part of the delta to figure out
1658 * the result size.
1659 */
1660 data = delta_head;
1661
1662 /* ignore base size */
1663 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1664
1665 /* Read the result size */
1666 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1667 }
1668
1669 static off_t get_delta_base(struct packed_git *p,
1670 struct pack_window **w_curs,
1671 off_t *curpos,
1672 enum object_type type,
1673 off_t delta_obj_offset)
1674 {
1675 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1676 off_t base_offset;
1677
1678 /* use_pack() assured us we have [base_info, base_info + 20)
1679 * as a range that we can look at without walking off the
1680 * end of the mapped window. Its actually the hash size
1681 * that is assured. An OFS_DELTA longer than the hash size
1682 * is stupid, as then a REF_DELTA would be smaller to store.
1683 */
1684 if (type == OBJ_OFS_DELTA) {
1685 unsigned used = 0;
1686 unsigned char c = base_info[used++];
1687 base_offset = c & 127;
1688 while (c & 128) {
1689 base_offset += 1;
1690 if (!base_offset || MSB(base_offset, 7))
1691 return 0; /* overflow */
1692 c = base_info[used++];
1693 base_offset = (base_offset << 7) + (c & 127);
1694 }
1695 base_offset = delta_obj_offset - base_offset;
1696 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1697 return 0; /* out of bound */
1698 *curpos += used;
1699 } else if (type == OBJ_REF_DELTA) {
1700 /* The base entry _must_ be in the same pack */
1701 base_offset = find_pack_entry_one(base_info, p);
1702 *curpos += 20;
1703 } else
1704 die("I am totally screwed");
1705 return base_offset;
1706 }
1707
1708 /*
1709 * Like get_delta_base above, but we return the sha1 instead of the pack
1710 * offset. This means it is cheaper for REF deltas (we do not have to do
1711 * the final object lookup), but more expensive for OFS deltas (we
1712 * have to load the revidx to convert the offset back into a sha1).
1713 */
1714 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1715 struct pack_window **w_curs,
1716 off_t curpos,
1717 enum object_type type,
1718 off_t delta_obj_offset)
1719 {
1720 if (type == OBJ_REF_DELTA) {
1721 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1722 return base;
1723 } else if (type == OBJ_OFS_DELTA) {
1724 struct revindex_entry *revidx;
1725 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1726 type, delta_obj_offset);
1727
1728 if (!base_offset)
1729 return NULL;
1730
1731 revidx = find_pack_revindex(p, base_offset);
1732 if (!revidx)
1733 return NULL;
1734
1735 return nth_packed_object_sha1(p, revidx->nr);
1736 } else
1737 return NULL;
1738 }
1739
1740 int unpack_object_header(struct packed_git *p,
1741 struct pack_window **w_curs,
1742 off_t *curpos,
1743 unsigned long *sizep)
1744 {
1745 unsigned char *base;
1746 unsigned long left;
1747 unsigned long used;
1748 enum object_type type;
1749
1750 /* use_pack() assures us we have [base, base + 20) available
1751 * as a range that we can look at. (Its actually the hash
1752 * size that is assured.) With our object header encoding
1753 * the maximum deflated object size is 2^137, which is just
1754 * insane, so we know won't exceed what we have been given.
1755 */
1756 base = use_pack(p, w_curs, *curpos, &left);
1757 used = unpack_object_header_buffer(base, left, &type, sizep);
1758 if (!used) {
1759 type = OBJ_BAD;
1760 } else
1761 *curpos += used;
1762
1763 return type;
1764 }
1765
1766 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1767 {
1768 int type;
1769 struct revindex_entry *revidx;
1770 const unsigned char *sha1;
1771 revidx = find_pack_revindex(p, obj_offset);
1772 if (!revidx)
1773 return OBJ_BAD;
1774 sha1 = nth_packed_object_sha1(p, revidx->nr);
1775 mark_bad_packed_object(p, sha1);
1776 type = sha1_object_info(sha1, NULL);
1777 if (type <= OBJ_NONE)
1778 return OBJ_BAD;
1779 return type;
1780 }
1781
1782 #define POI_STACK_PREALLOC 64
1783
1784 static enum object_type packed_to_object_type(struct packed_git *p,
1785 off_t obj_offset,
1786 enum object_type type,
1787 struct pack_window **w_curs,
1788 off_t curpos)
1789 {
1790 off_t small_poi_stack[POI_STACK_PREALLOC];
1791 off_t *poi_stack = small_poi_stack;
1792 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1793
1794 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1795 off_t base_offset;
1796 unsigned long size;
1797 /* Push the object we're going to leave behind */
1798 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1799 poi_stack_alloc = alloc_nr(poi_stack_nr);
1800 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1801 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1802 } else {
1803 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1804 }
1805 poi_stack[poi_stack_nr++] = obj_offset;
1806 /* If parsing the base offset fails, just unwind */
1807 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1808 if (!base_offset)
1809 goto unwind;
1810 curpos = obj_offset = base_offset;
1811 type = unpack_object_header(p, w_curs, &curpos, &size);
1812 if (type <= OBJ_NONE) {
1813 /* If getting the base itself fails, we first
1814 * retry the base, otherwise unwind */
1815 type = retry_bad_packed_offset(p, base_offset);
1816 if (type > OBJ_NONE)
1817 goto out;
1818 goto unwind;
1819 }
1820 }
1821
1822 switch (type) {
1823 case OBJ_BAD:
1824 case OBJ_COMMIT:
1825 case OBJ_TREE:
1826 case OBJ_BLOB:
1827 case OBJ_TAG:
1828 break;
1829 default:
1830 error("unknown object type %i at offset %"PRIuMAX" in %s",
1831 type, (uintmax_t)obj_offset, p->pack_name);
1832 type = OBJ_BAD;
1833 }
1834
1835 out:
1836 if (poi_stack != small_poi_stack)
1837 free(poi_stack);
1838 return type;
1839
1840 unwind:
1841 while (poi_stack_nr) {
1842 obj_offset = poi_stack[--poi_stack_nr];
1843 type = retry_bad_packed_offset(p, obj_offset);
1844 if (type > OBJ_NONE)
1845 goto out;
1846 }
1847 type = OBJ_BAD;
1848 goto out;
1849 }
1850
1851 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1852 struct object_info *oi)
1853 {
1854 struct pack_window *w_curs = NULL;
1855 unsigned long size;
1856 off_t curpos = obj_offset;
1857 enum object_type type;
1858
1859 /*
1860 * We always get the representation type, but only convert it to
1861 * a "real" type later if the caller is interested.
