Merge branch 'jk/fix-refresh-utime' into maint
[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 int foreach_alt_odb(alt_odb_fn fn, void *cb)
417 {
418 struct alternate_object_database *ent;
419 int r = 0;
420
421 prepare_alt_odb();
422 for (ent = alt_odb_list; ent; ent = ent->next) {
423 r = fn(ent, cb);
424 if (r)
425 break;
426 }
427 return r;
428 }
429
430 void prepare_alt_odb(void)
431 {
432 const char *alt;
433
434 if (alt_odb_tail)
435 return;
436
437 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
438 if (!alt) alt = "";
439
440 alt_odb_tail = &alt_odb_list;
441 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
442
443 read_info_alternates(get_object_directory(), 0);
444 }
445
446 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
447 static int freshen_file(const char *fn)
448 {
449 struct utimbuf t;
450 t.actime = t.modtime = time(NULL);
451 return !utime(fn, &t);
452 }
453
454 /*
455 * All of the check_and_freshen functions return 1 if the file exists and was
456 * freshened (if freshening was requested), 0 otherwise. If they return
457 * 0, you should not assume that it is safe to skip a write of the object (it
458 * either does not exist on disk, or has a stale mtime and may be subject to
459 * pruning).
460 */
461 static int check_and_freshen_file(const char *fn, int freshen)
462 {
463 if (access(fn, F_OK))
464 return 0;
465 if (freshen && !freshen_file(fn))
466 return 0;
467 return 1;
468 }
469
470 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
471 {
472 return check_and_freshen_file(sha1_file_name(sha1), freshen);
473 }
474
475 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
476 {
477 struct alternate_object_database *alt;
478 prepare_alt_odb();
479 for (alt = alt_odb_list; alt; alt = alt->next) {
480 fill_sha1_path(alt->name, sha1);
481 if (check_and_freshen_file(alt->base, freshen))
482 return 1;
483 }
484 return 0;
485 }
486
487 static int check_and_freshen(const unsigned char *sha1, int freshen)
488 {
489 return check_and_freshen_local(sha1, freshen) ||
490 check_and_freshen_nonlocal(sha1, freshen);
491 }
492
493 int has_loose_object_nonlocal(const unsigned char *sha1)
494 {
495 return check_and_freshen_nonlocal(sha1, 0);
496 }
497
498 static int has_loose_object(const unsigned char *sha1)
499 {
500 return check_and_freshen(sha1, 0);
501 }
502
503 static unsigned int pack_used_ctr;
504 static unsigned int pack_mmap_calls;
505 static unsigned int peak_pack_open_windows;
506 static unsigned int pack_open_windows;
507 static unsigned int pack_open_fds;
508 static unsigned int pack_max_fds;
509 static size_t peak_pack_mapped;
510 static size_t pack_mapped;
511 struct packed_git *packed_git;
512
513 void pack_report(void)
514 {
515 fprintf(stderr,
516 "pack_report: getpagesize() = %10" SZ_FMT "\n"
517 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
518 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
519 sz_fmt(getpagesize()),
520 sz_fmt(packed_git_window_size),
521 sz_fmt(packed_git_limit));
522 fprintf(stderr,
523 "pack_report: pack_used_ctr = %10u\n"
524 "pack_report: pack_mmap_calls = %10u\n"
525 "pack_report: pack_open_windows = %10u / %10u\n"
526 "pack_report: pack_mapped = "
527 "%10" SZ_FMT " / %10" SZ_FMT "\n",
528 pack_used_ctr,
529 pack_mmap_calls,
530 pack_open_windows, peak_pack_open_windows,
531 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
532 }
533
534 /*
535 * Open and mmap the index file at path, perform a couple of
536 * consistency checks, then record its information to p. Return 0 on
537 * success.
538 */
539 static int check_packed_git_idx(const char *path, struct packed_git *p)
540 {
541 void *idx_map;
542 struct pack_idx_header *hdr;
543 size_t idx_size;
544 uint32_t version, nr, i, *index;
545 int fd = git_open_noatime(path);
546 struct stat st;
547
548 if (fd < 0)
549 return -1;
550 if (fstat(fd, &st)) {
551 close(fd);
552 return -1;
553 }
554 idx_size = xsize_t(st.st_size);
555 if (idx_size < 4 * 256 + 20 + 20) {
556 close(fd);
557 return error("index file %s is too small", path);
558 }
559 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
560 close(fd);
561
562 hdr = idx_map;
563 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
564 version = ntohl(hdr->idx_version);
565 if (version < 2 || version > 2) {
566 munmap(idx_map, idx_size);
567 return error("index file %s is version %"PRIu32
568 " and is not supported by this binary"
569 " (try upgrading GIT to a newer version)",
570 path, version);
571 }
572 } else
573 version = 1;
574
575 nr = 0;
576 index = idx_map;
577 if (version > 1)
578 index += 2; /* skip index header */
579 for (i = 0; i < 256; i++) {
580 uint32_t n = ntohl(index[i]);
581 if (n < nr) {
582 munmap(idx_map, idx_size);
583 return error("non-monotonic index %s", path);
584 }
585 nr = n;
586 }
587
588 if (version == 1) {
589 /*
590 * Total size:
591 * - 256 index entries 4 bytes each
592 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
593 * - 20-byte SHA1 of the packfile
594 * - 20-byte SHA1 file checksum
595 */
596 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
597 munmap(idx_map, idx_size);
598 return error("wrong index v1 file size in %s", path);
599 }
600 } else if (version == 2) {
601 /*
602 * Minimum size:
603 * - 8 bytes of header
604 * - 256 index entries 4 bytes each
605 * - 20-byte sha1 entry * nr
606 * - 4-byte crc entry * nr
607 * - 4-byte offset entry * nr
608 * - 20-byte SHA1 of the packfile
609 * - 20-byte SHA1 file checksum
610 * And after the 4-byte offset table might be a
611 * variable sized table containing 8-byte entries
612 * for offsets larger than 2^31.
613 */
614 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
615 unsigned long max_size = min_size;
616 if (nr)
617 max_size += (nr - 1)*8;
618 if (idx_size < min_size || idx_size > max_size) {
619 munmap(idx_map, idx_size);
620 return error("wrong index v2 file size in %s", path);
621 }
622 if (idx_size != min_size &&
623 /*
624 * make sure we can deal with large pack offsets.
625 * 31-bit signed offset won't be enough, neither
626 * 32-bit unsigned one will be.
627 */
628 (sizeof(off_t) <= 4)) {
629 munmap(idx_map, idx_size);
630 return error("pack too large for current definition of off_t in %s", path);
631 }
632 }
633
634 p->index_version = version;
635 p->index_data = idx_map;
636 p->index_size = idx_size;
637 p->num_objects = nr;
638 return 0;
639 }
640
641 int open_pack_index(struct packed_git *p)
642 {
643 char *idx_name;
644 int ret;
645
646 if (p->index_data)
647 return 0;
648
649 idx_name = xstrdup(p->pack_name);
650 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
651 ret = check_packed_git_idx(idx_name, p);
652 free(idx_name);
653 return ret;
654 }
655
656 static void scan_windows(struct packed_git *p,
657 struct packed_git **lru_p,
658 struct pack_window **lru_w,
659 struct pack_window **lru_l)
660 {
661 struct pack_window *w, *w_l;
662
663 for (w_l = NULL, w = p->windows; w; w = w->next) {
664 if (!w->inuse_cnt) {
665 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
666 *lru_p = p;
667 *lru_w = w;
668 *lru_l = w_l;
669 }
670 }
671 w_l = w;
672 }
673 }
674
675 static int unuse_one_window(struct packed_git *current)
676 {
677 struct packed_git *p, *lru_p = NULL;
678 struct pack_window *lru_w = NULL, *lru_l = NULL;
679
680 if (current)
681 scan_windows(current, &lru_p, &lru_w, &lru_l);
682 for (p = packed_git; p; p = p->next)
683 scan_windows(p, &lru_p, &lru_w, &lru_l);
684 if (lru_p) {
685 munmap(lru_w->base, lru_w->len);
686 pack_mapped -= lru_w->len;
687 if (lru_l)
688 lru_l->next = lru_w->next;
689 else
690 lru_p->windows = lru_w->next;
691 free(lru_w);
692 pack_open_windows--;
693 return 1;
694 }
695 return 0;
696 }
697
698 void release_pack_memory(size_t need)
699 {
700 size_t cur = pack_mapped;
701 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
702 ; /* nothing */
703 }
704
705 static void mmap_limit_check(size_t length)
706 {
707 static size_t limit = 0;
708 if (!limit) {
709 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
710 if (!limit)
711 limit = SIZE_MAX;
712 }
713 if (length > limit)
714 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
715 (uintmax_t)length, (uintmax_t)limit);
716 }
717
718 void *xmmap_gently(void *start, size_t length,
719 int prot, int flags, int fd, off_t offset)
720 {
721 void *ret;
722
723 mmap_limit_check(length);
724 ret = mmap(start, length, prot, flags, fd, offset);
725 if (ret == MAP_FAILED) {
726 if (!length)
727 return NULL;
728 release_pack_memory(length);
729 ret = mmap(start, length, prot, flags, fd, offset);
730 }
731 return ret;
732 }
733
734 void *xmmap(void *start, size_t length,
735 int prot, int flags, int fd, off_t offset)
736 {
737 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
738 if (ret == MAP_FAILED)
739 die_errno("mmap failed");
740 return ret;
741 }
742
743 void close_pack_windows(struct packed_git *p)
744 {
745 while (p->windows) {
746 struct pack_window *w = p->windows;
747
748 if (w->inuse_cnt)
749 die("pack '%s' still has open windows to it",
750 p->pack_name);
751 munmap(w->base, w->len);
752 pack_mapped -= w->len;
753 pack_open_windows--;
754 p->windows = w->next;
755 free(w);
756 }
757 }
758
759 /*
760 * The LRU pack is the one with the oldest MRU window, preferring packs
761 * with no used windows, or the oldest mtime if it has no windows allocated.
762 */
763 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
764 {
765 struct pack_window *w, *this_mru_w;
766 int has_windows_inuse = 0;
767
768 /*
769 * Reject this pack if it has windows and the previously selected
770 * one does not. If this pack does not have windows, reject
771 * it if the pack file is newer than the previously selected one.
772 */
773 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
774 return;
775
776 for (w = this_mru_w = p->windows; w; w = w->next) {
777 /*
778 * Reject this pack if any of its windows are in use,
779 * but the previously selected pack did not have any
780 * inuse windows. Otherwise, record that this pack
781 * has windows in use.
782 */
783 if (w->inuse_cnt) {
784 if (*accept_windows_inuse)
785 has_windows_inuse = 1;
786 else
787 return;
788 }
789
790 if (w->last_used > this_mru_w->last_used)
791 this_mru_w = w;
792
793 /*
794 * Reject this pack if it has windows that have been
795 * used more recently than the previously selected pack.
796 * If the previously selected pack had windows inuse and
797 * we have not encountered a window in this pack that is
798 * inuse, skip this check since we prefer a pack with no
799 * inuse windows to one that has inuse windows.
800 */
801 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
802 this_mru_w->last_used > (*mru_w)->last_used)
803 return;
804 }
805
806 /*
807 * Select this pack.
