Sync with 2.3.9
[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 char *path;
381 int fd;
382
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
391 }
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
395
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
397
398 munmap(map, mapsz);
399 }
400
401 void add_to_alternates_file(const char *reference)
402 {
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
405 char *alt = mkpath("%s\n", reference);
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
411 }
412
413 int foreach_alt_odb(alt_odb_fn fn, void *cb)
414 {
415 struct alternate_object_database *ent;
416 int r = 0;
417
418 prepare_alt_odb();
419 for (ent = alt_odb_list; ent; ent = ent->next) {
420 r = fn(ent, cb);
421 if (r)
422 break;
423 }
424 return r;
425 }
426
427 void prepare_alt_odb(void)
428 {
429 const char *alt;
430
431 if (alt_odb_tail)
432 return;
433
434 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
435 if (!alt) alt = "";
436
437 alt_odb_tail = &alt_odb_list;
438 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
439
440 read_info_alternates(get_object_directory(), 0);
441 }
442
443 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
444 static int freshen_file(const char *fn)
445 {
446 struct utimbuf t;
447 t.actime = t.modtime = time(NULL);
448 return !utime(fn, &t);
449 }
450
451 /*
452 * All of the check_and_freshen functions return 1 if the file exists and was
453 * freshened (if freshening was requested), 0 otherwise. If they return
454 * 0, you should not assume that it is safe to skip a write of the object (it
455 * either does not exist on disk, or has a stale mtime and may be subject to
456 * pruning).
457 */
458 static int check_and_freshen_file(const char *fn, int freshen)
459 {
460 if (access(fn, F_OK))
461 return 0;
462 if (freshen && !freshen_file(fn))
463 return 0;
464 return 1;
465 }
466
467 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
468 {
469 return check_and_freshen_file(sha1_file_name(sha1), freshen);
470 }
471
472 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
473 {
474 struct alternate_object_database *alt;
475 prepare_alt_odb();
476 for (alt = alt_odb_list; alt; alt = alt->next) {
477 fill_sha1_path(alt->name, sha1);
478 if (check_and_freshen_file(alt->base, freshen))
479 return 1;
480 }
481 return 0;
482 }
483
484 static int check_and_freshen(const unsigned char *sha1, int freshen)
485 {
486 return check_and_freshen_local(sha1, freshen) ||
487 check_and_freshen_nonlocal(sha1, freshen);
488 }
489
490 int has_loose_object_nonlocal(const unsigned char *sha1)
491 {
492 return check_and_freshen_nonlocal(sha1, 0);
493 }
494
495 static int has_loose_object(const unsigned char *sha1)
496 {
497 return check_and_freshen(sha1, 0);
498 }
499
500 static unsigned int pack_used_ctr;
501 static unsigned int pack_mmap_calls;
502 static unsigned int peak_pack_open_windows;
503 static unsigned int pack_open_windows;
504 static unsigned int pack_open_fds;
505 static unsigned int pack_max_fds;
506 static size_t peak_pack_mapped;
507 static size_t pack_mapped;
508 struct packed_git *packed_git;
509
510 void pack_report(void)
511 {
512 fprintf(stderr,
513 "pack_report: getpagesize() = %10" SZ_FMT "\n"
514 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
515 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
516 sz_fmt(getpagesize()),
517 sz_fmt(packed_git_window_size),
518 sz_fmt(packed_git_limit));
519 fprintf(stderr,
520 "pack_report: pack_used_ctr = %10u\n"
521 "pack_report: pack_mmap_calls = %10u\n"
522 "pack_report: pack_open_windows = %10u / %10u\n"
523 "pack_report: pack_mapped = "
524 "%10" SZ_FMT " / %10" SZ_FMT "\n",
525 pack_used_ctr,
526 pack_mmap_calls,
527 pack_open_windows, peak_pack_open_windows,
528 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
529 }
530
531 /*
532 * Open and mmap the index file at path, perform a couple of
533 * consistency checks, then record its information to p. Return 0 on
534 * success.
535 */
536 static int check_packed_git_idx(const char *path, struct packed_git *p)
537 {
538 void *idx_map;
539 struct pack_idx_header *hdr;
540 size_t idx_size;
541 uint32_t version, nr, i, *index;
542 int fd = git_open_noatime(path);
543 struct stat st;
544
545 if (fd < 0)
546 return -1;
547 if (fstat(fd, &st)) {
548 close(fd);
549 return -1;
550 }
551 idx_size = xsize_t(st.st_size);
552 if (idx_size < 4 * 256 + 20 + 20) {
553 close(fd);
554 return error("index file %s is too small", path);
555 }
556 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
557 close(fd);
558
559 hdr = idx_map;
560 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
561 version = ntohl(hdr->idx_version);
562 if (version < 2 || version > 2) {
563 munmap(idx_map, idx_size);
564 return error("index file %s is version %"PRIu32
565 " and is not supported by this binary"
566 " (try upgrading GIT to a newer version)",
567 path, version);
568 }
569 } else
570 version = 1;
571
572 nr = 0;
573 index = idx_map;
574 if (version > 1)
575 index += 2; /* skip index header */
576 for (i = 0; i < 256; i++) {
577 uint32_t n = ntohl(index[i]);
578 if (n < nr) {
579 munmap(idx_map, idx_size);
580 return error("non-monotonic index %s", path);
581 }
582 nr = n;
583 }
584
585 if (version == 1) {
586 /*
587 * Total size:
588 * - 256 index entries 4 bytes each
589 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
590 * - 20-byte SHA1 of the packfile
591 * - 20-byte SHA1 file checksum
592 */
593 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
594 munmap(idx_map, idx_size);
595 return error("wrong index v1 file size in %s", path);
596 }
597 } else if (version == 2) {
598 /*
599 * Minimum size:
600 * - 8 bytes of header
601 * - 256 index entries 4 bytes each
602 * - 20-byte sha1 entry * nr
603 * - 4-byte crc entry * nr
604 * - 4-byte offset entry * nr
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 * And after the 4-byte offset table might be a
608 * variable sized table containing 8-byte entries
609 * for offsets larger than 2^31.
610 */
611 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
612 unsigned long max_size = min_size;
613 if (nr)
614 max_size += (nr - 1)*8;
615 if (idx_size < min_size || idx_size > max_size) {
616 munmap(idx_map, idx_size);
617 return error("wrong index v2 file size in %s", path);
618 }
619 if (idx_size != min_size &&
620 /*
621 * make sure we can deal with large pack offsets.
622 * 31-bit signed offset won't be enough, neither
623 * 32-bit unsigned one will be.
624 */
625 (sizeof(off_t) <= 4)) {
626 munmap(idx_map, idx_size);
627 return error("pack too large for current definition of off_t in %s", path);
628 }
629 }
630
631 p->index_version = version;
632 p->index_data = idx_map;
633 p->index_size = idx_size;
634 p->num_objects = nr;
635 return 0;
636 }
637
638 int open_pack_index(struct packed_git *p)
639 {
640 char *idx_name;
641 int ret;
642
643 if (p->index_data)
644 return 0;
645
646 idx_name = xstrdup(p->pack_name);
647 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
648 ret = check_packed_git_idx(idx_name, p);
649 free(idx_name);
650 return ret;
651 }
652
653 static void scan_windows(struct packed_git *p,
654 struct packed_git **lru_p,
655 struct pack_window **lru_w,
656 struct pack_window **lru_l)
657 {
658 struct pack_window *w, *w_l;
659
660 for (w_l = NULL, w = p->windows; w; w = w->next) {
661 if (!w->inuse_cnt) {
662 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
663 *lru_p = p;
664 *lru_w = w;
665 *lru_l = w_l;
666 }
667 }
668 w_l = w;
669 }
670 }
671
672 static int unuse_one_window(struct packed_git *current)
673 {
674 struct packed_git *p, *lru_p = NULL;
675 struct pack_window *lru_w = NULL, *lru_l = NULL;
676
677 if (current)
678 scan_windows(current, &lru_p, &lru_w, &lru_l);
679 for (p = packed_git; p; p = p->next)
680 scan_windows(p, &lru_p, &lru_w, &lru_l);
681 if (lru_p) {
682 munmap(lru_w->base, lru_w->len);
683 pack_mapped -= lru_w->len;
684 if (lru_l)
685 lru_l->next = lru_w->next;
686 else
687 lru_p->windows = lru_w->next;
688 free(lru_w);
689 pack_open_windows--;
690 return 1;
691 }
692 return 0;
693 }
694
695 void release_pack_memory(size_t need)
696 {
697 size_t cur = pack_mapped;
698 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
699 ; /* nothing */
700 }
701
702 static void mmap_limit_check(size_t length)
703 {
704 static size_t limit = 0;
705 if (!limit) {
706 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
707 if (!limit)
708 limit = SIZE_MAX;
709 }
710 if (length > limit)
711 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
712 (uintmax_t)length, (uintmax_t)limit);
713 }
714
715 void *xmmap_gently(void *start, size_t length,
716 int prot, int flags, int fd, off_t offset)
717 {
718 void *ret;
719
720 mmap_limit_check(length);
721 ret = mmap(start, length, prot, flags, fd, offset);
722 if (ret == MAP_FAILED) {
723 if (!length)
724 return NULL;
725 release_pack_memory(length);
726 ret = mmap(start, length, prot, flags, fd, offset);
727 }
728 return ret;
729 }
730
731 void *xmmap(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
733 {
734 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
735 if (ret == MAP_FAILED)
736 die_errno("mmap failed");
737 return ret;
738 }
739
740 void close_pack_windows(struct packed_git *p)
741 {
742 while (p->windows) {
743 struct pack_window *w = p->windows;
744
745 if (w->inuse_cnt)
746 die("pack '%s' still has open windows to it",
747 p->pack_name);
748 munmap(w->base, w->len);
749 pack_mapped -= w->len;
750 pack_open_windows--;
751 p->windows = w->next;
752 free(w);
753 }
754 }
755
756 /*
757 * The LRU pack is the one with the oldest MRU window, preferring packs
758 * with no used windows, or the oldest mtime if it has no windows allocated.