1862 */
1863 type = unpack_object_header(p, &w_curs, &curpos, &size);
1864
1865 if (oi->sizep) {
1866 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1867 off_t tmp_pos = curpos;
1868 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1869 type, obj_offset);
1870 if (!base_offset) {
1871 type = OBJ_BAD;
1872 goto out;
1873 }
1874 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1875 if (*oi->sizep == 0) {
1876 type = OBJ_BAD;
1877 goto out;
1878 }
1879 } else {
1880 *oi->sizep = size;
1881 }
1882 }
1883
1884 if (oi->disk_sizep) {
1885 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1886 *oi->disk_sizep = revidx[1].offset - obj_offset;
1887 }
1888
1889 if (oi->typep) {
1890 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1891 if (*oi->typep < 0) {
1892 type = OBJ_BAD;
1893 goto out;
1894 }
1895 }
1896
1897 if (oi->delta_base_sha1) {
1898 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1899 const unsigned char *base;
1900
1901 base = get_delta_base_sha1(p, &w_curs, curpos,
1902 type, obj_offset);
1903 if (!base) {
1904 type = OBJ_BAD;
1905 goto out;
1906 }
1907
1908 hashcpy(oi->delta_base_sha1, base);
1909 } else
1910 hashclr(oi->delta_base_sha1);
1911 }
1912
1913 out:
1914 unuse_pack(&w_curs);
1915 return type;
1916 }
1917
1918 static void *unpack_compressed_entry(struct packed_git *p,
1919 struct pack_window **w_curs,
1920 off_t curpos,
1921 unsigned long size)
1922 {
1923 int st;
1924 git_zstream stream;
1925 unsigned char *buffer, *in;
1926
1927 buffer = xmallocz_gently(size);
1928 if (!buffer)
1929 return NULL;
1930 memset(&stream, 0, sizeof(stream));
1931 stream.next_out = buffer;
1932 stream.avail_out = size + 1;
1933
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;
1947 }
1948
1949 return buffer;
1950 }
1951
1952 #define MAX_DELTA_CACHE (256)
1953
1954 static size_t delta_base_cached;
1955
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 };
1960
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];
1969
1970 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1971 {
1972 unsigned long hash;
1973
1974 hash = (unsigned long)p + (unsigned long)base_offset;
1975 hash += (hash >> 8) + (hash >> 16);
1976 return hash % MAX_DELTA_CACHE;
1977 }
1978
1979 static struct delta_base_cache_entry *
1980 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1981 {
1982 unsigned long hash = pack_entry_hash(p, base_offset);
1983 return delta_base_cache + hash;
1984 }
1985
1986 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1987 struct packed_git *p, off_t base_offset)
1988 {
1989 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1990 }
1991
1992 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1993 {
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);
1997 }
1998
1999 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2000 {
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;
2005 }
2006
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)
2009 {
2010 struct delta_base_cache_entry *ent;
2011 void *ret;
2012
2013 ent = get_delta_base_cache_entry(p, base_offset);
2014
2015 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2016 return unpack_entry(p, base_offset, type, base_size);
2017
2018 ret = ent->data;
2019
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;
2027 }
2028
2029 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2030 {
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;
2037 }
2038 }
2039
2040 void clear_delta_base_cache(void)
2041 {
2042 unsigned long p;
2043 for (p = 0; p < MAX_DELTA_CACHE; p++)
2044 release_delta_base_cache(&delta_base_cache[p]);
2045 }
2046
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)
2049 {
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;
2053
2054 release_delta_base_cache(ent);
2055 delta_base_cached += base_size;
2056
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);
2064 }
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);
2071 }
2072
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;
2082 }
2083
2084 static void *read_object(const unsigned char *sha1, enum object_type *type,
2085 unsigned long *size);
2086
2087 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2088 {
2089 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2090 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2091 p->pack_name, (uintmax_t)obj_offset);
2092 }
2093
2094 int do_check_packed_object_crc;
2095
2096 #define UNPACK_ENTRY_STACK_PREALLOC 64
2097 struct unpack_entry_stack_ent {
2098 off_t obj_offset;
2099 off_t curpos;
2100 unsigned long size;
2101 };
2102
2103 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2104 enum object_type *final_type, unsigned long *final_size)
2105 {
2106 struct pack_window *w_curs = NULL;
2107 off_t curpos = obj_offset;
2108 void *data = NULL;
2109 unsigned long size;