808 */
809 *mru_w = this_mru_w;
810 *lru_p = p;
811 *accept_windows_inuse = has_windows_inuse;
812 }
813
814 static int close_one_pack(void)
815 {
816 struct packed_git *p, *lru_p = NULL;
817 struct pack_window *mru_w = NULL;
818 int accept_windows_inuse = 1;
819
820 for (p = packed_git; p; p = p->next) {
821 if (p->pack_fd == -1)
822 continue;
823 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
824 }
825
826 if (lru_p) {
827 close(lru_p->pack_fd);
828 pack_open_fds--;
829 lru_p->pack_fd = -1;
830 return 1;
831 }
832
833 return 0;
834 }
835
836 void unuse_pack(struct pack_window **w_cursor)
837 {
838 struct pack_window *w = *w_cursor;
839 if (w) {
840 w->inuse_cnt--;
841 *w_cursor = NULL;
842 }
843 }
844
845 void close_pack_index(struct packed_git *p)
846 {
847 if (p->index_data) {
848 munmap((void *)p->index_data, p->index_size);
849 p->index_data = NULL;
850 }
851 }
852
853 /*
854 * This is used by git-repack in case a newly created pack happens to
855 * contain the same set of objects as an existing one. In that case
856 * the resulting file might be different even if its name would be the
857 * same. It is best to close any reference to the old pack before it is
858 * replaced on disk. Of course no index pointers or windows for given pack
859 * must subsist at this point. If ever objects from this pack are requested
860 * again, the new version of the pack will be reinitialized through
861 * reprepare_packed_git().
862 */
863 void free_pack_by_name(const char *pack_name)
864 {
865 struct packed_git *p, **pp = &packed_git;
866
867 while (*pp) {
868 p = *pp;
869 if (strcmp(pack_name, p->pack_name) == 0) {
870 clear_delta_base_cache();
871 close_pack_windows(p);
872 if (p->pack_fd != -1) {
873 close(p->pack_fd);
874 pack_open_fds--;
875 }
876 close_pack_index(p);
877 free(p->bad_object_sha1);
878 *pp = p->next;
879 if (last_found_pack == p)
880 last_found_pack = NULL;
881 free(p);
882 return;
883 }
884 pp = &p->next;
885 }
886 }
887
888 static unsigned int get_max_fd_limit(void)
889 {
890 #ifdef RLIMIT_NOFILE
891 {
892 struct rlimit lim;
893
894 if (!getrlimit(RLIMIT_NOFILE, &lim))
895 return lim.rlim_cur;
896 }
897 #endif
898
899 #ifdef _SC_OPEN_MAX
900 {
901 long open_max = sysconf(_SC_OPEN_MAX);
902 if (0 < open_max)
903 return open_max;
904 /*
905 * Otherwise, we got -1 for one of the two
906 * reasons:
907 *
908 * (1) sysconf() did not understand _SC_OPEN_MAX
909 * and signaled an error with -1; or
910 * (2) sysconf() said there is no limit.
911 *
912 * We _could_ clear errno before calling sysconf() to
913 * tell these two cases apart and return a huge number
914 * in the latter case to let the caller cap it to a
915 * value that is not so selfish, but letting the
916 * fallback OPEN_MAX codepath take care of these cases
917 * is a lot simpler.
918 */
919 }
920 #endif
921
922 #ifdef OPEN_MAX
923 return OPEN_MAX;
924 #else
925 return 1; /* see the caller ;-) */
926 #endif
927 }
928
929 /*
930 * Do not call this directly as this leaks p->pack_fd on error return;
931 * call open_packed_git() instead.
932 */
933 static int open_packed_git_1(struct packed_git *p)
934 {
935 struct stat st;
936 struct pack_header hdr;
937 unsigned char sha1[20];
938 unsigned char *idx_sha1;
939 long fd_flag;
940
941 if (!p->index_data && open_pack_index(p))
942 return error("packfile %s index unavailable", p->pack_name);
943
944 if (!pack_max_fds) {
945 unsigned int max_fds = get_max_fd_limit();
946
947 /* Save 3 for stdin/stdout/stderr, 22 for work */
948 if (25 < max_fds)
949 pack_max_fds = max_fds - 25;
950 else
951 pack_max_fds = 1;
952 }
953
954 while (pack_max_fds <= pack_open_fds && close_one_pack())
955 ; /* nothing */
956
957 p->pack_fd = git_open_noatime(p->pack_name);
958 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
959 return -1;
960 pack_open_fds++;
961
962 /* If we created the struct before we had the pack we lack size. */
963 if (!p->pack_size) {
964 if (!S_ISREG(st.st_mode))
965 return error("packfile %s not a regular file", p->pack_name);
966 p->pack_size = st.st_size;
967 } else if (p->pack_size != st.st_size)
968 return error("packfile %s size changed", p->pack_name);
969
970 /* We leave these file descriptors open with sliding mmap;
971 * there is no point keeping them open across exec(), though.
972 */
973 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
974 if (fd_flag < 0)
975 return error("cannot determine file descriptor flags");
976 fd_flag |= FD_CLOEXEC;
977 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
978 return error("cannot set FD_CLOEXEC");
979
980 /* Verify we recognize this pack file format. */
981 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
982 return error("file %s is far too short to be a packfile", p->pack_name);
983 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
984 return error("file %s is not a GIT packfile", p->pack_name);
985 if (!pack_version_ok(hdr.hdr_version))
986 return error("packfile %s is version %"PRIu32" and not"
987 " supported (try upgrading GIT to a newer version)",
988 p->pack_name, ntohl(hdr.hdr_version));
989
990 /* Verify the pack matches its index. */
991 if (p->num_objects != ntohl(hdr.hdr_entries))
992 return error("packfile %s claims to have %"PRIu32" objects"
993 " while index indicates %"PRIu32" objects",
994 p->pack_name, ntohl(hdr.hdr_entries),
995 p->num_objects);
996 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
997 return error("end of packfile %s is unavailable", p->pack_name);
998 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
999 return error("packfile %s signature is unavailable", p->pack_name);
1000 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1001 if (hashcmp(sha1, idx_sha1))
1002 return error("packfile %s does not match index", p->pack_name);
1003 return 0;
1004 }
1005
1006 static int open_packed_git(struct packed_git *p)
1007 {
1008 if (!open_packed_git_1(p))
1009 return 0;
1010 if (p->pack_fd != -1) {
1011 close(p->pack_fd);
1012 pack_open_fds--;
1013 p->pack_fd = -1;
1014 }
1015 return -1;
1016 }
1017
1018 static int in_window(struct pack_window *win, off_t offset)
1019 {
1020 /* We must promise at least 20 bytes (one hash) after the
1021 * offset is available from this window, otherwise the offset
1022 * is not actually in this window and a different window (which
1023 * has that one hash excess) must be used. This is to support
1024 * the object header and delta base parsing routines below.
1025 */
1026 off_t win_off = win->offset;
1027 return win_off <= offset
1028 && (offset + 20) <= (win_off + win->len);
1029 }
1030
1031 unsigned char *use_pack(struct packed_git *p,
1032 struct pack_window **w_cursor,
1033 off_t offset,
1034 unsigned long *left)
1035 {
1036 struct pack_window *win = *w_cursor;
1037
1038 /* Since packfiles end in a hash of their content and it's
1039 * pointless to ask for an offset into the middle of that
1040 * hash, and the in_window function above wouldn't match
1041 * don't allow an offset too close to the end of the file.
1042 */
1043 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1044 die("packfile %s cannot be accessed", p->pack_name);
1045 if (offset > (p->pack_size - 20))
1046 die("offset beyond end of packfile (truncated pack?)");
1047
1048 if (!win || !in_window(win, offset)) {
1049 if (win)
1050 win->inuse_cnt--;
1051 for (win = p->windows; win; win = win->next) {
1052 if (in_window(win, offset))
1053 break;
1054 }
1055 if (!win) {
1056 size_t window_align = packed_git_window_size / 2;
1057 off_t len;
1058
1059 if (p->pack_fd == -1 && open_packed_git(p))
1060 die("packfile %s cannot be accessed", p->pack_name);
1061
1062 win = xcalloc(1, sizeof(*win));
1063 win->offset = (offset / window_align) * window_align;
1064 len = p->pack_size - win->offset;
1065 if (len > packed_git_window_size)
1066 len = packed_git_window_size;
1067 win->len = (size_t)len;
1068 pack_mapped += win->len;
1069 while (packed_git_limit < pack_mapped
1070 && unuse_one_window(p))
1071 ; /* nothing */
1072 win->base = xmmap(NULL, win->len,
1073 PROT_READ, MAP_PRIVATE,
1074 p->pack_fd, win->offset);
1075 if (win->base == MAP_FAILED)
1076 die("packfile %s cannot be mapped: %s",
1077 p->pack_name,
1078 strerror(errno));
1079 if (!win->offset && win->len == p->pack_size
1080 && !p->do_not_close) {
1081 close(p->pack_fd);
1082 pack_open_fds--;
1083 p->pack_fd = -1;
1084 }
1085 pack_mmap_calls++;
1086 pack_open_windows++;
1087 if (pack_mapped > peak_pack_mapped)
1088 peak_pack_mapped = pack_mapped;
1089 if (pack_open_windows > peak_pack_open_windows)
1090 peak_pack_open_windows = pack_open_windows;
1091 win->next = p->windows;
1092 p->windows = win;
1093 }
1094 }
1095 if (win != *w_cursor) {
1096 win->last_used = pack_used_ctr++;
1097 win->inuse_cnt++;
1098 *w_cursor = win;
1099 }
1100 offset -= win->offset;
1101 if (left)
1102 *left = win->len - xsize_t(offset);
1103 return win->base + offset;
1104 }
1105
1106 static struct packed_git *alloc_packed_git(int extra)
1107 {
1108 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1109 memset(p, 0, sizeof(*p));
1110 p->pack_fd = -1;
1111 return p;
1112 }
1113
1114 static void try_to_free_pack_memory(size_t size)
1115 {
1116 release_pack_memory(size);
1117 }
1118
1119 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1120 {
1121 static int have_set_try_to_free_routine;
1122 struct stat st;
1123 struct packed_git *p = alloc_packed_git(path_len + 2);
1124
1125 if (!have_set_try_to_free_routine) {
1126 have_set_try_to_free_routine = 1;
1127 set_try_to_free_routine(try_to_free_pack_memory);
1128 }
1129
1130 /*
1131 * Make sure a corresponding .pack file exists and that
1132 * the index looks sane.
1133 */
1134 path_len -= strlen(".idx");
1135 if (path_len < 1) {
1136 free(p);
1137 return NULL;
1138 }
1139 memcpy(p->pack_name, path, path_len);
1140
1141 strcpy(p->pack_name + path_len, ".keep");
1142 if (!access(p->pack_name, F_OK))
1143 p->pack_keep = 1;
1144
1145 strcpy(p->pack_name + path_len, ".pack");
1146 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1147 free(p);
1148 return NULL;
1149 }
1150
1151 /* ok, it looks sane as far as we can check without
1152 * actually mapping the pack file.