759 */
760 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
761 {
762 struct pack_window *w, *this_mru_w;
763 int has_windows_inuse = 0;
764
765 /*
766 * Reject this pack if it has windows and the previously selected
767 * one does not. If this pack does not have windows, reject
768 * it if the pack file is newer than the previously selected one.
769 */
770 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
771 return;
772
773 for (w = this_mru_w = p->windows; w; w = w->next) {
774 /*
775 * Reject this pack if any of its windows are in use,
776 * but the previously selected pack did not have any
777 * inuse windows. Otherwise, record that this pack
778 * has windows in use.
779 */
780 if (w->inuse_cnt) {
781 if (*accept_windows_inuse)
782 has_windows_inuse = 1;
783 else
784 return;
785 }
786
787 if (w->last_used > this_mru_w->last_used)
788 this_mru_w = w;
789
790 /*
791 * Reject this pack if it has windows that have been
792 * used more recently than the previously selected pack.
793 * If the previously selected pack had windows inuse and
794 * we have not encountered a window in this pack that is
795 * inuse, skip this check since we prefer a pack with no
796 * inuse windows to one that has inuse windows.
797 */
798 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
799 this_mru_w->last_used > (*mru_w)->last_used)
800 return;
801 }
802
803 /*
804 * Select this pack.
805 */
806 *mru_w = this_mru_w;
807 *lru_p = p;
808 *accept_windows_inuse = has_windows_inuse;
809 }
810
811 static int close_one_pack(void)
812 {
813 struct packed_git *p, *lru_p = NULL;
814 struct pack_window *mru_w = NULL;
815 int accept_windows_inuse = 1;
816
817 for (p = packed_git; p; p = p->next) {
818 if (p->pack_fd == -1)
819 continue;
820 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
821 }
822
823 if (lru_p) {
824 close(lru_p->pack_fd);
825 pack_open_fds--;
826 lru_p->pack_fd = -1;
827 return 1;
828 }
829
830 return 0;
831 }
832
833 void unuse_pack(struct pack_window **w_cursor)
834 {
835 struct pack_window *w = *w_cursor;
836 if (w) {
837 w->inuse_cnt--;
838 *w_cursor = NULL;
839 }
840 }
841
842 void close_pack_index(struct packed_git *p)
843 {
844 if (p->index_data) {
845 munmap((void *)p->index_data, p->index_size);
846 p->index_data = NULL;
847 }
848 }
849
850 /*
851 * This is used by git-repack in case a newly created pack happens to
852 * contain the same set of objects as an existing one. In that case
853 * the resulting file might be different even if its name would be the
854 * same. It is best to close any reference to the old pack before it is
855 * replaced on disk. Of course no index pointers or windows for given pack
856 * must subsist at this point. If ever objects from this pack are requested
857 * again, the new version of the pack will be reinitialized through
858 * reprepare_packed_git().
859 */
860 void free_pack_by_name(const char *pack_name)
861 {
862 struct packed_git *p, **pp = &packed_git;
863
864 while (*pp) {
865 p = *pp;
866 if (strcmp(pack_name, p->pack_name) == 0) {
867 clear_delta_base_cache();
868 close_pack_windows(p);
869 if (p->pack_fd != -1) {
870 close(p->pack_fd);
871 pack_open_fds--;
872 }
873 close_pack_index(p);
874 free(p->bad_object_sha1);
875 *pp = p->next;
876 if (last_found_pack == p)
877 last_found_pack = NULL;
878 free(p);
879 return;
880 }
881 pp = &p->next;
882 }
883 }
884
885 static unsigned int get_max_fd_limit(void)
886 {
887 #ifdef RLIMIT_NOFILE
888 {
889 struct rlimit lim;
890
891 if (!getrlimit(RLIMIT_NOFILE, &lim))
892 return lim.rlim_cur;
893 }
894 #endif
895
896 #ifdef _SC_OPEN_MAX
897 {
898 long open_max = sysconf(_SC_OPEN_MAX);
899 if (0 < open_max)
900 return open_max;
901 /*
902 * Otherwise, we got -1 for one of the two
903 * reasons:
904 *
905 * (1) sysconf() did not understand _SC_OPEN_MAX
906 * and signaled an error with -1; or
907 * (2) sysconf() said there is no limit.
908 *
909 * We _could_ clear errno before calling sysconf() to
910 * tell these two cases apart and return a huge number
911 * in the latter case to let the caller cap it to a
912 * value that is not so selfish, but letting the
913 * fallback OPEN_MAX codepath take care of these cases
914 * is a lot simpler.
915 */
916 }
917 #endif
918
919 #ifdef OPEN_MAX
920 return OPEN_MAX;
921 #else
922 return 1; /* see the caller ;-) */
923 #endif
924 }
925
926 /*
927 * Do not call this directly as this leaks p->pack_fd on error return;
928 * call open_packed_git() instead.
929 */
930 static int open_packed_git_1(struct packed_git *p)
931 {
932 struct stat st;
933 struct pack_header hdr;
934 unsigned char sha1[20];
935 unsigned char *idx_sha1;
936 long fd_flag;
937
938 if (!p->index_data && open_pack_index(p))
939 return error("packfile %s index unavailable", p->pack_name);
940
941 if (!pack_max_fds) {
942 unsigned int max_fds = get_max_fd_limit();
943
944 /* Save 3 for stdin/stdout/stderr, 22 for work */
945 if (25 < max_fds)
946 pack_max_fds = max_fds - 25;
947 else
948 pack_max_fds = 1;
949 }
950
951 while (pack_max_fds <= pack_open_fds && close_one_pack())
952 ; /* nothing */
953
954 p->pack_fd = git_open_noatime(p->pack_name);
955 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
956 return -1;
957 pack_open_fds++;
958
959 /* If we created the struct before we had the pack we lack size. */
960 if (!p->pack_size) {
961 if (!S_ISREG(st.st_mode))
962 return error("packfile %s not a regular file", p->pack_name);
963 p->pack_size = st.st_size;
964 } else if (p->pack_size != st.st_size)
965 return error("packfile %s size changed", p->pack_name);
966
967 /* We leave these file descriptors open with sliding mmap;
968 * there is no point keeping them open across exec(), though.
969 */
970 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
971 if (fd_flag < 0)
972 return error("cannot determine file descriptor flags");
973 fd_flag |= FD_CLOEXEC;
974 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
975 return error("cannot set FD_CLOEXEC");
976
977 /* Verify we recognize this pack file format. */
978 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
979 return error("file %s is far too short to be a packfile", p->pack_name);
980 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
981 return error("file %s is not a GIT packfile", p->pack_name);
982 if (!pack_version_ok(hdr.hdr_version))
983 return error("packfile %s is version %"PRIu32" and not"
984 " supported (try upgrading GIT to a newer version)",
985 p->pack_name, ntohl(hdr.hdr_version));
986
987 /* Verify the pack matches its index. */
988 if (p->num_objects != ntohl(hdr.hdr_entries))
989 return error("packfile %s claims to have %"PRIu32" objects"
990 " while index indicates %"PRIu32" objects",
991 p->pack_name, ntohl(hdr.hdr_entries),
992 p->num_objects);
993 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
994 return error("end of packfile %s is unavailable", p->pack_name);
995 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
996 return error("packfile %s signature is unavailable", p->pack_name);
997 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
998 if (hashcmp(sha1, idx_sha1))
999 return error("packfile %s does not match index", p->pack_name);
1000 return 0;
1001 }
1002
1003 static int open_packed_git(struct packed_git *p)
1004 {
1005 if (!open_packed_git_1(p))
1006 return 0;
1007 if (p->pack_fd != -1) {
1008 close(p->pack_fd);
1009 pack_open_fds--;
1010 p->pack_fd = -1;
1011 }
1012 return -1;
1013 }
1014
1015 static int in_window(struct pack_window *win, off_t offset)
1016 {
1017 /* We must promise at least 20 bytes (one hash) after the
1018 * offset is available from this window, otherwise the offset
1019 * is not actually in this window and a different window (which
1020 * has that one hash excess) must be used. This is to support
1021 * the object header and delta base parsing routines below.