2110 enum object_type type;
2111 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2112 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2113 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2114 int base_from_cache = 0;
2115
2116 write_pack_access_log(p, obj_offset);
2117
2118 /* PHASE 1: drill down to the innermost base object */
2119 for (;;) {
2120 off_t base_offset;
2121 int i;
2122 struct delta_base_cache_entry *ent;
2123
2124 ent = get_delta_base_cache_entry(p, curpos);
2125 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2126 type = ent->type;
2127 data = ent->data;
2128 size = ent->size;
2129 clear_delta_base_cache_entry(ent);
2130 base_from_cache = 1;
2131 break;
2132 }
2133
2134 if (do_check_packed_object_crc && p->index_version > 1) {
2135 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2136 unsigned long len = revidx[1].offset - obj_offset;
2137 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2138 const unsigned char *sha1 =
2139 nth_packed_object_sha1(p, revidx->nr);
2140 error("bad packed object CRC for %s",
2141 sha1_to_hex(sha1));
2142 mark_bad_packed_object(p, sha1);
2143 unuse_pack(&w_curs);
2144 return NULL;
2145 }
2146 }
2147
2148 type = unpack_object_header(p, &w_curs, &curpos, &size);
2149 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2150 break;
2151
2152 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2153 if (!base_offset) {
2154 error("failed to validate delta base reference "
2155 "at offset %"PRIuMAX" from %s",
2156 (uintmax_t)curpos, p->pack_name);
2157 /* bail to phase 2, in hopes of recovery */
2158 data = NULL;
2159 break;
2160 }
2161
2162 /* push object, proceed to base */
2163 if (delta_stack_nr >= delta_stack_alloc
2164 && delta_stack == small_delta_stack) {
2165 delta_stack_alloc = alloc_nr(delta_stack_nr);
2166 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2167 memcpy(delta_stack, small_delta_stack,
2168 sizeof(*delta_stack)*delta_stack_nr);
2169 } else {
2170 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2171 }
2172 i = delta_stack_nr++;
2173 delta_stack[i].obj_offset = obj_offset;
2174 delta_stack[i].curpos = curpos;
2175 delta_stack[i].size = size;
2176
2177 curpos = obj_offset = base_offset;
2178 }
2179
2180 /* PHASE 2: handle the base */
2181 switch (type) {
2182 case OBJ_OFS_DELTA:
2183 case OBJ_REF_DELTA:
2184 if (data)
2185 die("BUG in unpack_entry: left loop at a valid delta");
2186 break;
2187 case OBJ_COMMIT:
2188 case OBJ_TREE:
2189 case OBJ_BLOB:
2190 case OBJ_TAG:
2191 if (!base_from_cache)
2192 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2193 break;
2194 default:
2195 data = NULL;
2196 error("unknown object type %i at offset %"PRIuMAX" in %s",
2197 type, (uintmax_t)obj_offset, p->pack_name);
2198 }
2199
2200 /* PHASE 3: apply deltas in order */
2201
2202 /* invariants:
2203 * 'data' holds the base data, or NULL if there was corruption
2204 */
2205 while (delta_stack_nr) {
2206 void *delta_data;
2207 void *base = data;
2208 unsigned long delta_size, base_size = size;
2209 int i;
2210
2211 data = NULL;
2212
2213 if (base)
2214 add_delta_base_cache(p, obj_offset, base, base_size, type);
2215
2216 if (!base) {
2217 /*
2218 * We're probably in deep shit, but let's try to fetch
2219 * the required base anyway from another pack or loose.
2220 * This is costly but should happen only in the presence
2221 * of a corrupted pack, and is better than failing outright.
2222 */
2223 struct revindex_entry *revidx;
2224 const unsigned char *base_sha1;
2225 revidx = find_pack_revindex(p, obj_offset);
2226 if (revidx) {
2227 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2228 error("failed to read delta base object %s"
2229 " at offset %"PRIuMAX" from %s",
2230 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2231 p->pack_name);
2232 mark_bad_packed_object(p, base_sha1);
2233 base = read_object(base_sha1, &type, &base_size);
2234 }
2235 }
2236
2237 i = --delta_stack_nr;
2238 obj_offset = delta_stack[i].obj_offset;
2239 curpos = delta_stack[i].curpos;
2240 delta_size = delta_stack[i].size;
2241
2242 if (!base)
2243 continue;
2244
2245 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2246
2247 if (!delta_data) {
2248 error("failed to unpack compressed delta "
2249 "at offset %"PRIuMAX" from %s",
2250 (uintmax_t)curpos, p->pack_name);
2251 data = NULL;
2252 continue;
2253 }
2254
2255 data = patch_delta(base, base_size,
2256 delta_data, delta_size,
2257 &size);
2258
2259 /*
2260 * We could not apply the delta; warn the user, but keep going.
2261 * Our failure will be noticed either in the next iteration of
2262 * the loop, or if this is the final delta, in the caller when
2263 * we return NULL. Those code paths will take care of making
2264 * a more explicit warning and retrying with another copy of
2265 * the object.