1153 */
1154 p->pack_size = st.st_size;
1155 p->pack_local = local;
1156 p->mtime = st.st_mtime;
1157 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1158 hashclr(p->sha1);
1159 return p;
1160 }
1161
1162 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1163 {
1164 const char *path = sha1_pack_name(sha1);
1165 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1166
1167 strcpy(p->pack_name, path);
1168 hashcpy(p->sha1, sha1);
1169 if (check_packed_git_idx(idx_path, p)) {
1170 free(p);
1171 return NULL;
1172 }
1173
1174 return p;
1175 }
1176
1177 void install_packed_git(struct packed_git *pack)
1178 {
1179 if (pack->pack_fd != -1)
1180 pack_open_fds++;
1181
1182 pack->next = packed_git;
1183 packed_git = pack;
1184 }
1185
1186 void (*report_garbage)(const char *desc, const char *path);
1187
1188 static void report_helper(const struct string_list *list,
1189 int seen_bits, int first, int last)
1190 {
1191 const char *msg;
1192 switch (seen_bits) {
1193 case 0:
1194 msg = "no corresponding .idx or .pack";
1195 break;
1196 case 1:
1197 msg = "no corresponding .idx";
1198 break;
1199 case 2:
1200 msg = "no corresponding .pack";
1201 break;
1202 default:
1203 return;
1204 }
1205 for (; first < last; first++)
1206 report_garbage(msg, list->items[first].string);
1207 }
1208
1209 static void report_pack_garbage(struct string_list *list)
1210 {
1211 int i, baselen = -1, first = 0, seen_bits = 0;
1212
1213 if (!report_garbage)
1214 return;
1215
1216 string_list_sort(list);
1217
1218 for (i = 0; i < list->nr; i++) {
1219 const char *path = list->items[i].string;
1220 if (baselen != -1 &&
1221 strncmp(path, list->items[first].string, baselen)) {
1222 report_helper(list, seen_bits, first, i);
1223 baselen = -1;
1224 seen_bits = 0;
1225 }
1226 if (baselen == -1) {
1227 const char *dot = strrchr(path, '.');
1228 if (!dot) {
1229 report_garbage("garbage found", path);
1230 continue;
1231 }
1232 baselen = dot - path + 1;
1233 first = i;
1234 }
1235 if (!strcmp(path + baselen, "pack"))
1236 seen_bits |= 1;
1237 else if (!strcmp(path + baselen, "idx"))
1238 seen_bits |= 2;
1239 }
1240 report_helper(list, seen_bits, first, list->nr);
1241 }
1242
1243 static void prepare_packed_git_one(char *objdir, int local)
1244 {
1245 struct strbuf path = STRBUF_INIT;
1246 size_t dirnamelen;
1247 DIR *dir;
1248 struct dirent *de;
1249 struct string_list garbage = STRING_LIST_INIT_DUP;
1250
1251 strbuf_addstr(&path, objdir);
1252 strbuf_addstr(&path, "/pack");
1253 dir = opendir(path.buf);
1254 if (!dir) {
1255 if (errno != ENOENT)
1256 error("unable to open object pack directory: %s: %s",
1257 path.buf, strerror(errno));
1258 strbuf_release(&path);
1259 return;
1260 }
1261 strbuf_addch(&path, '/');
1262 dirnamelen = path.len;
1263 while ((de = readdir(dir)) != NULL) {
1264 struct packed_git *p;
1265 size_t base_len;
1266
1267 if (is_dot_or_dotdot(de->d_name))
1268 continue;
1269
1270 strbuf_setlen(&path, dirnamelen);
1271 strbuf_addstr(&path, de->d_name);
1272
1273 base_len = path.len;
1274 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1275 /* Don't reopen a pack we already have. */
1276 for (p = packed_git; p; p = p->next) {
1277 size_t len;
1278 if (strip_suffix(p->pack_name, ".pack", &len) &&
1279 len == base_len &&
1280 !memcmp(p->pack_name, path.buf, len))
1281 break;
1282 }
1283 if (p == NULL &&
1284 /*
1285 * See if it really is a valid .idx file with
1286 * corresponding .pack file that we can map.
1287 */
1288 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1289 install_packed_git(p);
1290 }
1291
1292 if (!report_garbage)
1293 continue;
1294
1295 if (ends_with(de->d_name, ".idx") ||
1296 ends_with(de->d_name, ".pack") ||
1297 ends_with(de->d_name, ".bitmap") ||
1298 ends_with(de->d_name, ".keep"))
1299 string_list_append(&garbage, path.buf);
1300 else
1301 report_garbage("garbage found", path.buf);
1302 }
1303 closedir(dir);
1304 report_pack_garbage(&garbage);
1305 string_list_clear(&garbage, 0);
1306 strbuf_release(&path);
1307 }
1308
1309 static int sort_pack(const void *a_, const void *b_)
1310 {
1311 struct packed_git *a = *((struct packed_git **)a_);
1312 struct packed_git *b = *((struct packed_git **)b_);
1313 int st;
1314
1315 /*
1316 * Local packs tend to contain objects specific to our
1317 * variant of the project than remote ones. In addition,
1318 * remote ones could be on a network mounted filesystem.
1319 * Favor local ones for these reasons.
1320 */
1321 st = a->pack_local - b->pack_local;
1322 if (st)
1323 return -st;
1324
1325 /*
1326 * Younger packs tend to contain more recent objects,
1327 * and more recent objects tend to get accessed more
1328 * often.
1329 */
1330 if (a->mtime < b->mtime)
1331 return 1;
1332 else if (a->mtime == b->mtime)
1333 return 0;
1334 return -1;
1335 }
1336
1337 static void rearrange_packed_git(void)
1338 {
1339 struct packed_git **ary, *p;
1340 int i, n;
1341
1342 for (n = 0, p = packed_git; p; p = p->next)
1343 n++;
1344 if (n < 2)
1345 return;
1346
1347 /* prepare an array of packed_git for easier sorting */
1348 ary = xcalloc(n, sizeof(struct packed_git *));
1349 for (n = 0, p = packed_git; p; p = p->next)
1350 ary[n++] = p;
1351
1352 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1353
1354 /* link them back again */
1355 for (i = 0; i < n - 1; i++)
1356 ary[i]->next = ary[i + 1];
1357 ary[n - 1]->next = NULL;
1358 packed_git = ary[0];
1359
1360 free(ary);
1361 }
1362
1363 static int prepare_packed_git_run_once = 0;
1364 void prepare_packed_git(void)
1365 {
1366 struct alternate_object_database *alt;
1367
1368 if (prepare_packed_git_run_once)
1369 return;
1370 prepare_packed_git_one(get_object_directory(), 1);
1371 prepare_alt_odb();
1372 for (alt = alt_odb_list; alt; alt = alt->next) {
1373 alt->name[-1] = 0;
1374 prepare_packed_git_one(alt->base, 0);
1375 alt->name[-1] = '/';
1376 }
1377 rearrange_packed_git();
1378 prepare_packed_git_run_once = 1;
1379 }
1380
1381 void reprepare_packed_git(void)
1382 {
1383 prepare_packed_git_run_once = 0;
1384 prepare_packed_git();
1385 }
1386
1387 static void mark_bad_packed_object(struct packed_git *p,
1388 const unsigned char *sha1)
1389 {
1390 unsigned i;
1391 for (i = 0; i < p->num_bad_objects; i++)
1392 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1393 return;
1394 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1395 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1396 p->num_bad_objects++;
1397 }
1398
1399 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1400 {
1401 struct packed_git *p;
1402 unsigned i;
1403
1404 for (p = packed_git; p; p = p->next)
1405 for (i = 0; i < p->num_bad_objects; i++)
1406 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1407 return p;
1408 return NULL;
1409 }
1410
1411 /*
1412 * With an in-core object data in "map", rehash it to make sure the
1413 * object name actually matches "sha1" to detect object corruption.
1414 * With "map" == NULL, try reading the object named with "sha1" using
1415 * the streaming interface and rehash it to do the same.
1416 */
1417 int check_sha1_signature(const unsigned char *sha1, void *map,
1418 unsigned long size, const char *type)
1419 {
1420 unsigned char real_sha1[20];
1421 enum object_type obj_type;
1422 struct git_istream *st;
1423 git_SHA_CTX c;
1424 char hdr[32];
1425 int hdrlen;
1426
1427 if (map) {
1428 hash_sha1_file(map, size, type, real_sha1);
1429 return hashcmp(sha1, real_sha1) ? -1 : 0;
1430 }
1431
1432 st = open_istream(sha1, &obj_type, &size, NULL);
1433 if (!st)
1434 return -1;
1435
1436 /* Generate the header */
1437 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1438
1439 /* Sha1.. */
1440 git_SHA1_Init(&c);
1441 git_SHA1_Update(&c, hdr, hdrlen);
1442 for (;;) {
1443 char buf[1024 * 16];
1444 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1445
1446 if (readlen < 0) {
1447 close_istream(st);
1448 return -1;
1449 }
1450 if (!readlen)
1451 break;
1452 git_SHA1_Update(&c, buf, readlen);
1453 }
1454 git_SHA1_Final(real_sha1, &c);
1455 close_istream(st);
1456 return hashcmp(sha1, real_sha1) ? -1 : 0;
1457 }
1458
1459 int git_open_noatime(const char *name)
1460 {
1461 static int sha1_file_open_flag = O_NOATIME;
1462
1463 for (;;) {
1464 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1465 if (fd >= 0)
1466 return fd;
1467
1468 /* Might the failure be due to O_NOATIME? */
1469 if (errno != ENOENT && sha1_file_open_flag) {
1470 sha1_file_open_flag = 0;
1471 continue;
1472 }
1473
1474 return -1;
1475 }
1476 }
1477
1478 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1479 {
1480 struct alternate_object_database *alt;
1481
1482 if (!lstat(sha1_file_name(sha1), st))
1483 return 0;
1484
1485 prepare_alt_odb();
1486 errno = ENOENT;
1487 for (alt = alt_odb_list; alt; alt = alt->next) {
1488 fill_sha1_path(alt->name, sha1);
1489 if (!lstat(alt->base, st))
1490 return 0;
1491 }
1492
1493 return -1;
1494 }
1495
1496 static int open_sha1_file(const unsigned char *sha1)
1497 {
1498 int fd;
1499 struct alternate_object_database *alt;
1500 int most_interesting_errno;
1501
1502 fd = git_open_noatime(sha1_file_name(sha1));
1503 if (fd >= 0)
1504 return fd;
1505 most_interesting_errno = errno;
1506
1507 prepare_alt_odb();
1508 for (alt = alt_odb_list; alt; alt = alt->next) {
1509 fill_sha1_path(alt->name, sha1);
1510 fd = git_open_noatime(alt->base);
1511 if (fd >= 0)
1512 return fd;
1513 if (most_interesting_errno == ENOENT)
1514 most_interesting_errno = errno;
1515 }
1516 errno = most_interesting_errno;
1517 return -1;
1518 }
1519
1520 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1521 {
1522 void *map;
1523 int fd;
1524
1525 fd = open_sha1_file(sha1);
1526 map = NULL;
1527 if (fd >= 0) {
1528 struct stat st;
1529
1530 if (!fstat(fd, &st)) {
1531 *size = xsize_t(st.st_size);
1532 if (!*size) {
1533 /* mmap() is forbidden on empty files */
1534 error("object file %s is empty", sha1_file_name(sha1));
1535 return NULL;
1536 }
1537 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1538 }
1539 close(fd);
1540 }
1541 return map;
1542 }
1543
1544 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1545 unsigned long len, enum object_type *type, unsigned long *sizep)
1546 {
1547 unsigned shift;
1548 unsigned long size, c;
1549 unsigned long used = 0;
1550
1551 c = buf[used++];
1552 *type = (c >> 4) & 7;
1553 size = c & 15;
1554 shift = 4;
1555 while (c & 0x80) {
1556 if (len <= used || bitsizeof(long) <= shift) {
1557 error("bad object header");
1558 size = used = 0;
1559 break;
1560 }
1561 c = buf[used++];
1562 size += (c & 0x7f) << shift;
1563 shift += 7;
1564 }
1565 *sizep = size;
1566 return used;
1567 }
1568
1569 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1570 {
1571 /* Get the data stream */
1572 memset(stream, 0, sizeof(*stream));
1573 stream->next_in = map;
1574 stream->avail_in = mapsize;
1575 stream->next_out = buffer;
1576 stream->avail_out = bufsiz;
1577
1578 git_inflate_init(stream);
1579 return git_inflate(stream, 0);
1580 }
1581
1582 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1583 {
1584 int bytes = strlen(buffer) + 1;
1585 unsigned char *buf = xmallocz(size);
1586 unsigned long n;
1587 int status = Z_OK;
1588
1589 n = stream->total_out - bytes;
1590 if (n > size)
1591 n = size;
1592 memcpy(buf, (char *) buffer + bytes, n);
1593 bytes = n;
1594 if (bytes <= size) {
1595 /*
1596 * The above condition must be (bytes <= size), not
1597 * (bytes < size). In other words, even though we
1598 * expect no more output and set avail_out to zero,
1599 * the input zlib stream may have bytes that express
1600 * "this concludes the stream", and we *do* want to
1601 * eat that input.