1022 */
1023 off_t win_off = win->offset;
1024 return win_off <= offset
1025 && (offset + 20) <= (win_off + win->len);
1026 }
1027
1028 unsigned char *use_pack(struct packed_git *p,
1029 struct pack_window **w_cursor,
1030 off_t offset,
1031 unsigned long *left)
1032 {
1033 struct pack_window *win = *w_cursor;
1034
1035 /* Since packfiles end in a hash of their content and it's
1036 * pointless to ask for an offset into the middle of that
1037 * hash, and the in_window function above wouldn't match
1038 * don't allow an offset too close to the end of the file.
1039 */
1040 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1041 die("packfile %s cannot be accessed", p->pack_name);
1042 if (offset > (p->pack_size - 20))
1043 die("offset beyond end of packfile (truncated pack?)");
1044
1045 if (!win || !in_window(win, offset)) {
1046 if (win)
1047 win->inuse_cnt--;
1048 for (win = p->windows; win; win = win->next) {
1049 if (in_window(win, offset))
1050 break;
1051 }
1052 if (!win) {
1053 size_t window_align = packed_git_window_size / 2;
1054 off_t len;
1055
1056 if (p->pack_fd == -1 && open_packed_git(p))
1057 die("packfile %s cannot be accessed", p->pack_name);
1058
1059 win = xcalloc(1, sizeof(*win));
1060 win->offset = (offset / window_align) * window_align;
1061 len = p->pack_size - win->offset;
1062 if (len > packed_git_window_size)
1063 len = packed_git_window_size;
1064 win->len = (size_t)len;
1065 pack_mapped += win->len;
1066 while (packed_git_limit < pack_mapped
1067 && unuse_one_window(p))
1068 ; /* nothing */
1069 win->base = xmmap(NULL, win->len,
1070 PROT_READ, MAP_PRIVATE,
1071 p->pack_fd, win->offset);
1072 if (win->base == MAP_FAILED)
1073 die("packfile %s cannot be mapped: %s",
1074 p->pack_name,
1075 strerror(errno));
1076 if (!win->offset && win->len == p->pack_size
1077 && !p->do_not_close) {
1078 close(p->pack_fd);
1079 pack_open_fds--;
1080 p->pack_fd = -1;
1081 }
1082 pack_mmap_calls++;
1083 pack_open_windows++;
1084 if (pack_mapped > peak_pack_mapped)
1085 peak_pack_mapped = pack_mapped;
1086 if (pack_open_windows > peak_pack_open_windows)
1087 peak_pack_open_windows = pack_open_windows;
1088 win->next = p->windows;
1089 p->windows = win;
1090 }
1091 }
1092 if (win != *w_cursor) {
1093 win->last_used = pack_used_ctr++;
1094 win->inuse_cnt++;
1095 *w_cursor = win;
1096 }
1097 offset -= win->offset;
1098 if (left)
1099 *left = win->len - xsize_t(offset);
1100 return win->base + offset;
1101 }
1102
1103 static struct packed_git *alloc_packed_git(int extra)
1104 {
1105 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1106 memset(p, 0, sizeof(*p));
1107 p->pack_fd = -1;
1108 return p;
1109 }
1110
1111 static void try_to_free_pack_memory(size_t size)
1112 {
1113 release_pack_memory(size);
1114 }
1115
1116 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1117 {
1118 static int have_set_try_to_free_routine;
1119 struct stat st;
1120 struct packed_git *p = alloc_packed_git(path_len + 2);
1121
1122 if (!have_set_try_to_free_routine) {
1123 have_set_try_to_free_routine = 1;
1124 set_try_to_free_routine(try_to_free_pack_memory);
1125 }
1126
1127 /*
1128 * Make sure a corresponding .pack file exists and that
1129 * the index looks sane.
1130 */
1131 path_len -= strlen(".idx");
1132 if (path_len < 1) {
1133 free(p);
1134 return NULL;
1135 }
1136 memcpy(p->pack_name, path, path_len);
1137
1138 strcpy(p->pack_name + path_len, ".keep");
1139 if (!access(p->pack_name, F_OK))
1140 p->pack_keep = 1;
1141
1142 strcpy(p->pack_name + path_len, ".pack");
1143 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1144 free(p);
1145 return NULL;
1146 }
1147
1148 /* ok, it looks sane as far as we can check without
1149 * actually mapping the pack file.
1150 */
1151 p->pack_size = st.st_size;
1152 p->pack_local = local;
1153 p->mtime = st.st_mtime;
1154 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1155 hashclr(p->sha1);
1156 return p;
1157 }
1158
1159 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1160 {
1161 const char *path = sha1_pack_name(sha1);
1162 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1163
1164 strcpy(p->pack_name, path);
1165 hashcpy(p->sha1, sha1);
1166 if (check_packed_git_idx(idx_path, p)) {
1167 free(p);
1168 return NULL;
1169 }
1170
1171 return p;
1172 }
1173
1174 void install_packed_git(struct packed_git *pack)
1175 {
1176 if (pack->pack_fd != -1)
1177 pack_open_fds++;
1178
1179 pack->next = packed_git;
1180 packed_git = pack;
1181 }
1182
1183 void (*report_garbage)(const char *desc, const char *path);
1184
1185 static void report_helper(const struct string_list *list,
1186 int seen_bits, int first, int last)
1187 {
1188 const char *msg;
1189 switch (seen_bits) {
1190 case 0:
1191 msg = "no corresponding .idx or .pack";
1192 break;
1193 case 1:
1194 msg = "no corresponding .idx";
1195 break;
1196 case 2:
1197 msg = "no corresponding .pack";
1198 break;
1199 default:
1200 return;
1201 }
1202 for (; first < last; first++)
1203 report_garbage(msg, list->items[first].string);
1204 }
1205
1206 static void report_pack_garbage(struct string_list *list)
1207 {
1208 int i, baselen = -1, first = 0, seen_bits = 0;
1209
1210 if (!report_garbage)
1211 return;
1212
1213 string_list_sort(list);
1214
1215 for (i = 0; i < list->nr; i++) {
1216 const char *path = list->items[i].string;
1217 if (baselen != -1 &&
1218 strncmp(path, list->items[first].string, baselen)) {
1219 report_helper(list, seen_bits, first, i);
1220 baselen = -1;
1221 seen_bits = 0;
1222 }
1223 if (baselen == -1) {
1224 const char *dot = strrchr(path, '.');
1225 if (!dot) {
1226 report_garbage("garbage found", path);
1227 continue;
1228 }
1229 baselen = dot - path + 1;
1230 first = i;
1231 }
1232 if (!strcmp(path + baselen, "pack"))
1233 seen_bits |= 1;
1234 else if (!strcmp(path + baselen, "idx"))
1235 seen_bits |= 2;
1236 }
1237 report_helper(list, seen_bits, first, list->nr);
1238 }
1239
1240 static void prepare_packed_git_one(char *objdir, int local)
1241 {
1242 struct strbuf path = STRBUF_INIT;
1243 size_t dirnamelen;
1244 DIR *dir;
1245 struct dirent *de;
1246 struct string_list garbage = STRING_LIST_INIT_DUP;
1247
1248 strbuf_addstr(&path, objdir);
1249 strbuf_addstr(&path, "/pack");
1250 dir = opendir(path.buf);
1251 if (!dir) {
1252 if (errno != ENOENT)
1253 error("unable to open object pack directory: %s: %s",
1254 path.buf, strerror(errno));
1255 strbuf_release(&path);
1256 return;
1257 }
1258 strbuf_addch(&path, '/');
1259 dirnamelen = path.len;
1260 while ((de = readdir(dir)) != NULL) {
1261 struct packed_git *p;
1262 size_t base_len;
1263
1264 if (is_dot_or_dotdot(de->d_name))
1265 continue;
1266
1267 strbuf_setlen(&path, dirnamelen);
1268 strbuf_addstr(&path, de->d_name);
1269
1270 base_len = path.len;
1271 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1272 /* Don't reopen a pack we already have. */
1273 for (p = packed_git; p; p = p->next) {
1274 size_t len;
1275 if (strip_suffix(p->pack_name, ".pack", &len) &&
1276 len == base_len &&
1277 !memcmp(p->pack_name, path.buf, len))
1278 break;
1279 }
1280 if (p == NULL &&
1281 /*
1282 * See if it really is a valid .idx file with
1283 * corresponding .pack file that we can map.
1284 */
1285 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1286 install_packed_git(p);
1287 }
1288
1289 if (!report_garbage)
1290 continue;
1291
1292 if (ends_with(de->d_name, ".idx") ||
1293 ends_with(de->d_name, ".pack") ||
1294 ends_with(de->d_name, ".bitmap") ||
1295 ends_with(de->d_name, ".keep"))
1296 string_list_append(&garbage, path.buf);
1297 else
1298 report_garbage("garbage found", path.buf);
1299 }
1300 closedir(dir);
1301 report_pack_garbage(&garbage);
1302 string_list_clear(&garbage, 0);
1303 strbuf_release(&path);
1304 }
1305
1306 static int sort_pack(const void *a_, const void *b_)
1307 {
1308 struct packed_git *a = *((struct packed_git **)a_);
1309 struct packed_git *b = *((struct packed_git **)b_);
1310 int st;
1311
1312 /*
1313 * Local packs tend to contain objects specific to our
1314 * variant of the project than remote ones. In addition,
1315 * remote ones could be on a network mounted filesystem.
1316 * Favor local ones for these reasons.