2266 */
2267 if (!data)
2268 error("failed to apply delta");
2269
2270 free(delta_data);
2271 }
2272
2273 *final_type = type;
2274 *final_size = size;
2275
2276 unuse_pack(&w_curs);
2277
2278 if (delta_stack != small_delta_stack)
2279 free(delta_stack);
2280
2281 return data;
2282 }
2283
2284 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2285 uint32_t n)
2286 {
2287 const unsigned char *index = p->index_data;
2288 if (!index) {
2289 if (open_pack_index(p))
2290 return NULL;
2291 index = p->index_data;
2292 }
2293 if (n >= p->num_objects)
2294 return NULL;
2295 index += 4 * 256;
2296 if (p->index_version == 1) {
2297 return index + 24 * n + 4;
2298 } else {
2299 index += 8;
2300 return index + 20 * n;
2301 }
2302 }
2303
2304 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2305 {
2306 const unsigned char *index = p->index_data;
2307 index += 4 * 256;
2308 if (p->index_version == 1) {
2309 return ntohl(*((uint32_t *)(index + 24 * n)));
2310 } else {
2311 uint32_t off;
2312 index += 8 + p->num_objects * (20 + 4);
2313 off = ntohl(*((uint32_t *)(index + 4 * n)));
2314 if (!(off & 0x80000000))
2315 return off;
2316 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2317 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2318 ntohl(*((uint32_t *)(index + 4)));
2319 }
2320 }
2321
2322 off_t find_pack_entry_one(const unsigned char *sha1,
2323 struct packed_git *p)
2324 {
2325 const uint32_t *level1_ofs = p->index_data;
2326 const unsigned char *index = p->index_data;
2327 unsigned hi, lo, stride;
2328 static int use_lookup = -1;
2329 static int debug_lookup = -1;
2330
2331 if (debug_lookup < 0)
2332 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2333
2334 if (!index) {
2335 if (open_pack_index(p))
2336 return 0;
2337 level1_ofs = p->index_data;
2338 index = p->index_data;
2339 }
2340 if (p->index_version > 1) {
2341 level1_ofs += 2;
2342 index += 8;
2343 }
2344 index += 4 * 256;
2345 hi = ntohl(level1_ofs[*sha1]);
2346 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2347 if (p->index_version > 1) {
2348 stride = 20;
2349 } else {
2350 stride = 24;
2351 index += 4;
2352 }
2353
2354 if (debug_lookup)
2355 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2356 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2357
2358 if (use_lookup < 0)
2359 use_lookup = !!getenv("GIT_USE_LOOKUP");
2360 if (use_lookup) {
2361 int pos = sha1_entry_pos(index, stride, 0,
2362 lo, hi, p->num_objects, sha1);
2363 if (pos < 0)
2364 return 0;
2365 return nth_packed_object_offset(p, pos);
2366 }
2367
2368 do {
2369 unsigned mi = (lo + hi) / 2;
2370 int cmp = hashcmp(index + mi * stride, sha1);
2371
2372 if (debug_lookup)
2373 printf("lo %u hi %u rg %u mi %u\n",
2374 lo, hi, hi - lo, mi);
2375 if (!cmp)
2376 return nth_packed_object_offset(p, mi);
2377 if (cmp > 0)
2378 hi = mi;
2379 else
2380 lo = mi+1;
2381 } while (lo < hi);
2382 return 0;
2383 }
2384
2385 int is_pack_valid(struct packed_git *p)
2386 {
2387 /* An already open pack is known to be valid. */
2388 if (p->pack_fd != -1)
2389 return 1;
2390
2391 /* If the pack has one window completely covering the
2392 * file size, the pack is known to be valid even if
2393 * the descriptor is not currently open.
2394 */
2395 if (p->windows) {
2396 struct pack_window *w = p->windows;
2397
2398 if (!w->offset && w->len == p->pack_size)
2399 return 1;
2400 }
2401
2402 /* Force the pack to open to prove its valid. */
2403 return !open_packed_git(p);
2404 }
2405
2406 static int fill_pack_entry(const unsigned char *sha1,
2407 struct pack_entry *e,
2408 struct packed_git *p)
2409 {
2410 off_t offset;
2411
2412 if (p->num_bad_objects) {
2413 unsigned i;
2414 for (i = 0; i < p->num_bad_objects; i++)
2415 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2416 return 0;
2417 }
2418
2419 offset = find_pack_entry_one(sha1, p);
2420 if (!offset)
2421 return 0;
2422
2423 /*
2424 * We are about to tell the caller where they can locate the
2425 * requested object. We better make sure the packfile is
2426 * still here and can be accessed before supplying that
2427 * answer, as it may have been deleted since the index was
2428 * loaded!
2429 */
2430 if (!is_pack_valid(p)) {
2431 warning("packfile %s cannot be accessed", p->pack_name);
2432 return 0;
2433 }
2434 e->offset = offset;
2435 e->p = p;
2436 hashcpy(e->sha1, sha1);
2437 return 1;
2438 }
2439
2440 /*
2441 * Iff a pack file contains the object named by sha1, return true and
2442 * store its location to e.
2443 */
2444 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2445 {
2446 struct packed_git *p;
2447
2448 prepare_packed_git();
2449 if (!packed_git)
2450 return 0;
2451
2452 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2453 return 1;
2454
2455 for (p = packed_git; p; p = p->next) {
2456 if (p == last_found_pack)
2457 continue; /* we already checked this one */
2458
2459 if (fill_pack_entry(sha1, e, p)) {
2460 last_found_pack = p;
2461 return 1;
2462 }
2463 }
2464 return 0;
2465 }
2466
2467 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2468 struct packed_git *packs)
2469 {
2470 struct packed_git *p;
2471
2472 for (p = packs; p; p = p->next) {
2473 if (find_pack_entry_one(sha1, p))
2474 return p;
2475 }
2476 return NULL;
2477
2478 }
2479
2480 static int sha1_loose_object_info(const unsigned char *sha1,
2481 struct object_info *oi)
2482 {
2483 int status;
2484 unsigned long mapsize, size;
2485 void *map;
2486 git_zstream stream;
2487 char hdr[32];
2488
2489 if (oi->delta_base_sha1)
2490 hashclr(oi->delta_base_sha1);
2491
2492 /*
2493 * If we don't care about type or size, then we don't
2494 * need to look inside the object at all. Note that we
2495 * do not optimize out the stat call, even if the
2496 * caller doesn't care about the disk-size, since our
2497 * return value implicitly indicates whether the
2498 * object even exists.