1602 *
1603 * Otherwise we would not be able to test that we
1604 * consumed all the input to reach the expected size;
1605 * we also want to check that zlib tells us that all
1606 * went well with status == Z_STREAM_END at the end.
1607 */
1608 stream->next_out = buf + bytes;
1609 stream->avail_out = size - bytes;
1610 while (status == Z_OK)
1611 status = git_inflate(stream, Z_FINISH);
1612 }
1613 if (status == Z_STREAM_END && !stream->avail_in) {
1614 git_inflate_end(stream);
1615 return buf;
1616 }
1617
1618 if (status < 0)
1619 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1620 else if (stream->avail_in)
1621 error("garbage at end of loose object '%s'",
1622 sha1_to_hex(sha1));
1623 free(buf);
1624 return NULL;
1625 }
1626
1627 /*
1628 * We used to just use "sscanf()", but that's actually way
1629 * too permissive for what we want to check. So do an anal
1630 * object header parse by hand.
1631 */
1632 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1633 {
1634 char type[10];
1635 int i;
1636 unsigned long size;
1637
1638 /*
1639 * The type can be at most ten bytes (including the
1640 * terminating '\0' that we add), and is followed by
1641 * a space.
1642 */
1643 i = 0;
1644 for (;;) {
1645 char c = *hdr++;
1646 if (c == ' ')
1647 break;
1648 type[i++] = c;
1649 if (i >= sizeof(type))
1650 return -1;
1651 }
1652 type[i] = 0;
1653
1654 /*
1655 * The length must follow immediately, and be in canonical
1656 * decimal format (ie "010" is not valid).
1657 */
1658 size = *hdr++ - '0';
1659 if (size > 9)
1660 return -1;
1661 if (size) {
1662 for (;;) {
1663 unsigned long c = *hdr - '0';
1664 if (c > 9)
1665 break;
1666 hdr++;
1667 size = size * 10 + c;
1668 }
1669 }
1670 *sizep = size;
1671
1672 /*
1673 * The length must be followed by a zero byte
1674 */
1675 return *hdr ? -1 : type_from_string(type);
1676 }
1677
1678 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1679 {
1680 int ret;
1681 git_zstream stream;
1682 char hdr[8192];
1683
1684 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1685 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1686 return NULL;
1687
1688 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1689 }
1690
1691 unsigned long get_size_from_delta(struct packed_git *p,
1692 struct pack_window **w_curs,
1693 off_t curpos)
1694 {
1695 const unsigned char *data;
1696 unsigned char delta_head[20], *in;
1697 git_zstream stream;
1698 int st;
1699
1700 memset(&stream, 0, sizeof(stream));
1701 stream.next_out = delta_head;
1702 stream.avail_out = sizeof(delta_head);
1703
1704 git_inflate_init(&stream);
1705 do {
1706 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1707 stream.next_in = in;
1708 st = git_inflate(&stream, Z_FINISH);
1709 curpos += stream.next_in - in;
1710 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1711 stream.total_out < sizeof(delta_head));
1712 git_inflate_end(&stream);
1713 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1714 error("delta data unpack-initial failed");
1715 return 0;
1716 }
1717
1718 /* Examine the initial part of the delta to figure out
1719 * the result size.
1720 */
1721 data = delta_head;
1722
1723 /* ignore base size */
1724 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1725
1726 /* Read the result size */
1727 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1728 }
1729
1730 static off_t get_delta_base(struct packed_git *p,
1731 struct pack_window **w_curs,
1732 off_t *curpos,
1733 enum object_type type,
1734 off_t delta_obj_offset)
1735 {
1736 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1737 off_t base_offset;
1738
1739 /* use_pack() assured us we have [base_info, base_info + 20)
1740 * as a range that we can look at without walking off the
1741 * end of the mapped window. Its actually the hash size
1742 * that is assured. An OFS_DELTA longer than the hash size
1743 * is stupid, as then a REF_DELTA would be smaller to store.
1744 */
1745 if (type == OBJ_OFS_DELTA) {
1746 unsigned used = 0;
1747 unsigned char c = base_info[used++];
1748 base_offset = c & 127;
1749 while (c & 128) {
1750 base_offset += 1;
1751 if (!base_offset || MSB(base_offset, 7))
1752 return 0; /* overflow */
1753 c = base_info[used++];
1754 base_offset = (base_offset << 7) + (c & 127);
1755 }
1756 base_offset = delta_obj_offset - base_offset;
1757 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1758 return 0; /* out of bound */
1759 *curpos += used;
1760 } else if (type == OBJ_REF_DELTA) {
1761 /* The base entry _must_ be in the same pack */
1762 base_offset = find_pack_entry_one(base_info, p);
1763 *curpos += 20;
1764 } else
1765 die("I am totally screwed");
1766 return base_offset;
1767 }
1768
1769 /*
1770 * Like get_delta_base above, but we return the sha1 instead of the pack
1771 * offset. This means it is cheaper for REF deltas (we do not have to do
1772 * the final object lookup), but more expensive for OFS deltas (we
1773 * have to load the revidx to convert the offset back into a sha1).
1774 */
1775 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1776 struct pack_window **w_curs,
1777 off_t curpos,
1778 enum object_type type,
1779 off_t delta_obj_offset)
1780 {
1781 if (type == OBJ_REF_DELTA) {
1782 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1783 return base;
1784 } else if (type == OBJ_OFS_DELTA) {
1785 struct revindex_entry *revidx;
1786 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1787 type, delta_obj_offset);
1788
1789 if (!base_offset)
1790 return NULL;
1791
1792 revidx = find_pack_revindex(p, base_offset);
1793 if (!revidx)
1794 return NULL;
1795
1796 return nth_packed_object_sha1(p, revidx->nr);
1797 } else
1798 return NULL;
1799 }
1800
1801 int unpack_object_header(struct packed_git *p,
1802 struct pack_window **w_curs,
1803 off_t *curpos,
1804 unsigned long *sizep)
1805 {
1806 unsigned char *base;
1807 unsigned long left;
1808 unsigned long used;
1809 enum object_type type;
1810
1811 /* use_pack() assures us we have [base, base + 20) available
1812 * as a range that we can look at. (Its actually the hash
1813 * size that is assured.) With our object header encoding
1814 * the maximum deflated object size is 2^137, which is just
1815 * insane, so we know won't exceed what we have been given.
1816 */
1817 base = use_pack(p, w_curs, *curpos, &left);
1818 used = unpack_object_header_buffer(base, left, &type, sizep);
1819 if (!used) {
1820 type = OBJ_BAD;
1821 } else
1822 *curpos += used;
1823
1824 return type;
1825 }
1826
1827 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1828 {
1829 int type;
1830 struct revindex_entry *revidx;
1831 const unsigned char *sha1;
1832 revidx = find_pack_revindex(p, obj_offset);
1833 if (!revidx)
1834 return OBJ_BAD;
1835 sha1 = nth_packed_object_sha1(p, revidx->nr);
1836 mark_bad_packed_object(p, sha1);
1837 type = sha1_object_info(sha1, NULL);
1838 if (type <= OBJ_NONE)
1839 return OBJ_BAD;
1840 return type;
1841 }
1842
1843 #define POI_STACK_PREALLOC 64
1844
1845 static enum object_type packed_to_object_type(struct packed_git *p,
1846 off_t obj_offset,
1847 enum object_type type,
1848 struct pack_window **w_curs,
1849 off_t curpos)
1850 {
1851 off_t small_poi_stack[POI_STACK_PREALLOC];
1852 off_t *poi_stack = small_poi_stack;
1853 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1854
1855 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1856 off_t base_offset;
1857 unsigned long size;
1858 /* Push the object we're going to leave behind */
1859 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1860 poi_stack_alloc = alloc_nr(poi_stack_nr);
1861 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1862 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1863 } else {
1864 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1865 }
1866 poi_stack[poi_stack_nr++] = obj_offset;
1867 /* If parsing the base offset fails, just unwind */
1868 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1869 if (!base_offset)
1870 goto unwind;
1871 curpos = obj_offset = base_offset;
1872 type = unpack_object_header(p, w_curs, &curpos, &size);
1873 if (type <= OBJ_NONE) {
1874 /* If getting the base itself fails, we first
1875 * retry the base, otherwise unwind */
1876 type = retry_bad_packed_offset(p, base_offset);
1877 if (type > OBJ_NONE)
1878 goto out;
1879 goto unwind;
1880 }
1881 }
1882
1883 switch (type) {
1884 case OBJ_BAD:
1885 case OBJ_COMMIT:
1886 case OBJ_TREE:
1887 case OBJ_BLOB:
1888 case OBJ_TAG:
1889 break;
1890 default:
1891 error("unknown object type %i at offset %"PRIuMAX" in %s",
1892 type, (uintmax_t)obj_offset, p->pack_name);
1893 type = OBJ_BAD;
1894 }
1895
1896 out:
1897 if (poi_stack != small_poi_stack)
1898 free(poi_stack);
1899 return type;
1900
1901 unwind:
1902 while (poi_stack_nr) {
1903 obj_offset = poi_stack[--poi_stack_nr];
1904 type = retry_bad_packed_offset(p, obj_offset);
1905 if (type > OBJ_NONE)
1906 goto out;
1907 }
1908 type = OBJ_BAD;
1909 goto out;
1910 }
1911
1912 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1913 struct object_info *oi)
1914 {
1915 struct pack_window *w_curs = NULL;
1916 unsigned long size;
1917 off_t curpos = obj_offset;
1918 enum object_type type;
1919
1920 /*
1921 * We always get the representation type, but only convert it to
1922 * a "real" type later if the caller is interested.