1317 */
1318 st = a->pack_local - b->pack_local;
1319 if (st)
1320 return -st;
1321
1322 /*
1323 * Younger packs tend to contain more recent objects,
1324 * and more recent objects tend to get accessed more
1325 * often.
1326 */
1327 if (a->mtime < b->mtime)
1328 return 1;
1329 else if (a->mtime == b->mtime)
1330 return 0;
1331 return -1;
1332 }
1333
1334 static void rearrange_packed_git(void)
1335 {
1336 struct packed_git **ary, *p;
1337 int i, n;
1338
1339 for (n = 0, p = packed_git; p; p = p->next)
1340 n++;
1341 if (n < 2)
1342 return;
1343
1344 /* prepare an array of packed_git for easier sorting */
1345 ary = xcalloc(n, sizeof(struct packed_git *));
1346 for (n = 0, p = packed_git; p; p = p->next)
1347 ary[n++] = p;
1348
1349 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1350
1351 /* link them back again */
1352 for (i = 0; i < n - 1; i++)
1353 ary[i]->next = ary[i + 1];
1354 ary[n - 1]->next = NULL;
1355 packed_git = ary[0];
1356
1357 free(ary);
1358 }
1359
1360 static int prepare_packed_git_run_once = 0;
1361 void prepare_packed_git(void)
1362 {
1363 struct alternate_object_database *alt;
1364
1365 if (prepare_packed_git_run_once)
1366 return;
1367 prepare_packed_git_one(get_object_directory(), 1);
1368 prepare_alt_odb();
1369 for (alt = alt_odb_list; alt; alt = alt->next) {
1370 alt->name[-1] = 0;
1371 prepare_packed_git_one(alt->base, 0);
1372 alt->name[-1] = '/';
1373 }
1374 rearrange_packed_git();
1375 prepare_packed_git_run_once = 1;
1376 }
1377
1378 void reprepare_packed_git(void)
1379 {
1380 prepare_packed_git_run_once = 0;
1381 prepare_packed_git();
1382 }
1383
1384 static void mark_bad_packed_object(struct packed_git *p,
1385 const unsigned char *sha1)
1386 {
1387 unsigned i;
1388 for (i = 0; i < p->num_bad_objects; i++)
1389 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1390 return;
1391 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1392 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1393 p->num_bad_objects++;
1394 }
1395
1396 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1397 {
1398 struct packed_git *p;
1399 unsigned i;
1400
1401 for (p = packed_git; p; p = p->next)
1402 for (i = 0; i < p->num_bad_objects; i++)
1403 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1404 return p;
1405 return NULL;
1406 }
1407
1408 /*
1409 * With an in-core object data in "map", rehash it to make sure the
1410 * object name actually matches "sha1" to detect object corruption.
1411 * With "map" == NULL, try reading the object named with "sha1" using
1412 * the streaming interface and rehash it to do the same.
1413 */
1414 int check_sha1_signature(const unsigned char *sha1, void *map,
1415 unsigned long size, const char *type)
1416 {
1417 unsigned char real_sha1[20];
1418 enum object_type obj_type;
1419 struct git_istream *st;
1420 git_SHA_CTX c;
1421 char hdr[32];
1422 int hdrlen;
1423
1424 if (map) {
1425 hash_sha1_file(map, size, type, real_sha1);
1426 return hashcmp(sha1, real_sha1) ? -1 : 0;
1427 }
1428
1429 st = open_istream(sha1, &obj_type, &size, NULL);
1430 if (!st)
1431 return -1;
1432
1433 /* Generate the header */
1434 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1435
1436 /* Sha1.. */
1437 git_SHA1_Init(&c);
1438 git_SHA1_Update(&c, hdr, hdrlen);
1439 for (;;) {
1440 char buf[1024 * 16];
1441 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1442
1443 if (readlen < 0) {
1444 close_istream(st);
1445 return -1;
1446 }
1447 if (!readlen)
1448 break;
1449 git_SHA1_Update(&c, buf, readlen);
1450 }
1451 git_SHA1_Final(real_sha1, &c);
1452 close_istream(st);
1453 return hashcmp(sha1, real_sha1) ? -1 : 0;
1454 }
1455
1456 int git_open_noatime(const char *name)
1457 {
1458 static int sha1_file_open_flag = O_NOATIME;
1459
1460 for (;;) {
1461 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1462 if (fd >= 0)
1463 return fd;
1464
1465 /* Might the failure be due to O_NOATIME? */
1466 if (errno != ENOENT && sha1_file_open_flag) {
1467 sha1_file_open_flag = 0;
1468 continue;
1469 }
1470
1471 return -1;
1472 }
1473 }
1474
1475 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1476 {
1477 struct alternate_object_database *alt;
1478
1479 if (!lstat(sha1_file_name(sha1), st))
1480 return 0;
1481
1482 prepare_alt_odb();
1483 errno = ENOENT;
1484 for (alt = alt_odb_list; alt; alt = alt->next) {
1485 fill_sha1_path(alt->name, sha1);
1486 if (!lstat(alt->base, st))
1487 return 0;
1488 }
1489
1490 return -1;
1491 }
1492
1493 static int open_sha1_file(const unsigned char *sha1)
1494 {
1495 int fd;
1496 struct alternate_object_database *alt;
1497 int most_interesting_errno;
1498
1499 fd = git_open_noatime(sha1_file_name(sha1));
1500 if (fd >= 0)
1501 return fd;
1502 most_interesting_errno = errno;
1503
1504 prepare_alt_odb();
1505 for (alt = alt_odb_list; alt; alt = alt->next) {
1506 fill_sha1_path(alt->name, sha1);
1507 fd = git_open_noatime(alt->base);
1508 if (fd >= 0)
1509 return fd;
1510 if (most_interesting_errno == ENOENT)
1511 most_interesting_errno = errno;
1512 }
1513 errno = most_interesting_errno;
1514 return -1;
1515 }
1516
1517 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1518 {
1519 void *map;
1520 int fd;
1521
1522 fd = open_sha1_file(sha1);
1523 map = NULL;
1524 if (fd >= 0) {
1525 struct stat st;
1526
1527 if (!fstat(fd, &st)) {
1528 *size = xsize_t(st.st_size);
1529 if (!*size) {
1530 /* mmap() is forbidden on empty files */
1531 error("object file %s is empty", sha1_file_name(sha1));
1532 return NULL;
1533 }
1534 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1535 }
1536 close(fd);
1537 }
1538 return map;
1539 }
1540
1541 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1542 unsigned long len, enum object_type *type, unsigned long *sizep)
1543 {
1544 unsigned shift;
1545 unsigned long size, c;
1546 unsigned long used = 0;
1547
1548 c = buf[used++];
1549 *type = (c >> 4) & 7;
1550 size = c & 15;
1551 shift = 4;
1552 while (c & 0x80) {
1553 if (len <= used || bitsizeof(long) <= shift) {
1554 error("bad object header");
1555 size = used = 0;
1556 break;
1557 }
1558 c = buf[used++];
1559 size += (c & 0x7f) << shift;
1560 shift += 7;
1561 }
1562 *sizep = size;
1563 return used;
1564 }
1565
1566 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1567 {
1568 /* Get the data stream */
1569 memset(stream, 0, sizeof(*stream));
1570 stream->next_in = map;
1571 stream->avail_in = mapsize;
1572 stream->next_out = buffer;
1573 stream->avail_out = bufsiz;
1574
1575 git_inflate_init(stream);
1576 return git_inflate(stream, 0);
1577 }
1578
1579 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1580 {
1581 int bytes = strlen(buffer) + 1;
1582 unsigned char *buf = xmallocz(size);
1583 unsigned long n;
1584 int status = Z_OK;
1585
1586 n = stream->total_out - bytes;
1587 if (n > size)
1588 n = size;
1589 memcpy(buf, (char *) buffer + bytes, n);
1590 bytes = n;
1591 if (bytes <= size) {
1592 /*
1593 * The above condition must be (bytes <= size), not
1594 * (bytes < size). In other words, even though we
1595 * expect no more output and set avail_out to zero,
1596 * the input zlib stream may have bytes that express
1597 * "this concludes the stream", and we *do* want to
1598 * eat that input.
1599 *
1600 * Otherwise we would not be able to test that we
1601 * consumed all the input to reach the expected size;
1602 * we also want to check that zlib tells us that all
1603 * went well with status == Z_STREAM_END at the end.
1604 */
1605 stream->next_out = buf + bytes;
1606 stream->avail_out = size - bytes;
1607 while (status == Z_OK)
1608 status = git_inflate(stream, Z_FINISH);
1609 }
1610 if (status == Z_STREAM_END && !stream->avail_in) {
1611 git_inflate_end(stream);
1612 return buf;
1613 }
1614
1615 if (status < 0)
1616 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1617 else if (stream->avail_in)
1618 error("garbage at end of loose object '%s'",
1619 sha1_to_hex(sha1));
1620 free(buf);
1621 return NULL;
1622 }
1623
1624 /*
1625 * We used to just use "sscanf()", but that's actually way
1626 * too permissive for what we want to check. So do an anal
1627 * object header parse by hand.
1628 */
1629 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1630 {
1631 char type[10];
1632 int i;
1633 unsigned long size;
1634
1635 /*
1636 * The type can be at most ten bytes (including the
1637 * terminating '\0' that we add), and is followed by
1638 * a space.