2499 */
2500 if (!oi->typep && !oi->sizep) {
2501 struct stat st;
2502 if (stat_sha1_file(sha1, &st) < 0)
2503 return -1;
2504 if (oi->disk_sizep)
2505 *oi->disk_sizep = st.st_size;
2506 return 0;
2507 }
2508
2509 map = map_sha1_file(sha1, &mapsize);
2510 if (!map)
2511 return -1;
2512 if (oi->disk_sizep)
2513 *oi->disk_sizep = mapsize;
2514 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2515 status = error("unable to unpack %s header",
2516 sha1_to_hex(sha1));
2517 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2518 status = error("unable to parse %s header", sha1_to_hex(sha1));
2519 else if (oi->sizep)
2520 *oi->sizep = size;
2521 git_inflate_end(&stream);
2522 munmap(map, mapsize);
2523 if (oi->typep)
2524 *oi->typep = status;
2525 return 0;
2526 }
2527
2528 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2529 {
2530 struct cached_object *co;
2531 struct pack_entry e;
2532 int rtype;
2533 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2534
2535 co = find_cached_object(real);
2536 if (co) {
2537 if (oi->typep)
2538 *(oi->typep) = co->type;
2539 if (oi->sizep)
2540 *(oi->sizep) = co->size;
2541 if (oi->disk_sizep)
2542 *(oi->disk_sizep) = 0;
2543 if (oi->delta_base_sha1)
2544 hashclr(oi->delta_base_sha1);
2545 oi->whence = OI_CACHED;
2546 return 0;
2547 }
2548
2549 if (!find_pack_entry(real, &e)) {
2550 /* Most likely it's a loose object. */
2551 if (!sha1_loose_object_info(real, oi)) {
2552 oi->whence = OI_LOOSE;
2553 return 0;
2554 }
2555
2556 /* Not a loose object; someone else may have just packed it. */
2557 reprepare_packed_git();
2558 if (!find_pack_entry(real, &e))
2559 return -1;
2560 }
2561
2562 rtype = packed_object_info(e.p, e.offset, oi);
2563 if (rtype < 0) {
2564 mark_bad_packed_object(e.p, real);
2565 return sha1_object_info_extended(real, oi, 0);
2566 } else if (in_delta_base_cache(e.p, e.offset)) {
2567 oi->whence = OI_DBCACHED;
2568 } else {
2569 oi->whence = OI_PACKED;
2570 oi->u.packed.offset = e.offset;
2571 oi->u.packed.pack = e.p;
2572 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2573 rtype == OBJ_OFS_DELTA);
2574 }
2575
2576 return 0;
2577 }
2578
2579 /* returns enum object_type or negative */
2580 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2581 {
2582 enum object_type type;
2583 struct object_info oi = {NULL};
2584
2585 oi.typep = &type;
2586 oi.sizep = sizep;
2587 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2588 return -1;
2589 return type;
2590 }
2591
2592 static void *read_packed_sha1(const unsigned char *sha1,
2593 enum object_type *type, unsigned long *size)
2594 {
2595 struct pack_entry e;
2596 void *data;
2597
2598 if (!find_pack_entry(sha1, &e))
2599 return NULL;
2600 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2601 if (!data) {
2602 /*
2603 * We're probably in deep shit, but let's try to fetch
2604 * the required object anyway from another pack or loose.
2605 * This should happen only in the presence of a corrupted
2606 * pack, and is better than failing outright.
2607 */
2608 error("failed to read object %s at offset %"PRIuMAX" from %s",
2609 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2610 mark_bad_packed_object(e.p, sha1);
2611 data = read_object(sha1, type, size);
2612 }
2613 return data;
2614 }
2615
2616 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2617 unsigned char *sha1)
2618 {
2619 struct cached_object *co;
2620
2621 hash_sha1_file(buf, len, typename(type), sha1);
2622 if (has_sha1_file(sha1) || find_cached_object(sha1))
2623 return 0;
2624 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2625 co = &cached_objects[cached_object_nr++];
2626 co->size = len;
2627 co->type = type;
2628 co->buf = xmalloc(len);
2629 memcpy(co->buf, buf, len);
2630 hashcpy(co->sha1, sha1);
2631 return 0;
2632 }
2633
2634 static void *read_object(const unsigned char *sha1, enum object_type *type,
2635 unsigned long *size)
2636 {
2637 unsigned long mapsize;
2638 void *map, *buf;
2639 struct cached_object *co;
2640
2641 co = find_cached_object(sha1);
2642 if (co) {
2643 *type = co->type;
2644 *size = co->size;
2645 return xmemdupz(co->buf, co->size);
2646 }
2647
2648 buf = read_packed_sha1(sha1, type, size);
2649 if (buf)
2650 return buf;
2651 map = map_sha1_file(sha1, &mapsize);
2652 if (map) {
2653 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2654 munmap(map, mapsize);
2655 return buf;
2656 }
2657 reprepare_packed_git();
2658 return read_packed_sha1(sha1, type, size);
2659 }
2660
2661 /*
2662 * This function dies on corrupt objects; the callers who want to
2663 * deal with them should arrange to call read_object() and give error
2664 * messages themselves.