1923 */
1924 type = unpack_object_header(p, &w_curs, &curpos, &size);
1925
1926 if (oi->sizep) {
1927 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1928 off_t tmp_pos = curpos;
1929 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1930 type, obj_offset);
1931 if (!base_offset) {
1932 type = OBJ_BAD;
1933 goto out;
1934 }
1935 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1936 if (*oi->sizep == 0) {
1937 type = OBJ_BAD;
1938 goto out;
1939 }
1940 } else {
1941 *oi->sizep = size;
1942 }
1943 }
1944
1945 if (oi->disk_sizep) {
1946 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1947 *oi->disk_sizep = revidx[1].offset - obj_offset;
1948 }
1949
1950 if (oi->typep) {
1951 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1952 if (*oi->typep < 0) {
1953 type = OBJ_BAD;
1954 goto out;
1955 }
1956 }
1957
1958 if (oi->delta_base_sha1) {
1959 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1960 const unsigned char *base;
1961
1962 base = get_delta_base_sha1(p, &w_curs, curpos,
1963 type, obj_offset);
1964 if (!base) {
1965 type = OBJ_BAD;
1966 goto out;
1967 }
1968
1969 hashcpy(oi->delta_base_sha1, base);
1970 } else
1971 hashclr(oi->delta_base_sha1);
1972 }
1973
1974 out:
1975 unuse_pack(&w_curs);
1976 return type;
1977 }
1978
1979 static void *unpack_compressed_entry(struct packed_git *p,
1980 struct pack_window **w_curs,
1981 off_t curpos,
1982 unsigned long size)
1983 {
1984 int st;
1985 git_zstream stream;
1986 unsigned char *buffer, *in;
1987
1988 buffer = xmallocz_gently(size);
1989 if (!buffer)
1990 return NULL;
1991 memset(&stream, 0, sizeof(stream));
1992 stream.next_out = buffer;
1993 stream.avail_out = size + 1;
1994
1995 git_inflate_init(&stream);
1996 do {
1997 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1998 stream.next_in = in;
1999 st = git_inflate(&stream, Z_FINISH);
2000 if (!stream.avail_out)
2001 break; /* the payload is larger than it should be */
2002 curpos += stream.next_in - in;
2003 } while (st == Z_OK || st == Z_BUF_ERROR);
2004 git_inflate_end(&stream);
2005 if ((st != Z_STREAM_END) || stream.total_out != size) {
2006 free(buffer);
2007 return NULL;
2008 }
2009
2010 return buffer;
2011 }
2012
2013 #define MAX_DELTA_CACHE (256)
2014
2015 static size_t delta_base_cached;
2016
2017 static struct delta_base_cache_lru_list {
2018 struct delta_base_cache_lru_list *prev;
2019 struct delta_base_cache_lru_list *next;
2020 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2021
2022 static struct delta_base_cache_entry {
2023 struct delta_base_cache_lru_list lru;
2024 void *data;
2025 struct packed_git *p;
2026 off_t base_offset;
2027 unsigned long size;
2028 enum object_type type;
2029 } delta_base_cache[MAX_DELTA_CACHE];
2030
2031 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2032 {
2033 unsigned long hash;
2034
2035 hash = (unsigned long)p + (unsigned long)base_offset;
2036 hash += (hash >> 8) + (hash >> 16);
2037 return hash % MAX_DELTA_CACHE;
2038 }
2039
2040 static struct delta_base_cache_entry *
2041 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2042 {
2043 unsigned long hash = pack_entry_hash(p, base_offset);
2044 return delta_base_cache + hash;
2045 }
2046
2047 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2048 struct packed_git *p, off_t base_offset)
2049 {
2050 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2051 }
2052
2053 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2054 {
2055 struct delta_base_cache_entry *ent;
2056 ent = get_delta_base_cache_entry(p, base_offset);
2057 return eq_delta_base_cache_entry(ent, p, base_offset);
2058 }
2059
2060 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2061 {
2062 ent->data = NULL;
2063 ent->lru.next->prev = ent->lru.prev;
2064 ent->lru.prev->next = ent->lru.next;
2065 delta_base_cached -= ent->size;
2066 }
2067
2068 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2069 unsigned long *base_size, enum object_type *type, int keep_cache)
2070 {
2071 struct delta_base_cache_entry *ent;
2072 void *ret;
2073
2074 ent = get_delta_base_cache_entry(p, base_offset);
2075
2076 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2077 return unpack_entry(p, base_offset, type, base_size);
2078
2079 ret = ent->data;
2080
2081 if (!keep_cache)
2082 clear_delta_base_cache_entry(ent);
2083 else
2084 ret = xmemdupz(ent->data, ent->size);
2085 *type = ent->type;
2086 *base_size = ent->size;
2087 return ret;
2088 }
2089
2090 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2091 {
2092 if (ent->data) {
2093 free(ent->data);
2094 ent->data = NULL;
2095 ent->lru.next->prev = ent->lru.prev;
2096 ent->lru.prev->next = ent->lru.next;
2097 delta_base_cached -= ent->size;
2098 }
2099 }
2100
2101 void clear_delta_base_cache(void)
2102 {
2103 unsigned long p;
2104 for (p = 0; p < MAX_DELTA_CACHE; p++)
2105 release_delta_base_cache(&delta_base_cache[p]);
2106 }
2107
2108 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2109 void *base, unsigned long base_size, enum object_type type)
2110 {
2111 unsigned long hash = pack_entry_hash(p, base_offset);
2112 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2113 struct delta_base_cache_lru_list *lru;
2114
2115 release_delta_base_cache(ent);
2116 delta_base_cached += base_size;
2117
2118 for (lru = delta_base_cache_lru.next;
2119 delta_base_cached > delta_base_cache_limit
2120 && lru != &delta_base_cache_lru;
2121 lru = lru->next) {
2122 struct delta_base_cache_entry *f = (void *)lru;
2123 if (f->type == OBJ_BLOB)
2124 release_delta_base_cache(f);
2125 }
2126 for (lru = delta_base_cache_lru.next;
2127 delta_base_cached > delta_base_cache_limit
2128 && lru != &delta_base_cache_lru;
2129 lru = lru->next) {
2130 struct delta_base_cache_entry *f = (void *)lru;
2131 release_delta_base_cache(f);
2132 }
2133
2134 ent->p = p;
2135 ent->base_offset = base_offset;
2136 ent->type = type;
2137 ent->data = base;
2138 ent->size = base_size;
2139 ent->lru.next = &delta_base_cache_lru;
2140 ent->lru.prev = delta_base_cache_lru.prev;
2141 delta_base_cache_lru.prev->next = &ent->lru;
2142 delta_base_cache_lru.prev = &ent->lru;
2143 }
2144
2145 static void *read_object(const unsigned char *sha1, enum object_type *type,
2146 unsigned long *size);
2147
2148 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2149 {
2150 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2151 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2152 p->pack_name, (uintmax_t)obj_offset);
2153 }
2154
2155 int do_check_packed_object_crc;
2156
2157 #define UNPACK_ENTRY_STACK_PREALLOC 64
2158 struct unpack_entry_stack_ent {
2159 off_t obj_offset;
2160 off_t curpos;
2161 unsigned long size;
2162 };
2163
2164 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2165 enum object_type *final_type, unsigned long *final_size)
2166 {
2167 struct pack_window *w_curs = NULL;
2168 off_t curpos = obj_offset;
2169 void *data = NULL;
2170 unsigned long size;
2171 enum object_type type;
2172 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2173 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2174 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2175 int base_from_cache = 0;
2176
2177 write_pack_access_log(p, obj_offset);
2178
2179 /* PHASE 1: drill down to the innermost base object */
2180 for (;;) {
2181 off_t base_offset;
2182 int i;
2183 struct delta_base_cache_entry *ent;
2184
2185 ent = get_delta_base_cache_entry(p, curpos);
2186 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2187 type = ent->type;
2188 data = ent->data;
2189 size = ent->size;
2190 clear_delta_base_cache_entry(ent);
2191 base_from_cache = 1;
2192 break;
2193 }
2194
2195 if (do_check_packed_object_crc && p->index_version > 1) {
2196 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2197 unsigned long len = revidx[1].offset - obj_offset;
2198 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2199 const unsigned char *sha1 =
2200 nth_packed_object_sha1(p, revidx->nr);
2201 error("bad packed object CRC for %s",
2202 sha1_to_hex(sha1));
2203 mark_bad_packed_object(p, sha1);
2204 unuse_pack(&w_curs);
2205 return NULL;
2206 }
2207 }
2208
2209 type = unpack_object_header(p, &w_curs, &curpos, &size);
2210 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2211 break;
2212
2213 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2214 if (!base_offset) {
2215 error("failed to validate delta base reference "
2216 "at offset %"PRIuMAX" from %s",
2217 (uintmax_t)curpos, p->pack_name);
2218 /* bail to phase 2, in hopes of recovery */
2219 data = NULL;
2220 break;
2221 }
2222
2223 /* push object, proceed to base */
2224 if (delta_stack_nr >= delta_stack_alloc
2225 && delta_stack == small_delta_stack) {
2226 delta_stack_alloc = alloc_nr(delta_stack_nr);
2227 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2228 memcpy(delta_stack, small_delta_stack,
2229 sizeof(*delta_stack)*delta_stack_nr);
2230 } else {
2231 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2232 }
2233 i = delta_stack_nr++;
2234 delta_stack[i].obj_offset = obj_offset;
2235 delta_stack[i].curpos = curpos;
2236 delta_stack[i].size = size;
2237
2238 curpos = obj_offset = base_offset;
2239 }
2240
2241 /* PHASE 2: handle the base */
2242 switch (type) {
2243 case OBJ_OFS_DELTA:
2244 case OBJ_REF_DELTA:
2245 if (data)
2246 die("BUG in unpack_entry: left loop at a valid delta");
2247 break;
2248 case OBJ_COMMIT:
2249 case OBJ_TREE:
2250 case OBJ_BLOB:
2251 case OBJ_TAG:
2252 if (!base_from_cache)
2253 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2254 break;
2255 default:
2256 data = NULL;
2257 error("unknown object type %i at offset %"PRIuMAX" in %s",
2258 type, (uintmax_t)obj_offset, p->pack_name);
2259 }
2260
2261 /* PHASE 3: apply deltas in order */
2262
2263 /* invariants:
2264 * 'data' holds the base data, or NULL if there was corruption
2265 */
2266 while (delta_stack_nr) {
2267 void *delta_data;
2268 void *base = data;
2269 unsigned long delta_size, base_size = size;
2270 int i;
2271
2272 data = NULL;
2273
2274 if (base)
2275 add_delta_base_cache(p, obj_offset, base, base_size, type);
2276
2277 if (!base) {
2278 /*
2279 * We're probably in deep shit, but let's try to fetch
2280 * the required base anyway from another pack or loose.
2281 * This is costly but should happen only in the presence
2282 * of a corrupted pack, and is better than failing outright.
2283 */
2284 struct revindex_entry *revidx;
2285 const unsigned char *base_sha1;
2286 revidx = find_pack_revindex(p, obj_offset);
2287 if (revidx) {
2288 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2289 error("failed to read delta base object %s"
2290 " at offset %"PRIuMAX" from %s",
2291 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2292 p->pack_name);
2293 mark_bad_packed_object(p, base_sha1);
2294 base = read_object(base_sha1, &type, &base_size);
2295 }
2296 }
2297
2298 i = --delta_stack_nr;
2299 obj_offset = delta_stack[i].obj_offset;
2300 curpos = delta_stack[i].curpos;
2301 delta_size = delta_stack[i].size;
2302
2303 if (!base)
2304 continue;
2305
2306 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2307
2308 if (!delta_data) {
2309 error("failed to unpack compressed delta "
2310 "at offset %"PRIuMAX" from %s",
2311 (uintmax_t)curpos, p->pack_name);
2312 data = NULL;
2313 continue;
2314 }
2315
2316 data = patch_delta(base, base_size,
2317 delta_data, delta_size,
2318 &size);
2319
2320 /*
2321 * We could not apply the delta; warn the user, but keep going.
2322 * Our failure will be noticed either in the next iteration of
2323 * the loop, or if this is the final delta, in the caller when
2324 * we return NULL. Those code paths will take care of making
2325 * a more explicit warning and retrying with another copy of
2326 * the object.