1639 */
1640 i = 0;
1641 for (;;) {
1642 char c = *hdr++;
1643 if (c == ' ')
1644 break;
1645 type[i++] = c;
1646 if (i >= sizeof(type))
1647 return -1;
1648 }
1649 type[i] = 0;
1650
1651 /*
1652 * The length must follow immediately, and be in canonical
1653 * decimal format (ie "010" is not valid).
1654 */
1655 size = *hdr++ - '0';
1656 if (size > 9)
1657 return -1;
1658 if (size) {
1659 for (;;) {
1660 unsigned long c = *hdr - '0';
1661 if (c > 9)
1662 break;
1663 hdr++;
1664 size = size * 10 + c;
1665 }
1666 }
1667 *sizep = size;
1668
1669 /*
1670 * The length must be followed by a zero byte
1671 */
1672 return *hdr ? -1 : type_from_string(type);
1673 }
1674
1675 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1676 {
1677 int ret;
1678 git_zstream stream;
1679 char hdr[8192];
1680
1681 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1682 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1683 return NULL;
1684
1685 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1686 }
1687
1688 unsigned long get_size_from_delta(struct packed_git *p,
1689 struct pack_window **w_curs,
1690 off_t curpos)
1691 {
1692 const unsigned char *data;
1693 unsigned char delta_head[20], *in;
1694 git_zstream stream;
1695 int st;
1696
1697 memset(&stream, 0, sizeof(stream));
1698 stream.next_out = delta_head;
1699 stream.avail_out = sizeof(delta_head);
1700
1701 git_inflate_init(&stream);
1702 do {
1703 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1704 stream.next_in = in;
1705 st = git_inflate(&stream, Z_FINISH);
1706 curpos += stream.next_in - in;
1707 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1708 stream.total_out < sizeof(delta_head));
1709 git_inflate_end(&stream);
1710 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1711 error("delta data unpack-initial failed");
1712 return 0;
1713 }
1714
1715 /* Examine the initial part of the delta to figure out
1716 * the result size.
1717 */
1718 data = delta_head;
1719
1720 /* ignore base size */
1721 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1722
1723 /* Read the result size */
1724 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1725 }
1726
1727 static off_t get_delta_base(struct packed_git *p,
1728 struct pack_window **w_curs,
1729 off_t *curpos,
1730 enum object_type type,
1731 off_t delta_obj_offset)
1732 {
1733 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1734 off_t base_offset;
1735
1736 /* use_pack() assured us we have [base_info, base_info + 20)
1737 * as a range that we can look at without walking off the
1738 * end of the mapped window. Its actually the hash size
1739 * that is assured. An OFS_DELTA longer than the hash size
1740 * is stupid, as then a REF_DELTA would be smaller to store.
1741 */
1742 if (type == OBJ_OFS_DELTA) {
1743 unsigned used = 0;
1744 unsigned char c = base_info[used++];
1745 base_offset = c & 127;
1746 while (c & 128) {
1747 base_offset += 1;
1748 if (!base_offset || MSB(base_offset, 7))
1749 return 0; /* overflow */
1750 c = base_info[used++];
1751 base_offset = (base_offset << 7) + (c & 127);
1752 }
1753 base_offset = delta_obj_offset - base_offset;
1754 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1755 return 0; /* out of bound */
1756 *curpos += used;
1757 } else if (type == OBJ_REF_DELTA) {
1758 /* The base entry _must_ be in the same pack */
1759 base_offset = find_pack_entry_one(base_info, p);
1760 *curpos += 20;
1761 } else
1762 die("I am totally screwed");
1763 return base_offset;
1764 }
1765
1766 /*
1767 * Like get_delta_base above, but we return the sha1 instead of the pack
1768 * offset. This means it is cheaper for REF deltas (we do not have to do
1769 * the final object lookup), but more expensive for OFS deltas (we
1770 * have to load the revidx to convert the offset back into a sha1).
1771 */
1772 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1773 struct pack_window **w_curs,
1774 off_t curpos,
1775 enum object_type type,
1776 off_t delta_obj_offset)
1777 {
1778 if (type == OBJ_REF_DELTA) {
1779 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1780 return base;
1781 } else if (type == OBJ_OFS_DELTA) {
1782 struct revindex_entry *revidx;
1783 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1784 type, delta_obj_offset);
1785
1786 if (!base_offset)
1787 return NULL;
1788
1789 revidx = find_pack_revindex(p, base_offset);
1790 if (!revidx)
1791 return NULL;
1792
1793 return nth_packed_object_sha1(p, revidx->nr);
1794 } else
1795 return NULL;
1796 }
1797
1798 int unpack_object_header(struct packed_git *p,
1799 struct pack_window **w_curs,
1800 off_t *curpos,
1801 unsigned long *sizep)
1802 {
1803 unsigned char *base;
1804 unsigned long left;
1805 unsigned long used;
1806 enum object_type type;
1807
1808 /* use_pack() assures us we have [base, base + 20) available
1809 * as a range that we can look at. (Its actually the hash
1810 * size that is assured.) With our object header encoding
1811 * the maximum deflated object size is 2^137, which is just
1812 * insane, so we know won't exceed what we have been given.
1813 */
1814 base = use_pack(p, w_curs, *curpos, &left);
1815 used = unpack_object_header_buffer(base, left, &type, sizep);
1816 if (!used) {
1817 type = OBJ_BAD;
1818 } else
1819 *curpos += used;
1820
1821 return type;
1822 }
1823
1824 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1825 {
1826 int type;
1827 struct revindex_entry *revidx;
1828 const unsigned char *sha1;
1829 revidx = find_pack_revindex(p, obj_offset);
1830 if (!revidx)
1831 return OBJ_BAD;
1832 sha1 = nth_packed_object_sha1(p, revidx->nr);
1833 mark_bad_packed_object(p, sha1);
1834 type = sha1_object_info(sha1, NULL);
1835 if (type <= OBJ_NONE)
1836 return OBJ_BAD;
1837 return type;
1838 }
1839
1840 #define POI_STACK_PREALLOC 64
1841
1842 static enum object_type packed_to_object_type(struct packed_git *p,
1843 off_t obj_offset,
1844 enum object_type type,
1845 struct pack_window **w_curs,
1846 off_t curpos)
1847 {
1848 off_t small_poi_stack[POI_STACK_PREALLOC];
1849 off_t *poi_stack = small_poi_stack;
1850 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1851
1852 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1853 off_t base_offset;
1854 unsigned long size;
1855 /* Push the object we're going to leave behind */
1856 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1857 poi_stack_alloc = alloc_nr(poi_stack_nr);
1858 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1859 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1860 } else {
1861 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1862 }
1863 poi_stack[poi_stack_nr++] = obj_offset;
1864 /* If parsing the base offset fails, just unwind */
1865 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1866 if (!base_offset)
1867 goto unwind;
1868 curpos = obj_offset = base_offset;
1869 type = unpack_object_header(p, w_curs, &curpos, &size);
1870 if (type <= OBJ_NONE) {
1871 /* If getting the base itself fails, we first
1872 * retry the base, otherwise unwind */
1873 type = retry_bad_packed_offset(p, base_offset);
1874 if (type > OBJ_NONE)
1875 goto out;
1876 goto unwind;
1877 }
1878 }
1879
1880 switch (type) {
1881 case OBJ_BAD:
1882 case OBJ_COMMIT:
1883 case OBJ_TREE:
1884 case OBJ_BLOB:
1885 case OBJ_TAG:
1886 break;
1887 default:
1888 error("unknown object type %i at offset %"PRIuMAX" in %s",
1889 type, (uintmax_t)obj_offset, p->pack_name);
1890 type = OBJ_BAD;
1891 }
1892
1893 out:
1894 if (poi_stack != small_poi_stack)
1895 free(poi_stack);
1896 return type;
1897
1898 unwind:
1899 while (poi_stack_nr) {
1900 obj_offset = poi_stack[--poi_stack_nr];
1901 type = retry_bad_packed_offset(p, obj_offset);
1902 if (type > OBJ_NONE)
1903 goto out;
1904 }
1905 type = OBJ_BAD;
1906 goto out;
1907 }
1908
1909 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1910 struct object_info *oi)
1911 {
1912 struct pack_window *w_curs = NULL;
1913 unsigned long size;
1914 off_t curpos = obj_offset;
1915 enum object_type type;
1916
1917 /*
1918 * We always get the representation type, but only convert it to
1919 * a "real" type later if the caller is interested.