2665 */
2666 void *read_sha1_file_extended(const unsigned char *sha1,
2667 enum object_type *type,
2668 unsigned long *size,
2669 unsigned flag)
2670 {
2671 void *data;
2672 const struct packed_git *p;
2673 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2674
2675 errno = 0;
2676 data = read_object(repl, type, size);
2677 if (data)
2678 return data;
2679
2680 if (errno && errno != ENOENT)
2681 die_errno("failed to read object %s", sha1_to_hex(sha1));
2682
2683 /* die if we replaced an object with one that does not exist */
2684 if (repl != sha1)
2685 die("replacement %s not found for %s",
2686 sha1_to_hex(repl), sha1_to_hex(sha1));
2687
2688 if (has_loose_object(repl)) {
2689 const char *path = sha1_file_name(sha1);
2690
2691 die("loose object %s (stored in %s) is corrupt",
2692 sha1_to_hex(repl), path);
2693 }
2694
2695 if ((p = has_packed_and_bad(repl)) != NULL)
2696 die("packed object %s (stored in %s) is corrupt",
2697 sha1_to_hex(repl), p->pack_name);
2698
2699 return NULL;
2700 }
2701
2702 void *read_object_with_reference(const unsigned char *sha1,
2703 const char *required_type_name,
2704 unsigned long *size,
2705 unsigned char *actual_sha1_return)
2706 {
2707 enum object_type type, required_type;
2708 void *buffer;
2709 unsigned long isize;
2710 unsigned char actual_sha1[20];
2711
2712 required_type = type_from_string(required_type_name);
2713 hashcpy(actual_sha1, sha1);
2714 while (1) {
2715 int ref_length = -1;
2716 const char *ref_type = NULL;
2717
2718 buffer = read_sha1_file(actual_sha1, &type, &isize);
2719 if (!buffer)
2720 return NULL;
2721 if (type == required_type) {
2722 *size = isize;
2723 if (actual_sha1_return)
2724 hashcpy(actual_sha1_return, actual_sha1);
2725 return buffer;
2726 }
2727 /* Handle references */
2728 else if (type == OBJ_COMMIT)
2729 ref_type = "tree ";
2730 else if (type == OBJ_TAG)
2731 ref_type = "object ";
2732 else {
2733 free(buffer);
2734 return NULL;
2735 }
2736 ref_length = strlen(ref_type);
2737
2738 if (ref_length + 40 > isize ||
2739 memcmp(buffer, ref_type, ref_length) ||
2740 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2741 free(buffer);
2742 return NULL;
2743 }
2744 free(buffer);
2745 /* Now we have the ID of the referred-to object in
2746 * actual_sha1. Check again. */
2747 }
2748 }
2749
2750 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2751 const char *type, unsigned char *sha1,
2752 char *hdr, int *hdrlen)
2753 {
2754 git_SHA_CTX c;
2755
2756 /* Generate the header */
2757 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2758
2759 /* Sha1.. */
2760 git_SHA1_Init(&c);
2761 git_SHA1_Update(&c, hdr, *hdrlen);
2762 git_SHA1_Update(&c, buf, len);
2763 git_SHA1_Final(sha1, &c);
2764 }
2765
2766 /*
2767 * Move the just written object into its final resting place.
2768 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2769 * "moving" is only a part of what it does, when no patch between
2770 * master to pu changes the call sites of this function.
2771 */
2772 int move_temp_to_file(const char *tmpfile, const char *filename)
2773 {
2774 int ret = 0;
2775
2776 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2777 goto try_rename;
2778 else if (link(tmpfile, filename))
2779 ret = errno;
2780
2781 /*
2782 * Coda hack - coda doesn't like cross-directory links,
2783 * so we fall back to a rename, which will mean that it
2784 * won't be able to check collisions, but that's not a
2785 * big deal.
2786 *
2787 * The same holds for FAT formatted media.
2788 *
2789 * When this succeeds, we just return. We have nothing
2790 * left to unlink.
2791 */
2792 if (ret && ret != EEXIST) {
2793 try_rename:
2794 if (!rename(tmpfile, filename))
2795 goto out;
2796 ret = errno;
2797 }
2798 unlink_or_warn(tmpfile);
2799 if (ret) {
2800 if (ret != EEXIST) {
2801 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2802 }
2803 /* FIXME!!! Collision check here ? */
2804 }
2805
2806 out:
2807 if (adjust_shared_perm(filename))
2808 return error("unable to set permission to '%s'", filename);
2809 return 0;
2810 }
2811
2812 static int write_buffer(int fd, const void *buf, size_t len)
2813 {
2814 if (write_in_full(fd, buf, len) < 0)
2815 return error("file write error (%s)", strerror(errno));
2816 return 0;
2817 }
2818
2819 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2820 unsigned char *sha1)
2821 {
2822 char hdr[32];
2823 int hdrlen;
2824 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2825 return 0;
2826 }
2827
2828 /* Finalize a file on disk, and close it. */
2829 static void close_sha1_file(int fd)
2830 {
2831 if (fsync_object_files)
2832 fsync_or_die(fd, "sha1 file");
2833 if (close(fd) != 0)
2834 die_errno("error when closing sha1 file");
2835 }
2836
2837 /* Size of directory component, including the ending '/' */
2838 static inline int directory_size(const char *filename)
2839 {
2840 const char *s = strrchr(filename, '/');
2841 if (!s)
2842 return 0;
2843 return s - filename + 1;
2844 }
2845
2846 /*
2847 * This creates a temporary file in the same directory as the final
2848 * 'filename'
2849 *
2850 * We want to avoid cross-directory filename renames, because those
2851 * can have problems on various filesystems (FAT, NFS, Coda).