2327 */
2328 if (!data)
2329 error("failed to apply delta");
2330
2331 free(delta_data);
2332 }
2333
2334 *final_type = type;
2335 *final_size = size;
2336
2337 unuse_pack(&w_curs);
2338
2339 if (delta_stack != small_delta_stack)
2340 free(delta_stack);
2341
2342 return data;
2343 }
2344
2345 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2346 uint32_t n)
2347 {
2348 const unsigned char *index = p->index_data;
2349 if (!index) {
2350 if (open_pack_index(p))
2351 return NULL;
2352 index = p->index_data;
2353 }
2354 if (n >= p->num_objects)
2355 return NULL;
2356 index += 4 * 256;
2357 if (p->index_version == 1) {
2358 return index + 24 * n + 4;
2359 } else {
2360 index += 8;
2361 return index + 20 * n;
2362 }
2363 }
2364
2365 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2366 {
2367 const unsigned char *index = p->index_data;
2368 index += 4 * 256;
2369 if (p->index_version == 1) {
2370 return ntohl(*((uint32_t *)(index + 24 * n)));
2371 } else {
2372 uint32_t off;
2373 index += 8 + p->num_objects * (20 + 4);
2374 off = ntohl(*((uint32_t *)(index + 4 * n)));
2375 if (!(off & 0x80000000))
2376 return off;
2377 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2378 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2379 ntohl(*((uint32_t *)(index + 4)));
2380 }
2381 }
2382
2383 off_t find_pack_entry_one(const unsigned char *sha1,
2384 struct packed_git *p)
2385 {
2386 const uint32_t *level1_ofs = p->index_data;
2387 const unsigned char *index = p->index_data;
2388 unsigned hi, lo, stride;
2389 static int use_lookup = -1;
2390 static int debug_lookup = -1;
2391
2392 if (debug_lookup < 0)
2393 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2394
2395 if (!index) {
2396 if (open_pack_index(p))
2397 return 0;
2398 level1_ofs = p->index_data;
2399 index = p->index_data;
2400 }
2401 if (p->index_version > 1) {
2402 level1_ofs += 2;
2403 index += 8;
2404 }
2405 index += 4 * 256;
2406 hi = ntohl(level1_ofs[*sha1]);
2407 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2408 if (p->index_version > 1) {
2409 stride = 20;
2410 } else {
2411 stride = 24;
2412 index += 4;
2413 }
2414
2415 if (debug_lookup)
2416 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2417 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2418
2419 if (use_lookup < 0)
2420 use_lookup = !!getenv("GIT_USE_LOOKUP");
2421 if (use_lookup) {
2422 int pos = sha1_entry_pos(index, stride, 0,
2423 lo, hi, p->num_objects, sha1);
2424 if (pos < 0)
2425 return 0;
2426 return nth_packed_object_offset(p, pos);
2427 }
2428
2429 do {
2430 unsigned mi = (lo + hi) / 2;
2431 int cmp = hashcmp(index + mi * stride, sha1);
2432
2433 if (debug_lookup)
2434 printf("lo %u hi %u rg %u mi %u\n",
2435 lo, hi, hi - lo, mi);
2436 if (!cmp)
2437 return nth_packed_object_offset(p, mi);
2438 if (cmp > 0)
2439 hi = mi;
2440 else
2441 lo = mi+1;
2442 } while (lo < hi);
2443 return 0;
2444 }
2445
2446 int is_pack_valid(struct packed_git *p)
2447 {
2448 /* An already open pack is known to be valid. */
2449 if (p->pack_fd != -1)
2450 return 1;
2451
2452 /* If the pack has one window completely covering the
2453 * file size, the pack is known to be valid even if
2454 * the descriptor is not currently open.
2455 */
2456 if (p->windows) {
2457 struct pack_window *w = p->windows;
2458
2459 if (!w->offset && w->len == p->pack_size)
2460 return 1;
2461 }
2462
2463 /* Force the pack to open to prove its valid. */
2464 return !open_packed_git(p);
2465 }
2466
2467 static int fill_pack_entry(const unsigned char *sha1,
2468 struct pack_entry *e,
2469 struct packed_git *p)
2470 {
2471 off_t offset;
2472
2473 if (p->num_bad_objects) {
2474 unsigned i;
2475 for (i = 0; i < p->num_bad_objects; i++)
2476 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2477 return 0;
2478 }
2479
2480 offset = find_pack_entry_one(sha1, p);
2481 if (!offset)
2482 return 0;
2483
2484 /*
2485 * We are about to tell the caller where they can locate the
2486 * requested object. We better make sure the packfile is
2487 * still here and can be accessed before supplying that
2488 * answer, as it may have been deleted since the index was
2489 * loaded!
2490 */
2491 if (!is_pack_valid(p))
2492 return 0;
2493 e->offset = offset;
2494 e->p = p;
2495 hashcpy(e->sha1, sha1);
2496 return 1;
2497 }
2498
2499 /*
2500 * Iff a pack file contains the object named by sha1, return true and
2501 * store its location to e.
2502 */
2503 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2504 {
2505 struct packed_git *p;
2506
2507 prepare_packed_git();
2508 if (!packed_git)
2509 return 0;
2510
2511 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2512 return 1;
2513
2514 for (p = packed_git; p; p = p->next) {
2515 if (p == last_found_pack)
2516 continue; /* we already checked this one */
2517
2518 if (fill_pack_entry(sha1, e, p)) {
2519 last_found_pack = p;
2520 return 1;
2521 }
2522 }
2523 return 0;
2524 }
2525
2526 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2527 struct packed_git *packs)
2528 {
2529 struct packed_git *p;
2530
2531 for (p = packs; p; p = p->next) {
2532 if (find_pack_entry_one(sha1, p))
2533 return p;
2534 }
2535 return NULL;
2536
2537 }
2538
2539 static int sha1_loose_object_info(const unsigned char *sha1,
2540 struct object_info *oi)
2541 {
2542 int status;
2543 unsigned long mapsize, size;
2544 void *map;
2545 git_zstream stream;
2546 char hdr[32];
2547
2548 if (oi->delta_base_sha1)
2549 hashclr(oi->delta_base_sha1);
2550
2551 /*
2552 * If we don't care about type or size, then we don't
2553 * need to look inside the object at all. Note that we
2554 * do not optimize out the stat call, even if the
2555 * caller doesn't care about the disk-size, since our
2556 * return value implicitly indicates whether the
2557 * object even exists.
2558 */
2559 if (!oi->typep && !oi->sizep) {
2560 struct stat st;
2561 if (stat_sha1_file(sha1, &st) < 0)
2562 return -1;
2563 if (oi->disk_sizep)
2564 *oi->disk_sizep = st.st_size;
2565 return 0;
2566 }
2567
2568 map = map_sha1_file(sha1, &mapsize);
2569 if (!map)
2570 return -1;
2571 if (oi->disk_sizep)
2572 *oi->disk_sizep = mapsize;
2573 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2574 status = error("unable to unpack %s header",
2575 sha1_to_hex(sha1));
2576 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2577 status = error("unable to parse %s header", sha1_to_hex(sha1));
2578 else if (oi->sizep)
2579 *oi->sizep = size;
2580 git_inflate_end(&stream);
2581 munmap(map, mapsize);
2582 if (oi->typep)
2583 *oi->typep = status;
2584 return 0;
2585 }
2586
2587 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2588 {
2589 struct cached_object *co;
2590 struct pack_entry e;
2591 int rtype;
2592 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2593
2594 co = find_cached_object(real);
2595 if (co) {
2596 if (oi->typep)
2597 *(oi->typep) = co->type;
2598 if (oi->sizep)
2599 *(oi->sizep) = co->size;
2600 if (oi->disk_sizep)
2601 *(oi->disk_sizep) = 0;
2602 if (oi->delta_base_sha1)
2603 hashclr(oi->delta_base_sha1);
2604 oi->whence = OI_CACHED;
2605 return 0;
2606 }
2607
2608 if (!find_pack_entry(real, &e)) {
2609 /* Most likely it's a loose object. */
2610 if (!sha1_loose_object_info(real, oi)) {
2611 oi->whence = OI_LOOSE;
2612 return 0;
2613 }
2614
2615 /* Not a loose object; someone else may have just packed it. */
2616 reprepare_packed_git();
2617 if (!find_pack_entry(real, &e))
2618 return -1;
2619 }
2620
2621 rtype = packed_object_info(e.p, e.offset, oi);
2622 if (rtype < 0) {
2623 mark_bad_packed_object(e.p, real);
2624 return sha1_object_info_extended(real, oi, 0);
2625 } else if (in_delta_base_cache(e.p, e.offset)) {
2626 oi->whence = OI_DBCACHED;
2627 } else {
2628 oi->whence = OI_PACKED;
2629 oi->u.packed.offset = e.offset;
2630 oi->u.packed.pack = e.p;
2631 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2632 rtype == OBJ_OFS_DELTA);
2633 }
2634
2635 return 0;
2636 }
2637
2638 /* returns enum object_type or negative */
2639 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2640 {
2641 enum object_type type;
2642 struct object_info oi = {NULL};
2643
2644 oi.typep = &type;
2645 oi.sizep = sizep;
2646 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2647 return -1;
2648 return type;
2649 }
2650
2651 static void *read_packed_sha1(const unsigned char *sha1,
2652 enum object_type *type, unsigned long *size)
2653 {
2654 struct pack_entry e;
2655 void *data;
2656
2657 if (!find_pack_entry(sha1, &e))
2658 return NULL;
2659 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2660 if (!data) {
2661 /*
2662 * We're probably in deep shit, but let's try to fetch
2663 * the required object anyway from another pack or loose.
2664 * This should happen only in the presence of a corrupted
2665 * pack, and is better than failing outright.
2666 */
2667 error("failed to read object %s at offset %"PRIuMAX" from %s",
2668 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2669 mark_bad_packed_object(e.p, sha1);
2670 data = read_object(sha1, type, size);
2671 }
2672 return data;
2673 }
2674
2675 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2676 unsigned char *sha1)
2677 {
2678 struct cached_object *co;
2679
2680 hash_sha1_file(buf, len, typename(type), sha1);
2681 if (has_sha1_file(sha1) || find_cached_object(sha1))
2682 return 0;
2683 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2684 co = &cached_objects[cached_object_nr++];
2685 co->size = len;
2686 co->type = type;
2687 co->buf = xmalloc(len);
2688 memcpy(co->buf, buf, len);
2689 hashcpy(co->sha1, sha1);
2690 return 0;
2691 }
2692
2693 static void *read_object(const unsigned char *sha1, enum object_type *type,
2694 unsigned long *size)
2695 {
2696 unsigned long mapsize;
2697 void *map, *buf;
2698 struct cached_object *co;
2699
2700 co = find_cached_object(sha1);
2701 if (co) {
2702 *type = co->type;
2703 *size = co->size;
2704 return xmemdupz(co->buf, co->size);
2705 }
2706
2707 buf = read_packed_sha1(sha1, type, size);
2708 if (buf)
2709 return buf;
2710 map = map_sha1_file(sha1, &mapsize);
2711 if (map) {
2712 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2713 munmap(map, mapsize);
2714 return buf;
2715 }
2716 reprepare_packed_git();
2717 return read_packed_sha1(sha1, type, size);
2718 }
2719
2720 /*
2721 * This function dies on corrupt objects; the callers who want to
2722 * deal with them should arrange to call read_object() and give error
2723 * messages themselves.