1920 */
1921 type = unpack_object_header(p, &w_curs, &curpos, &size);
1922
1923 if (oi->sizep) {
1924 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1925 off_t tmp_pos = curpos;
1926 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1927 type, obj_offset);
1928 if (!base_offset) {
1929 type = OBJ_BAD;
1930 goto out;
1931 }
1932 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1933 if (*oi->sizep == 0) {
1934 type = OBJ_BAD;
1935 goto out;
1936 }
1937 } else {
1938 *oi->sizep = size;
1939 }
1940 }
1941
1942 if (oi->disk_sizep) {
1943 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1944 *oi->disk_sizep = revidx[1].offset - obj_offset;
1945 }
1946
1947 if (oi->typep) {
1948 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1949 if (*oi->typep < 0) {
1950 type = OBJ_BAD;
1951 goto out;
1952 }
1953 }
1954
1955 if (oi->delta_base_sha1) {
1956 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1957 const unsigned char *base;
1958
1959 base = get_delta_base_sha1(p, &w_curs, curpos,
1960 type, obj_offset);
1961 if (!base) {
1962 type = OBJ_BAD;
1963 goto out;
1964 }
1965
1966 hashcpy(oi->delta_base_sha1, base);
1967 } else
1968 hashclr(oi->delta_base_sha1);
1969 }
1970
1971 out:
1972 unuse_pack(&w_curs);
1973 return type;
1974 }
1975
1976 static void *unpack_compressed_entry(struct packed_git *p,
1977 struct pack_window **w_curs,
1978 off_t curpos,
1979 unsigned long size)
1980 {
1981 int st;
1982 git_zstream stream;
1983 unsigned char *buffer, *in;
1984
1985 buffer = xmallocz_gently(size);
1986 if (!buffer)
1987 return NULL;
1988 memset(&stream, 0, sizeof(stream));
1989 stream.next_out = buffer;
1990 stream.avail_out = size + 1;
1991
1992 git_inflate_init(&stream);
1993 do {
1994 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1995 stream.next_in = in;
1996 st = git_inflate(&stream, Z_FINISH);
1997 if (!stream.avail_out)
1998 break; /* the payload is larger than it should be */
1999 curpos += stream.next_in - in;
2000 } while (st == Z_OK || st == Z_BUF_ERROR);
2001 git_inflate_end(&stream);
2002 if ((st != Z_STREAM_END) || stream.total_out != size) {
2003 free(buffer);
2004 return NULL;
2005 }
2006
2007 return buffer;
2008 }
2009
2010 #define MAX_DELTA_CACHE (256)
2011
2012 static size_t delta_base_cached;
2013
2014 static struct delta_base_cache_lru_list {
2015 struct delta_base_cache_lru_list *prev;
2016 struct delta_base_cache_lru_list *next;
2017 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2018
2019 static struct delta_base_cache_entry {
2020 struct delta_base_cache_lru_list lru;
2021 void *data;
2022 struct packed_git *p;
2023 off_t base_offset;
2024 unsigned long size;
2025 enum object_type type;
2026 } delta_base_cache[MAX_DELTA_CACHE];
2027
2028 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2029 {
2030 unsigned long hash;
2031
2032 hash = (unsigned long)p + (unsigned long)base_offset;
2033 hash += (hash >> 8) + (hash >> 16);
2034 return hash % MAX_DELTA_CACHE;
2035 }
2036
2037 static struct delta_base_cache_entry *
2038 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2039 {
2040 unsigned long hash = pack_entry_hash(p, base_offset);
2041 return delta_base_cache + hash;
2042 }
2043
2044 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2045 struct packed_git *p, off_t base_offset)
2046 {
2047 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2048 }
2049
2050 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2051 {
2052 struct delta_base_cache_entry *ent;
2053 ent = get_delta_base_cache_entry(p, base_offset);
2054 return eq_delta_base_cache_entry(ent, p, base_offset);
2055 }
2056
2057 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2058 {
2059 ent->data = NULL;
2060 ent->lru.next->prev = ent->lru.prev;
2061 ent->lru.prev->next = ent->lru.next;
2062 delta_base_cached -= ent->size;
2063 }
2064
2065 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2066 unsigned long *base_size, enum object_type *type, int keep_cache)
2067 {
2068 struct delta_base_cache_entry *ent;
2069 void *ret;
2070
2071 ent = get_delta_base_cache_entry(p, base_offset);
2072
2073 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2074 return unpack_entry(p, base_offset, type, base_size);
2075
2076 ret = ent->data;
2077
2078 if (!keep_cache)
2079 clear_delta_base_cache_entry(ent);
2080 else
2081 ret = xmemdupz(ent->data, ent->size);
2082 *type = ent->type;
2083 *base_size = ent->size;
2084 return ret;
2085 }
2086
2087 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2088 {
2089 if (ent->data) {
2090 free(ent->data);
2091 ent->data = NULL;
2092 ent->lru.next->prev = ent->lru.prev;
2093 ent->lru.prev->next = ent->lru.next;
2094 delta_base_cached -= ent->size;
2095 }
2096 }
2097
2098 void clear_delta_base_cache(void)
2099 {
2100 unsigned long p;
2101 for (p = 0; p < MAX_DELTA_CACHE; p++)
2102 release_delta_base_cache(&delta_base_cache[p]);
2103 }
2104
2105 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2106 void *base, unsigned long base_size, enum object_type type)
2107 {
2108 unsigned long hash = pack_entry_hash(p, base_offset);
2109 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2110 struct delta_base_cache_lru_list *lru;
2111
2112 release_delta_base_cache(ent);
2113 delta_base_cached += base_size;
2114
2115 for (lru = delta_base_cache_lru.next;
2116 delta_base_cached > delta_base_cache_limit
2117 && lru != &delta_base_cache_lru;
2118 lru = lru->next) {
2119 struct delta_base_cache_entry *f = (void *)lru;
2120 if (f->type == OBJ_BLOB)
2121 release_delta_base_cache(f);
2122 }
2123 for (lru = delta_base_cache_lru.next;
2124 delta_base_cached > delta_base_cache_limit
2125 && lru != &delta_base_cache_lru;
2126 lru = lru->next) {
2127 struct delta_base_cache_entry *f = (void *)lru;
2128 release_delta_base_cache(f);
2129 }
2130
2131 ent->p = p;
2132 ent->base_offset = base_offset;
2133 ent->type = type;
2134 ent->data = base;
2135 ent->size = base_size;
2136 ent->lru.next = &delta_base_cache_lru;
2137 ent->lru.prev = delta_base_cache_lru.prev;
2138 delta_base_cache_lru.prev->next = &ent->lru;
2139 delta_base_cache_lru.prev = &ent->lru;
2140 }
2141
2142 static void *read_object(const unsigned char *sha1, enum object_type *type,
2143 unsigned long *size);
2144
2145 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2146 {
2147 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2148 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2149 p->pack_name, (uintmax_t)obj_offset);
2150 }
2151
2152 int do_check_packed_object_crc;
2153
2154 #define UNPACK_ENTRY_STACK_PREALLOC 64
2155 struct unpack_entry_stack_ent {
2156 off_t obj_offset;
2157 off_t curpos;
2158 unsigned long size;
2159 };
2160
2161 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2162 enum object_type *final_type, unsigned long *final_size)
2163 {
2164 struct pack_window *w_curs = NULL;
2165 off_t curpos = obj_offset;
2166 void *data = NULL;
2167 unsigned long size;
2168 enum object_type type;
2169 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2170 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2171 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2172 int base_from_cache = 0;
2173
2174 write_pack_access_log(p, obj_offset);
2175
2176 /* PHASE 1: drill down to the innermost base object */
2177 for (;;) {
2178 off_t base_offset;
2179 int i;
2180 struct delta_base_cache_entry *ent;
2181
2182 ent = get_delta_base_cache_entry(p, curpos);
2183 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2184 type = ent->type;
2185 data = ent->data;
2186 size = ent->size;
2187 clear_delta_base_cache_entry(ent);
2188 base_from_cache = 1;
2189 break;
2190 }
2191
2192 if (do_check_packed_object_crc && p->index_version > 1) {
2193 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2194 unsigned long len = revidx[1].offset - obj_offset;
2195 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2196 const unsigned char *sha1 =
2197 nth_packed_object_sha1(p, revidx->nr);
2198 error("bad packed object CRC for %s",
2199 sha1_to_hex(sha1));
2200 mark_bad_packed_object(p, sha1);
2201 unuse_pack(&w_curs);
2202 return NULL;
2203 }
2204 }
2205
2206 type = unpack_object_header(p, &w_curs, &curpos, &size);
2207 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2208 break;
2209
2210 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2211 if (!base_offset) {
2212 error("failed to validate delta base reference "
2213 "at offset %"PRIuMAX" from %s",
2214 (uintmax_t)curpos, p->pack_name);
2215 /* bail to phase 2, in hopes of recovery */
2216 data = NULL;
2217 break;
2218 }
2219
2220 /* push object, proceed to base */
2221 if (delta_stack_nr >= delta_stack_alloc
2222 && delta_stack == small_delta_stack) {
2223 delta_stack_alloc = alloc_nr(delta_stack_nr);
2224 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2225 memcpy(delta_stack, small_delta_stack,
2226 sizeof(*delta_stack)*delta_stack_nr);
2227 } else {
2228 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2229 }
2230 i = delta_stack_nr++;
2231 delta_stack[i].obj_offset = obj_offset;
2232 delta_stack[i].curpos = curpos;
2233 delta_stack[i].size = size;
2234
2235 curpos = obj_offset = base_offset;
2236 }
2237
2238 /* PHASE 2: handle the base */
2239 switch (type) {
2240 case OBJ_OFS_DELTA:
2241 case OBJ_REF_DELTA:
2242 if (data)
2243 die("BUG in unpack_entry: left loop at a valid delta");
2244 break;
2245 case OBJ_COMMIT:
2246 case OBJ_TREE:
2247 case OBJ_BLOB:
2248 case OBJ_TAG:
2249 if (!base_from_cache)
2250 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2251 break;
2252 default:
2253 data = NULL;
2254 error("unknown object type %i at offset %"PRIuMAX" in %s",
2255 type, (uintmax_t)obj_offset, p->pack_name);
2256 }
2257
2258 /* PHASE 3: apply deltas in order */
2259
2260 /* invariants:
2261 * 'data' holds the base data, or NULL if there was corruption
2262 */
2263 while (delta_stack_nr) {
2264 void *delta_data;
2265 void *base = data;
2266 unsigned long delta_size, base_size = size;
2267 int i;
2268
2269 data = NULL;
2270
2271 if (base)
2272 add_delta_base_cache(p, obj_offset, base, base_size, type);
2273
2274 if (!base) {
2275 /*
2276 * We're probably in deep shit, but let's try to fetch
2277 * the required base anyway from another pack or loose.