2852 */
2853 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2854 {
2855 int fd, dirlen = directory_size(filename);
2856
2857 if (dirlen + 20 > bufsiz) {
2858 errno = ENAMETOOLONG;
2859 return -1;
2860 }
2861 memcpy(buffer, filename, dirlen);
2862 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2863 fd = git_mkstemp_mode(buffer, 0444);
2864 if (fd < 0 && dirlen && errno == ENOENT) {
2865 /* Make sure the directory exists */
2866 memcpy(buffer, filename, dirlen);
2867 buffer[dirlen-1] = 0;
2868 if (mkdir(buffer, 0777) && errno != EEXIST)
2869 return -1;
2870 if (adjust_shared_perm(buffer))
2871 return -1;
2872
2873 /* Try again */
2874 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2875 fd = git_mkstemp_mode(buffer, 0444);
2876 }
2877 return fd;
2878 }
2879
2880 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2881 const void *buf, unsigned long len, time_t mtime)
2882 {
2883 int fd, ret;
2884 unsigned char compressed[4096];
2885 git_zstream stream;
2886 git_SHA_CTX c;
2887 unsigned char parano_sha1[20];
2888 static char tmp_file[PATH_MAX];
2889 const char *filename = sha1_file_name(sha1);
2890
2891 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2892 if (fd < 0) {
2893 if (errno == EACCES)
2894 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2895 else
2896 return error("unable to create temporary file: %s", strerror(errno));
2897 }
2898
2899 /* Set it up */
2900 memset(&stream, 0, sizeof(stream));
2901 git_deflate_init(&stream, zlib_compression_level);
2902 stream.next_out = compressed;
2903 stream.avail_out = sizeof(compressed);
2904 git_SHA1_Init(&c);
2905
2906 /* First header.. */
2907 stream.next_in = (unsigned char *)hdr;
2908 stream.avail_in = hdrlen;
2909 while (git_deflate(&stream, 0) == Z_OK)
2910 ; /* nothing */
2911 git_SHA1_Update(&c, hdr, hdrlen);
2912
2913 /* Then the data itself.. */
2914 stream.next_in = (void *)buf;
2915 stream.avail_in = len;
2916 do {
2917 unsigned char *in0 = stream.next_in;
2918 ret = git_deflate(&stream, Z_FINISH);
2919 git_SHA1_Update(&c, in0, stream.next_in - in0);
2920 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2921 die("unable to write sha1 file");
2922 stream.next_out = compressed;
2923 stream.avail_out = sizeof(compressed);
2924 } while (ret == Z_OK);
2925
2926 if (ret != Z_STREAM_END)
2927 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2928 ret = git_deflate_end_gently(&stream);
2929 if (ret != Z_OK)
2930 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2931 git_SHA1_Final(parano_sha1, &c);
2932 if (hashcmp(sha1, parano_sha1) != 0)
2933 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2934
2935 close_sha1_file(fd);
2936
2937 if (mtime) {
2938 struct utimbuf utb;
2939 utb.actime = mtime;
2940 utb.modtime = mtime;
2941 if (utime(tmp_file, &utb) < 0)
2942 warning("failed utime() on %s: %s",
2943 tmp_file, strerror(errno));
2944 }
2945
2946 return move_temp_to_file(tmp_file, filename);
2947 }
2948
2949 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2950 {
2951 unsigned char sha1[20];
2952 char hdr[32];
2953 int hdrlen;
2954
2955 /* Normally if we have it in the pack then we do not bother writing
2956 * it out into .git/objects/??/?{38} file.
2957 */
2958 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2959 if (returnsha1)
2960 hashcpy(returnsha1, sha1);
2961 if (has_sha1_file(sha1))
2962 return 0;
2963 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2964 }
2965
2966 int force_object_loose(const unsigned char *sha1, time_t mtime)
2967 {
2968 void *buf;
2969 unsigned long len;
2970 enum object_type type;
2971 char hdr[32];
2972 int hdrlen;
2973 int ret;
2974
2975 if (has_loose_object(sha1))
2976 return 0;
2977 buf = read_packed_sha1(sha1, &type, &len);
2978 if (!buf)
2979 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2980 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2981 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2982 free(buf);
2983
2984 return ret;
2985 }
2986
2987 int has_pack_index(const unsigned char *sha1)
2988 {
2989 struct stat st;
2990 if (stat(sha1_pack_index_name(sha1), &st))
2991 return 0;
2992 return 1;
2993 }
2994
2995 int has_sha1_pack(const unsigned char *sha1)
2996 {
2997 struct pack_entry e;
2998 return find_pack_entry(sha1, &e);
2999 }
3000
3001 int has_sha1_file(const unsigned char *sha1)
3002 {
3003 struct pack_entry e;
3004
3005 if (find_pack_entry(sha1, &e))
3006 return 1;
3007 if (has_loose_object(sha1))
3008 return 1;
3009 reprepare_packed_git();
3010 return find_pack_entry(sha1, &e);
3011 }
3012
3013 static void check_tree(const void *buf, size_t size)
3014 {
3015 struct tree_desc desc;
3016 struct name_entry entry;
3017
3018 init_tree_desc(&desc, buf, size);
3019 while (tree_entry(&desc, &entry))
3020 /* do nothing
3021 * tree_entry() will die() on malformed entries */
3022 ;
3023 }
3024
3025 static void check_commit(const void *buf, size_t size)
3026 {
3027 struct commit c;
3028 memset(&c, 0, sizeof(c));
3029 if (parse_commit_buffer(&c, buf, size))
3030 die("corrupt commit");
3031 }
3032
3033 static void check_tag(const void *buf, size_t size)
3034 {
3035 struct tag t;
3036 memset(&t, 0, sizeof(t));
3037 if (parse_tag_buffer(&t, buf, size))
3038 die("corrupt tag");
3039 }
3040
3041 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3042 enum object_type type,
3043 const char *path, unsigned flags)
3044 {
3045 int ret, re_allocated = 0;