2724 */
2725 void *read_sha1_file_extended(const unsigned char *sha1,
2726 enum object_type *type,
2727 unsigned long *size,
2728 unsigned flag)
2729 {
2730 void *data;
2731 const struct packed_git *p;
2732 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2733
2734 errno = 0;
2735 data = read_object(repl, type, size);
2736 if (data)
2737 return data;
2738
2739 if (errno && errno != ENOENT)
2740 die_errno("failed to read object %s", sha1_to_hex(sha1));
2741
2742 /* die if we replaced an object with one that does not exist */
2743 if (repl != sha1)
2744 die("replacement %s not found for %s",
2745 sha1_to_hex(repl), sha1_to_hex(sha1));
2746
2747 if (has_loose_object(repl)) {
2748 const char *path = sha1_file_name(sha1);
2749
2750 die("loose object %s (stored in %s) is corrupt",
2751 sha1_to_hex(repl), path);
2752 }
2753
2754 if ((p = has_packed_and_bad(repl)) != NULL)
2755 die("packed object %s (stored in %s) is corrupt",
2756 sha1_to_hex(repl), p->pack_name);
2757
2758 return NULL;
2759 }
2760
2761 void *read_object_with_reference(const unsigned char *sha1,
2762 const char *required_type_name,
2763 unsigned long *size,
2764 unsigned char *actual_sha1_return)
2765 {
2766 enum object_type type, required_type;
2767 void *buffer;
2768 unsigned long isize;
2769 unsigned char actual_sha1[20];
2770
2771 required_type = type_from_string(required_type_name);
2772 hashcpy(actual_sha1, sha1);
2773 while (1) {
2774 int ref_length = -1;
2775 const char *ref_type = NULL;
2776
2777 buffer = read_sha1_file(actual_sha1, &type, &isize);
2778 if (!buffer)
2779 return NULL;
2780 if (type == required_type) {
2781 *size = isize;
2782 if (actual_sha1_return)
2783 hashcpy(actual_sha1_return, actual_sha1);
2784 return buffer;
2785 }
2786 /* Handle references */
2787 else if (type == OBJ_COMMIT)
2788 ref_type = "tree ";
2789 else if (type == OBJ_TAG)
2790 ref_type = "object ";
2791 else {
2792 free(buffer);
2793 return NULL;
2794 }
2795 ref_length = strlen(ref_type);
2796
2797 if (ref_length + 40 > isize ||
2798 memcmp(buffer, ref_type, ref_length) ||
2799 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2800 free(buffer);
2801 return NULL;
2802 }
2803 free(buffer);
2804 /* Now we have the ID of the referred-to object in
2805 * actual_sha1. Check again. */
2806 }
2807 }
2808
2809 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2810 const char *type, unsigned char *sha1,
2811 char *hdr, int *hdrlen)
2812 {
2813 git_SHA_CTX c;
2814
2815 /* Generate the header */
2816 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2817
2818 /* Sha1.. */
2819 git_SHA1_Init(&c);
2820 git_SHA1_Update(&c, hdr, *hdrlen);
2821 git_SHA1_Update(&c, buf, len);
2822 git_SHA1_Final(sha1, &c);
2823 }
2824
2825 /*
2826 * Move the just written object into its final resting place.
2827 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2828 * "moving" is only a part of what it does, when no patch between
2829 * master to pu changes the call sites of this function.
2830 */
2831 int move_temp_to_file(const char *tmpfile, const char *filename)
2832 {
2833 int ret = 0;
2834
2835 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2836 goto try_rename;
2837 else if (link(tmpfile, filename))
2838 ret = errno;
2839
2840 /*
2841 * Coda hack - coda doesn't like cross-directory links,
2842 * so we fall back to a rename, which will mean that it
2843 * won't be able to check collisions, but that's not a
2844 * big deal.
2845 *
2846 * The same holds for FAT formatted media.
2847 *
2848 * When this succeeds, we just return. We have nothing
2849 * left to unlink.
2850 */
2851 if (ret && ret != EEXIST) {
2852 try_rename:
2853 if (!rename(tmpfile, filename))
2854 goto out;
2855 ret = errno;
2856 }
2857 unlink_or_warn(tmpfile);
2858 if (ret) {
2859 if (ret != EEXIST) {
2860 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2861 }
2862 /* FIXME!!! Collision check here ? */
2863 }
2864
2865 out:
2866 if (adjust_shared_perm(filename))
2867 return error("unable to set permission to '%s'", filename);
2868 return 0;
2869 }
2870
2871 static int write_buffer(int fd, const void *buf, size_t len)
2872 {
2873 if (write_in_full(fd, buf, len) < 0)
2874 return error("file write error (%s)", strerror(errno));
2875 return 0;
2876 }
2877
2878 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2879 unsigned char *sha1)
2880 {
2881 char hdr[32];
2882 int hdrlen;
2883 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2884 return 0;
2885 }
2886
2887 /* Finalize a file on disk, and close it. */
2888 static void close_sha1_file(int fd)
2889 {
2890 if (fsync_object_files)
2891 fsync_or_die(fd, "sha1 file");
2892 if (close(fd) != 0)
2893 die_errno("error when closing sha1 file");
2894 }
2895
2896 /* Size of directory component, including the ending '/' */
2897 static inline int directory_size(const char *filename)
2898 {
2899 const char *s = strrchr(filename, '/');
2900 if (!s)
2901 return 0;
2902 return s - filename + 1;
2903 }
2904
2905 /*
2906 * This creates a temporary file in the same directory as the final
2907 * 'filename'
2908 *
2909 * We want to avoid cross-directory filename renames, because those
2910 * can have problems on various filesystems (FAT, NFS, Coda).
2911 */
2912 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2913 {
2914 int fd, dirlen = directory_size(filename);
2915
2916 if (dirlen + 20 > bufsiz) {
2917 errno = ENAMETOOLONG;
2918 return -1;
2919 }
2920 memcpy(buffer, filename, dirlen);
2921 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2922 fd = git_mkstemp_mode(buffer, 0444);
2923 if (fd < 0 && dirlen && errno == ENOENT) {
2924 /* Make sure the directory exists */
2925 memcpy(buffer, filename, dirlen);
2926 buffer[dirlen-1] = 0;
2927 if (mkdir(buffer, 0777) && errno != EEXIST)
2928 return -1;
2929 if (adjust_shared_perm(buffer))
2930 return -1;
2931
2932 /* Try again */
2933 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2934 fd = git_mkstemp_mode(buffer, 0444);
2935 }
2936 return fd;
2937 }
2938
2939 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2940 const void *buf, unsigned long len, time_t mtime)
2941 {
2942 int fd, ret;
2943 unsigned char compressed[4096];
2944 git_zstream stream;
2945 git_SHA_CTX c;
2946 unsigned char parano_sha1[20];
2947 static char tmp_file[PATH_MAX];
2948 const char *filename = sha1_file_name(sha1);
2949
2950 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2951 if (fd < 0) {
2952 if (errno == EACCES)
2953 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2954 else
2955 return error("unable to create temporary file: %s", strerror(errno));
2956 }
2957
2958 /* Set it up */
2959 git_deflate_init(&stream, zlib_compression_level);
2960 stream.next_out = compressed;
2961 stream.avail_out = sizeof(compressed);
2962 git_SHA1_Init(&c);
2963
2964 /* First header.. */
2965 stream.next_in = (unsigned char *)hdr;
2966 stream.avail_in = hdrlen;
2967 while (git_deflate(&stream, 0) == Z_OK)
2968 ; /* nothing */
2969 git_SHA1_Update(&c, hdr, hdrlen);
2970
2971 /* Then the data itself.. */
2972 stream.next_in = (void *)buf;
2973 stream.avail_in = len;
2974 do {
2975 unsigned char *in0 = stream.next_in;
2976 ret = git_deflate(&stream, Z_FINISH);
2977 git_SHA1_Update(&c, in0, stream.next_in - in0);
2978 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2979 die("unable to write sha1 file");
2980 stream.next_out = compressed;
2981 stream.avail_out = sizeof(compressed);
2982 } while (ret == Z_OK);
2983
2984 if (ret != Z_STREAM_END)
2985 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2986 ret = git_deflate_end_gently(&stream);
2987 if (ret != Z_OK)
2988 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2989 git_SHA1_Final(parano_sha1, &c);
2990 if (hashcmp(sha1, parano_sha1) != 0)
2991 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2992
2993 close_sha1_file(fd);
2994
2995 if (mtime) {
2996 struct utimbuf utb;
2997 utb.actime = mtime;
2998 utb.modtime = mtime;
2999 if (utime(tmp_file, &utb) < 0)
3000 warning("failed utime() on %s: %s",
3001 tmp_file, strerror(errno));
3002 }
3003
3004 return move_temp_to_file(tmp_file, filename);
3005 }
3006
3007 static int freshen_loose_object(const unsigned char *sha1)
3008 {
3009 return check_and_freshen(sha1, 1);
3010 }
3011
3012 static int freshen_packed_object(const unsigned char *sha1)
3013 {
3014 struct pack_entry e;
3015 if (!find_pack_entry(sha1, &e))
3016 return 0;
3017 if (e.p->freshened)
3018 return 1;
3019 if (!freshen_file(e.p->pack_name))
3020 return 0;
3021 e.p->freshened = 1;
3022 return 1;
3023 }
3024
3025 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3026 {
3027 char hdr[32];
3028 int hdrlen;
3029
3030 /* Normally if we have it in the pack then we do not bother writing
3031 * it out into .git/objects/??/?{38} file.