2278 * This is costly but should happen only in the presence
2279 * of a corrupted pack, and is better than failing outright.
2280 */
2281 struct revindex_entry *revidx;
2282 const unsigned char *base_sha1;
2283 revidx = find_pack_revindex(p, obj_offset);
2284 if (revidx) {
2285 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2286 error("failed to read delta base object %s"
2287 " at offset %"PRIuMAX" from %s",
2288 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2289 p->pack_name);
2290 mark_bad_packed_object(p, base_sha1);
2291 base = read_object(base_sha1, &type, &base_size);
2292 }
2293 }
2294
2295 i = --delta_stack_nr;
2296 obj_offset = delta_stack[i].obj_offset;
2297 curpos = delta_stack[i].curpos;
2298 delta_size = delta_stack[i].size;
2299
2300 if (!base)
2301 continue;
2302
2303 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2304
2305 if (!delta_data) {
2306 error("failed to unpack compressed delta "
2307 "at offset %"PRIuMAX" from %s",
2308 (uintmax_t)curpos, p->pack_name);
2309 data = NULL;
2310 continue;
2311 }
2312
2313 data = patch_delta(base, base_size,
2314 delta_data, delta_size,
2315 &size);
2316
2317 /*
2318 * We could not apply the delta; warn the user, but keep going.
2319 * Our failure will be noticed either in the next iteration of
2320 * the loop, or if this is the final delta, in the caller when
2321 * we return NULL. Those code paths will take care of making
2322 * a more explicit warning and retrying with another copy of
2323 * the object.
2324 */
2325 if (!data)
2326 error("failed to apply delta");
2327
2328 free(delta_data);
2329 }
2330
2331 *final_type = type;
2332 *final_size = size;
2333
2334 unuse_pack(&w_curs);
2335
2336 if (delta_stack != small_delta_stack)
2337 free(delta_stack);
2338
2339 return data;
2340 }
2341
2342 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2343 uint32_t n)
2344 {
2345 const unsigned char *index = p->index_data;
2346 if (!index) {
2347 if (open_pack_index(p))
2348 return NULL;
2349 index = p->index_data;
2350 }
2351 if (n >= p->num_objects)
2352 return NULL;
2353 index += 4 * 256;
2354 if (p->index_version == 1) {
2355 return index + 24 * n + 4;
2356 } else {
2357 index += 8;
2358 return index + 20 * n;
2359 }
2360 }
2361
2362 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2363 {
2364 const unsigned char *index = p->index_data;
2365 index += 4 * 256;
2366 if (p->index_version == 1) {
2367 return ntohl(*((uint32_t *)(index + 24 * n)));
2368 } else {
2369 uint32_t off;
2370 index += 8 + p->num_objects * (20 + 4);
2371 off = ntohl(*((uint32_t *)(index + 4 * n)));
2372 if (!(off & 0x80000000))
2373 return off;
2374 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2375 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2376 ntohl(*((uint32_t *)(index + 4)));
2377 }
2378 }
2379
2380 off_t find_pack_entry_one(const unsigned char *sha1,
2381 struct packed_git *p)
2382 {
2383 const uint32_t *level1_ofs = p->index_data;
2384 const unsigned char *index = p->index_data;
2385 unsigned hi, lo, stride;
2386 static int use_lookup = -1;
2387 static int debug_lookup = -1;
2388
2389 if (debug_lookup < 0)
2390 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2391
2392 if (!index) {
2393 if (open_pack_index(p))
2394 return 0;
2395 level1_ofs = p->index_data;
2396 index = p->index_data;
2397 }
2398 if (p->index_version > 1) {
2399 level1_ofs += 2;
2400 index += 8;
2401 }
2402 index += 4 * 256;
2403 hi = ntohl(level1_ofs[*sha1]);
2404 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2405 if (p->index_version > 1) {
2406 stride = 20;
2407 } else {
2408 stride = 24;
2409 index += 4;
2410 }
2411
2412 if (debug_lookup)
2413 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2414 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2415
2416 if (use_lookup < 0)
2417 use_lookup = !!getenv("GIT_USE_LOOKUP");
2418 if (use_lookup) {
2419 int pos = sha1_entry_pos(index, stride, 0,
2420 lo, hi, p->num_objects, sha1);
2421 if (pos < 0)
2422 return 0;
2423 return nth_packed_object_offset(p, pos);
2424 }
2425
2426 do {
2427 unsigned mi = (lo + hi) / 2;
2428 int cmp = hashcmp(index + mi * stride, sha1);
2429
2430 if (debug_lookup)
2431 printf("lo %u hi %u rg %u mi %u\n",
2432 lo, hi, hi - lo, mi);
2433 if (!cmp)
2434 return nth_packed_object_offset(p, mi);
2435 if (cmp > 0)
2436 hi = mi;
2437 else
2438 lo = mi+1;
2439 } while (lo < hi);
2440 return 0;
2441 }
2442
2443 int is_pack_valid(struct packed_git *p)
2444 {
2445 /* An already open pack is known to be valid. */
2446 if (p->pack_fd != -1)
2447 return 1;
2448
2449 /* If the pack has one window completely covering the
2450 * file size, the pack is known to be valid even if
2451 * the descriptor is not currently open.
2452 */
2453 if (p->windows) {
2454 struct pack_window *w = p->windows;
2455
2456 if (!w->offset && w->len == p->pack_size)
2457 return 1;
2458 }
2459
2460 /* Force the pack to open to prove its valid. */
2461 return !open_packed_git(p);
2462 }
2463
2464 static int fill_pack_entry(const unsigned char *sha1,
2465 struct pack_entry *e,
2466 struct packed_git *p)
2467 {
2468 off_t offset;
2469
2470 if (p->num_bad_objects) {
2471 unsigned i;
2472 for (i = 0; i < p->num_bad_objects; i++)
2473 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2474 return 0;
2475 }
2476
2477 offset = find_pack_entry_one(sha1, p);
2478 if (!offset)
2479 return 0;
2480
2481 /*
2482 * We are about to tell the caller where they can locate the
2483 * requested object. We better make sure the packfile is
2484 * still here and can be accessed before supplying that
2485 * answer, as it may have been deleted since the index was
2486 * loaded!
2487 */
2488 if (!is_pack_valid(p))
2489 return 0;
2490 e->offset = offset;
2491 e->p = p;
2492 hashcpy(e->sha1, sha1);
2493 return 1;
2494 }
2495
2496 /*
2497 * Iff a pack file contains the object named by sha1, return true and
2498 * store its location to e.
2499 */
2500 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2501 {
2502 struct packed_git *p;
2503
2504 prepare_packed_git();
2505 if (!packed_git)
2506 return 0;
2507
2508 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2509 return 1;
2510
2511 for (p = packed_git; p; p = p->next) {
2512 if (p == last_found_pack)
2513 continue; /* we already checked this one */
2514
2515 if (fill_pack_entry(sha1, e, p)) {
2516 last_found_pack = p;
2517 return 1;
2518 }
2519 }
2520 return 0;
2521 }
2522
2523 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2524 struct packed_git *packs)
2525 {
2526 struct packed_git *p;
2527
2528 for (p = packs; p; p = p->next) {
2529 if (find_pack_entry_one(sha1, p))
2530 return p;
2531 }
2532 return NULL;
2533
2534 }
2535
2536 static int sha1_loose_object_info(const unsigned char *sha1,
2537 struct object_info *oi)
2538 {
2539 int status;
2540 unsigned long mapsize, size;
2541 void *map;
2542 git_zstream stream;
2543 char hdr[32];
2544
2545 if (oi->delta_base_sha1)
2546 hashclr(oi->delta_base_sha1);
2547
2548 /*
2549 * If we don't care about type or size, then we don't
2550 * need to look inside the object at all. Note that we
2551 * do not optimize out the stat call, even if the
2552 * caller doesn't care about the disk-size, since our
2553 * return value implicitly indicates whether the
2554 * object even exists.