3046 int write_object = flags & HASH_WRITE_OBJECT;
3047
3048 if (!type)
3049 type = OBJ_BLOB;
3050
3051 /*
3052 * Convert blobs to git internal format
3053 */
3054 if ((type == OBJ_BLOB) && path) {
3055 struct strbuf nbuf = STRBUF_INIT;
3056 if (convert_to_git(path, buf, size, &nbuf,
3057 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3058 buf = strbuf_detach(&nbuf, &size);
3059 re_allocated = 1;
3060 }
3061 }
3062 if (flags & HASH_FORMAT_CHECK) {
3063 if (type == OBJ_TREE)
3064 check_tree(buf, size);
3065 if (type == OBJ_COMMIT)
3066 check_commit(buf, size);
3067 if (type == OBJ_TAG)
3068 check_tag(buf, size);
3069 }
3070
3071 if (write_object)
3072 ret = write_sha1_file(buf, size, typename(type), sha1);
3073 else
3074 ret = hash_sha1_file(buf, size, typename(type), sha1);
3075 if (re_allocated)
3076 free(buf);
3077 return ret;
3078 }
3079
3080 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3081 const char *path, unsigned flags)
3082 {
3083 struct strbuf sbuf = STRBUF_INIT;
3084 int ret;
3085
3086 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3087 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3088 else
3089 ret = -1;
3090 strbuf_release(&sbuf);
3091 return ret;
3092 }
3093
3094 #define SMALL_FILE_SIZE (32*1024)
3095
3096 static int index_core(unsigned char *sha1, int fd, size_t size,
3097 enum object_type type, const char *path,
3098 unsigned flags)
3099 {
3100 int ret;
3101
3102 if (!size) {
3103 ret = index_mem(sha1, NULL, size, type, path, flags);
3104 } else if (size <= SMALL_FILE_SIZE) {
3105 char *buf = xmalloc(size);
3106 if (size == read_in_full(fd, buf, size))
3107 ret = index_mem(sha1, buf, size, type, path, flags);
3108 else
3109 ret = error("short read %s", strerror(errno));
3110 free(buf);
3111 } else {
3112 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3113 ret = index_mem(sha1, buf, size, type, path, flags);
3114 munmap(buf, size);
3115 }
3116 return ret;
3117 }
3118
3119 /*
3120 * This creates one packfile per large blob unless bulk-checkin
3121 * machinery is "plugged".
3122 *
3123 * This also bypasses the usual "convert-to-git" dance, and that is on
3124 * purpose. We could write a streaming version of the converting
3125 * functions and insert that before feeding the data to fast-import
3126 * (or equivalent in-core API described above). However, that is
3127 * somewhat complicated, as we do not know the size of the filter
3128 * result, which we need to know beforehand when writing a git object.
3129 * Since the primary motivation for trying to stream from the working
3130 * tree file and to avoid mmaping it in core is to deal with large
3131 * binary blobs, they generally do not want to get any conversion, and
3132 * callers should avoid this code path when filters are requested.
3133 */
3134 static int index_stream(unsigned char *sha1, int fd, size_t size,
3135 enum object_type type, const char *path,
3136 unsigned flags)
3137 {
3138 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3139 }
3140
3141 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3142 enum object_type type, const char *path, unsigned flags)
3143 {
3144 int ret;
3145 size_t size = xsize_t(st->st_size);
3146
3147 if (!S_ISREG(st->st_mode))
3148 ret = index_pipe(sha1, fd, type, path, flags);
3149 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3150 (path && would_convert_to_git(path, NULL, 0, 0)))
3151 ret = index_core(sha1, fd, size, type, path, flags);
3152 else
3153 ret = index_stream(sha1, fd, size, type, path, flags);
3154 close(fd);
3155 return ret;
3156 }
3157
3158 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3159 {
3160 int fd;
3161 struct strbuf sb = STRBUF_INIT;
3162
3163 switch (st->st_mode & S_IFMT) {
3164 case S_IFREG:
3165 fd = open(path, O_RDONLY);
3166 if (fd < 0)
3167 return error("open(\"%s\"): %s", path,
3168 strerror(errno));
3169 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3170 return error("%s: failed to insert into database",
3171 path);
3172 break;
3173 case S_IFLNK:
3174 if (strbuf_readlink(&sb, path, st->st_size)) {
3175 char *errstr = strerror(errno);
3176 return error("readlink(\"%s\"): %s", path,
3177 errstr);
3178 }
3179 if (!(flags & HASH_WRITE_OBJECT))
3180 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3181 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3182 return error("%s: failed to insert into database",
3183 path);
3184 strbuf_release(&sb);
3185 break;
3186 case S_IFDIR:
3187 return resolve_gitlink_ref(path, "HEAD", sha1);
3188 default:
3189 return error("%s: unsupported file type", path);
3190 }
3191 return 0;
3192 }
3193
3194 int read_pack_header(int fd, struct pack_header *header)
3195 {
3196 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3197 /* "eof before pack header was fully read" */
3198 return PH_ERROR_EOF;
3199
3200 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3201 /* "protocol error (pack signature mismatch detected)" */
3202 return PH_ERROR_PACK_SIGNATURE;
3203 if (!pack_version_ok(header->hdr_version))
3204 /* "protocol error (pack version unsupported)" */
3205 return PH_ERROR_PROTOCOL;
3206 return 0;
3207 }
3208
3209 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3210 {
3211 enum object_type type = sha1_object_info(sha1, NULL);
3212 if (type < 0)
3213 die("%s is not a valid object", sha1_to_hex(sha1));
3214 if (type != expect)
3215 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3216 typename(expect));
3217 }