3032 */
3033 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3034 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3035 return 0;
3036 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3037 }
3038
3039 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3040 unsigned char *sha1, unsigned flags)
3041 {
3042 char *header;
3043 int hdrlen, status = 0;
3044
3045 /* type string, SP, %lu of the length plus NUL must fit this */
3046 header = xmalloc(strlen(type) + 32);
3047 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3048
3049 if (!(flags & HASH_WRITE_OBJECT))
3050 goto cleanup;
3051 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3052 goto cleanup;
3053 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3054
3055 cleanup:
3056 free(header);
3057 return status;
3058 }
3059
3060 int force_object_loose(const unsigned char *sha1, time_t mtime)
3061 {
3062 void *buf;
3063 unsigned long len;
3064 enum object_type type;
3065 char hdr[32];
3066 int hdrlen;
3067 int ret;
3068
3069 if (has_loose_object(sha1))
3070 return 0;
3071 buf = read_packed_sha1(sha1, &type, &len);
3072 if (!buf)
3073 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3074 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3075 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3076 free(buf);
3077
3078 return ret;
3079 }
3080
3081 int has_pack_index(const unsigned char *sha1)
3082 {
3083 struct stat st;
3084 if (stat(sha1_pack_index_name(sha1), &st))
3085 return 0;
3086 return 1;
3087 }
3088
3089 int has_sha1_pack(const unsigned char *sha1)
3090 {
3091 struct pack_entry e;
3092 return find_pack_entry(sha1, &e);
3093 }
3094
3095 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3096 {
3097 struct pack_entry e;
3098
3099 if (find_pack_entry(sha1, &e))
3100 return 1;
3101 if (has_loose_object(sha1))
3102 return 1;
3103 if (flags & HAS_SHA1_QUICK)
3104 return 0;
3105 reprepare_packed_git();
3106 return find_pack_entry(sha1, &e);
3107 }
3108
3109 static void check_tree(const void *buf, size_t size)
3110 {
3111 struct tree_desc desc;
3112 struct name_entry entry;
3113
3114 init_tree_desc(&desc, buf, size);
3115 while (tree_entry(&desc, &entry))
3116 /* do nothing
3117 * tree_entry() will die() on malformed entries */
3118 ;
3119 }
3120
3121 static void check_commit(const void *buf, size_t size)
3122 {
3123 struct commit c;
3124 memset(&c, 0, sizeof(c));
3125 if (parse_commit_buffer(&c, buf, size))
3126 die("corrupt commit");
3127 }
3128
3129 static void check_tag(const void *buf, size_t size)
3130 {
3131 struct tag t;
3132 memset(&t, 0, sizeof(t));
3133 if (parse_tag_buffer(&t, buf, size))
3134 die("corrupt tag");
3135 }
3136
3137 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3138 enum object_type type,
3139 const char *path, unsigned flags)
3140 {
3141 int ret, re_allocated = 0;
3142 int write_object = flags & HASH_WRITE_OBJECT;
3143
3144 if (!type)
3145 type = OBJ_BLOB;
3146
3147 /*
3148 * Convert blobs to git internal format
3149 */
3150 if ((type == OBJ_BLOB) && path) {
3151 struct strbuf nbuf = STRBUF_INIT;
3152 if (convert_to_git(path, buf, size, &nbuf,
3153 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3154 buf = strbuf_detach(&nbuf, &size);
3155 re_allocated = 1;
3156 }
3157 }
3158 if (flags & HASH_FORMAT_CHECK) {
3159 if (type == OBJ_TREE)
3160 check_tree(buf, size);
3161 if (type == OBJ_COMMIT)
3162 check_commit(buf, size);
3163 if (type == OBJ_TAG)
3164 check_tag(buf, size);
3165 }
3166
3167 if (write_object)
3168 ret = write_sha1_file(buf, size, typename(type), sha1);
3169 else
3170 ret = hash_sha1_file(buf, size, typename(type), sha1);
3171 if (re_allocated)
3172 free(buf);
3173 return ret;
3174 }
3175
3176 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3177 const char *path, unsigned flags)
3178 {
3179 int ret;
3180 const int write_object = flags & HASH_WRITE_OBJECT;
3181 struct strbuf sbuf = STRBUF_INIT;
3182
3183 assert(path);
3184 assert(would_convert_to_git_filter_fd(path));
3185
3186 convert_to_git_filter_fd(path, fd, &sbuf,
3187 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3188
3189 if (write_object)
3190 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3191 sha1);
3192 else
3193 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3194 sha1);
3195 strbuf_release(&sbuf);
3196 return ret;
3197 }
3198
3199 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3200 const char *path, unsigned flags)
3201 {
3202 struct strbuf sbuf = STRBUF_INIT;
3203 int ret;
3204
3205 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3206 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3207 else
3208 ret = -1;
3209 strbuf_release(&sbuf);
3210 return ret;
3211 }
3212
3213 #define SMALL_FILE_SIZE (32*1024)
3214
3215 static int index_core(unsigned char *sha1, int fd, size_t size,
3216 enum object_type type, const char *path,
3217 unsigned flags)
3218 {
3219 int ret;
3220
3221 if (!size) {
3222 ret = index_mem(sha1, "", size, type, path, flags);
3223 } else if (size <= SMALL_FILE_SIZE) {
3224 char *buf = xmalloc(size);
3225 if (size == read_in_full(fd, buf, size))
3226 ret = index_mem(sha1, buf, size, type, path, flags);
3227 else
3228 ret = error("short read %s", strerror(errno));
3229 free(buf);
3230 } else {
3231 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3232 ret = index_mem(sha1, buf, size, type, path, flags);
3233 munmap(buf, size);
3234 }
3235 return ret;
3236 }
3237
3238 /*
3239 * This creates one packfile per large blob unless bulk-checkin
3240 * machinery is "plugged".
3241 *
3242 * This also bypasses the usual "convert-to-git" dance, and that is on
3243 * purpose. We could write a streaming version of the converting
3244 * functions and insert that before feeding the data to fast-import
3245 * (or equivalent in-core API described above). However, that is
3246 * somewhat complicated, as we do not know the size of the filter
3247 * result, which we need to know beforehand when writing a git object.
3248 * Since the primary motivation for trying to stream from the working
3249 * tree file and to avoid mmaping it in core is to deal with large
3250 * binary blobs, they generally do not want to get any conversion, and
3251 * callers should avoid this code path when filters are requested.
3252 */
3253 static int index_stream(unsigned char *sha1, int fd, size_t size,
3254 enum object_type type, const char *path,
3255 unsigned flags)
3256 {
3257 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3258 }
3259
3260 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3261 enum object_type type, const char *path, unsigned flags)
3262 {
3263 int ret;
3264
3265 /*
3266 * Call xsize_t() only when needed to avoid potentially unnecessary
3267 * die() for large files.
3268 */
3269 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3270 ret = index_stream_convert_blob(sha1, fd, path, flags);
3271 else if (!S_ISREG(st->st_mode))
3272 ret = index_pipe(sha1, fd, type, path, flags);
3273 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3274 (path && would_convert_to_git(path)))
3275 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3276 flags);
3277 else
3278 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3279 flags);
3280 close(fd);
3281 return ret;
3282 }
3283
3284 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3285 {
3286 int fd;
3287 struct strbuf sb = STRBUF_INIT;
3288
3289 switch (st->st_mode & S_IFMT) {
3290 case S_IFREG:
3291 fd = open(path, O_RDONLY);
3292 if (fd < 0)
3293 return error("open(\"%s\"): %s", path,
3294 strerror(errno));
3295 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3296 return error("%s: failed to insert into database",
3297 path);
3298 break;
3299 case S_IFLNK:
3300 if (strbuf_readlink(&sb, path, st->st_size)) {
3301 char *errstr = strerror(errno);
3302 return error("readlink(\"%s\"): %s", path,
3303 errstr);
3304 }
3305 if (!(flags & HASH_WRITE_OBJECT))
3306 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3307 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3308 return error("%s: failed to insert into database",
3309 path);
3310 strbuf_release(&sb);
3311 break;
3312 case S_IFDIR:
3313 return resolve_gitlink_ref(path, "HEAD", sha1);
3314 default:
3315 return error("%s: unsupported file type", path);
3316 }
3317 return 0;
3318 }
3319
3320 int read_pack_header(int fd, struct pack_header *header)
3321 {
3322 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3323 /* "eof before pack header was fully read" */
3324 return PH_ERROR_EOF;
3325
3326 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3327 /* "protocol error (pack signature mismatch detected)" */
3328 return PH_ERROR_PACK_SIGNATURE;
3329 if (!pack_version_ok(header->hdr_version))
3330 /* "protocol error (pack version unsupported)" */
3331 return PH_ERROR_PROTOCOL;
3332 return 0;
3333 }
3334
3335 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3336 {
3337 enum object_type type = sha1_object_info(sha1, NULL);
3338 if (type < 0)
3339 die("%s is not a valid object", sha1_to_hex(sha1));
3340 if (type != expect)
3341 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3342 typename(expect));
3343 }
3344
3345 static int for_each_file_in_obj_subdir(int subdir_nr,
3346 struct strbuf *path,
3347 each_loose_object_fn obj_cb,
3348 each_loose_cruft_fn cruft_cb,
3349 each_loose_subdir_fn subdir_cb,
3350 void *data)
3351 {
3352 size_t baselen = path->len;
3353 DIR *dir = opendir(path->buf);
3354 struct dirent *de;
3355 int r = 0;
3356
3357 if (!dir) {
3358 if (errno == ENOENT)
3359 return 0;
3360 return error("unable to open %s: %s", path->buf, strerror(errno));
3361 }
3362
3363 while ((de = readdir(dir))) {
3364 if (is_dot_or_dotdot(de->d_name))
3365 continue;
3366
3367 strbuf_setlen(path, baselen);
3368 strbuf_addf(path, "/%s", de->d_name);
3369
3370 if (strlen(de->d_name) == 38) {
3371 char hex[41];
3372 unsigned char sha1[20];
3373
3374 snprintf(hex, sizeof(hex), "%02x%s",
3375 subdir_nr, de->d_name);
3376 if (!get_sha1_hex(hex, sha1)) {
3377 if (obj_cb) {
3378 r = obj_cb(sha1, path->buf, data);
3379 if (r)
3380 break;
3381 }
3382 continue;
3383 }
3384 }
3385
3386 if (cruft_cb) {
3387 r = cruft_cb(de->d_name, path->buf, data);
3388 if (r)
3389 break;
3390 }
3391 }
3392 strbuf_setlen(path, baselen);
3393
3394 if (!r && subdir_cb)
3395 r = subdir_cb(subdir_nr, path->buf, data);
3396
3397 closedir(dir);
3398 return r;
3399 }
3400
3401 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3402 each_loose_object_fn obj_cb,
3403 each_loose_cruft_fn cruft_cb,
3404 each_loose_subdir_fn subdir_cb,
3405 void *data)
3406 {
3407 size_t baselen = path->len;
3408 int r = 0;
3409 int i;
3410
3411 for (i = 0; i < 256; i++) {
3412 strbuf_addf(path, "/%02x", i);
3413 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3414 subdir_cb, data);
3415 strbuf_setlen(path, baselen);
3416 if (r)
3417 break;
3418 }
3419
3420 return r;
3421 }
3422
3423 int for_each_loose_file_in_objdir(const char *path,
3424 each_loose_object_fn obj_cb,
3425 each_loose_cruft_fn cruft_cb,
3426 each_loose_subdir_fn subdir_cb,
3427 void *data)
3428 {
3429 struct strbuf buf = STRBUF_INIT;
3430 int r;
3431
3432 strbuf_addstr(&buf, path);
3433 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3434 subdir_cb, data);
3435 strbuf_release(&buf);
3436
3437 return r;
3438 }
3439
3440 struct loose_alt_odb_data {
3441 each_loose_object_fn *cb;
3442 void *data;
3443 };
3444
3445 static int loose_from_alt_odb(struct alternate_object_database *alt,
3446 void *vdata)
3447 {
3448 struct loose_alt_odb_data *data = vdata;
3449 struct strbuf buf = STRBUF_INIT;
3450 int r;
3451
3452 /* copy base not including trailing '/' */
3453 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3454 r = for_each_loose_file_in_objdir_buf(&buf,
3455 data->cb, NULL, NULL,
3456 data->data);
3457 strbuf_release(&buf);
3458 return r;
3459 }
3460
3461 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3462 {
3463 struct loose_alt_odb_data alt;
3464 int r;
3465
3466 r = for_each_loose_file_in_objdir(get_object_directory(),
3467 cb, NULL, NULL, data);
3468 if (r)
3469 return r;
3470
3471 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3472 return 0;
3473
3474 alt.cb = cb;
3475 alt.data = data;
3476 return foreach_alt_odb(loose_from_alt_odb, &alt);
3477 }
3478
3479 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3480 {
3481 uint32_t i;
3482 int r = 0;
3483
3484 for (i = 0; i < p->num_objects; i++) {
3485 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3486
3487 if (!sha1)
3488 return error("unable to get sha1 of object %u in %s",
3489 i, p->pack_name);
3490
3491 r = cb(sha1, p, i, data);
3492 if (r)
3493 break;
3494 }
3495 return r;
3496 }
3497
3498 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3499 {
3500 struct packed_git *p;
3501 int r = 0;
3502
3503 prepare_packed_git();
3504 for (p = packed_git; p; p = p->next) {
3505 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3506 continue;
3507 r = for_each_object_in_pack(p, cb, data);
3508 if (r)
3509 break;
3510 }
3511 return r;
3512 }