2555 */
2556 if (!oi->typep && !oi->sizep) {
2557 struct stat st;
2558 if (stat_sha1_file(sha1, &st) < 0)
2559 return -1;
2560 if (oi->disk_sizep)
2561 *oi->disk_sizep = st.st_size;
2562 return 0;
2563 }
2564
2565 map = map_sha1_file(sha1, &mapsize);
2566 if (!map)
2567 return -1;
2568 if (oi->disk_sizep)
2569 *oi->disk_sizep = mapsize;
2570 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2571 status = error("unable to unpack %s header",
2572 sha1_to_hex(sha1));
2573 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2574 status = error("unable to parse %s header", sha1_to_hex(sha1));
2575 else if (oi->sizep)
2576 *oi->sizep = size;
2577 git_inflate_end(&stream);
2578 munmap(map, mapsize);
2579 if (oi->typep)
2580 *oi->typep = status;
2581 return 0;
2582 }
2583
2584 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2585 {
2586 struct cached_object *co;
2587 struct pack_entry e;
2588 int rtype;
2589 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2590
2591 co = find_cached_object(real);
2592 if (co) {
2593 if (oi->typep)
2594 *(oi->typep) = co->type;
2595 if (oi->sizep)
2596 *(oi->sizep) = co->size;
2597 if (oi->disk_sizep)
2598 *(oi->disk_sizep) = 0;
2599 if (oi->delta_base_sha1)
2600 hashclr(oi->delta_base_sha1);
2601 oi->whence = OI_CACHED;
2602 return 0;
2603 }
2604
2605 if (!find_pack_entry(real, &e)) {
2606 /* Most likely it's a loose object. */
2607 if (!sha1_loose_object_info(real, oi)) {
2608 oi->whence = OI_LOOSE;
2609 return 0;
2610 }
2611
2612 /* Not a loose object; someone else may have just packed it. */
2613 reprepare_packed_git();
2614 if (!find_pack_entry(real, &e))
2615 return -1;
2616 }
2617
2618 rtype = packed_object_info(e.p, e.offset, oi);
2619 if (rtype < 0) {
2620 mark_bad_packed_object(e.p, real);
2621 return sha1_object_info_extended(real, oi, 0);
2622 } else if (in_delta_base_cache(e.p, e.offset)) {
2623 oi->whence = OI_DBCACHED;
2624 } else {
2625 oi->whence = OI_PACKED;
2626 oi->u.packed.offset = e.offset;
2627 oi->u.packed.pack = e.p;
2628 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2629 rtype == OBJ_OFS_DELTA);
2630 }
2631
2632 return 0;
2633 }
2634
2635 /* returns enum object_type or negative */
2636 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2637 {
2638 enum object_type type;
2639 struct object_info oi = {NULL};
2640
2641 oi.typep = &type;
2642 oi.sizep = sizep;
2643 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2644 return -1;
2645 return type;
2646 }
2647
2648 static void *read_packed_sha1(const unsigned char *sha1,
2649 enum object_type *type, unsigned long *size)
2650 {
2651 struct pack_entry e;
2652 void *data;
2653
2654 if (!find_pack_entry(sha1, &e))
2655 return NULL;
2656 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2657 if (!data) {
2658 /*
2659 * We're probably in deep shit, but let's try to fetch
2660 * the required object anyway from another pack or loose.
2661 * This should happen only in the presence of a corrupted
2662 * pack, and is better than failing outright.
2663 */
2664 error("failed to read object %s at offset %"PRIuMAX" from %s",
2665 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2666 mark_bad_packed_object(e.p, sha1);
2667 data = read_object(sha1, type, size);
2668 }
2669 return data;
2670 }
2671
2672 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2673 unsigned char *sha1)
2674 {
2675 struct cached_object *co;
2676
2677 hash_sha1_file(buf, len, typename(type), sha1);
2678 if (has_sha1_file(sha1) || find_cached_object(sha1))
2679 return 0;
2680 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2681 co = &cached_objects[cached_object_nr++];
2682 co->size = len;
2683 co->type = type;
2684 co->buf = xmalloc(len);
2685 memcpy(co->buf, buf, len);
2686 hashcpy(co->sha1, sha1);
2687 return 0;
2688 }
2689
2690 static void *read_object(const unsigned char *sha1, enum object_type *type,
2691 unsigned long *size)
2692 {
2693 unsigned long mapsize;
2694 void *map, *buf;
2695 struct cached_object *co;
2696
2697 co = find_cached_object(sha1);
2698 if (co) {
2699 *type = co->type;
2700 *size = co->size;
2701 return xmemdupz(co->buf, co->size);
2702 }
2703
2704 buf = read_packed_sha1(sha1, type, size);
2705 if (buf)
2706 return buf;
2707 map = map_sha1_file(sha1, &mapsize);
2708 if (map) {
2709 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2710 munmap(map, mapsize);
2711 return buf;
2712 }
2713 reprepare_packed_git();
2714 return read_packed_sha1(sha1, type, size);
2715 }
2716
2717 /*
2718 * This function dies on corrupt objects; the callers who want to
2719 * deal with them should arrange to call read_object() and give error
2720 * messages themselves.
2721 */
2722 void *read_sha1_file_extended(const unsigned char *sha1,
2723 enum object_type *type,
2724 unsigned long *size,
2725 unsigned flag)
2726 {
2727 void *data;
2728 const struct packed_git *p;
2729 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2730
2731 errno = 0;
2732 data = read_object(repl, type, size);
2733 if (data)
2734 return data;
2735
2736 if (errno && errno != ENOENT)
2737 die_errno("failed to read object %s", sha1_to_hex(sha1));
2738
2739 /* die if we replaced an object with one that does not exist */
2740 if (repl != sha1)
2741 die("replacement %s not found for %s",
2742 sha1_to_hex(repl), sha1_to_hex(sha1));
2743
2744 if (has_loose_object(repl)) {
2745 const char *path = sha1_file_name(sha1);
2746
2747 die("loose object %s (stored in %s) is corrupt",
2748 sha1_to_hex(repl), path);
2749 }
2750
2751 if ((p = has_packed_and_bad(repl)) != NULL)
2752 die("packed object %s (stored in %s) is corrupt",
2753 sha1_to_hex(repl), p->pack_name);
2754
2755 return NULL;
2756 }
2757
2758 void *read_object_with_reference(const unsigned char *sha1,
2759 const char *required_type_name,
2760 unsigned long *size,
2761 unsigned char *actual_sha1_return)
2762 {
2763 enum object_type type, required_type;
2764 void *buffer;
2765 unsigned long isize;
2766 unsigned char actual_sha1[20];
2767
2768 required_type = type_from_string(required_type_name);
2769 hashcpy(actual_sha1, sha1);
2770 while (1) {
2771 int ref_length = -1;
2772 const char *ref_type = NULL;
2773
2774 buffer = read_sha1_file(actual_sha1, &type, &isize);
2775 if (!buffer)
2776 return NULL;
2777 if (type == required_type) {
2778 *size = isize;
2779 if (actual_sha1_return)
2780 hashcpy(actual_sha1_return, actual_sha1);
2781 return buffer;
2782 }
2783 /* Handle references */
2784 else if (type == OBJ_COMMIT)
2785 ref_type = "tree ";
2786 else if (type == OBJ_TAG)
2787 ref_type = "object ";
2788 else {
2789 free(buffer);
2790 return NULL;
2791 }
2792 ref_length = strlen(ref_type);
2793
2794 if (ref_length + 40 > isize ||
2795 memcmp(buffer, ref_type, ref_length) ||
2796 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2797 free(buffer);
2798 return NULL;
2799 }
2800 free(buffer);
2801 /* Now we have the ID of the referred-to object in
2802 * actual_sha1. Check again. */
2803 }
2804 }
2805
2806 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2807 const char *type, unsigned char *sha1,
2808 char *hdr, int *hdrlen)
2809 {
2810 git_SHA_CTX c;
2811
2812 /* Generate the header */
2813 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2814
2815 /* Sha1.. */
2816 git_SHA1_Init(&c);
2817 git_SHA1_Update(&c, hdr, *hdrlen);
2818 git_SHA1_Update(&c, buf, len);
2819 git_SHA1_Final(sha1, &c);
2820 }
2821
2822 /*
2823 * Move the just written object into its final resting place.
2824 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2825 * "moving" is only a part of what it does, when no patch between
2826 * master to pu changes the call sites of this function.
2827 */
2828 int move_temp_to_file(const char *tmpfile, const char *filename)
2829 {
2830 int ret = 0;
2831
2832 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2833 goto try_rename;
2834 else if (link(tmpfile, filename))
2835 ret = errno;
2836
2837 /*
2838 * Coda hack - coda doesn't like cross-directory links,
2839 * so we fall back to a rename, which will mean that it
2840 * won't be able to check collisions, but that's not a
2841 * big deal.
2842 *
2843 * The same holds for FAT formatted media.
2844 *
2845 * When this succeeds, we just return. We have nothing
2846 * left to unlink.
2847 */
2848 if (ret && ret != EEXIST) {
2849 try_rename:
2850 if (!rename(tmpfile, filename))
2851 goto out;
2852 ret = errno;
2853 }
2854 unlink_or_warn(tmpfile);
2855 if (ret) {
2856 if (ret != EEXIST) {
2857 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2858 }
2859 /* FIXME!!! Collision check here ? */
2860 }
2861
2862 out:
2863 if (adjust_shared_perm(filename))
2864 return error("unable to set permission to '%s'", filename);
2865 return 0;
2866 }
2867
2868 static int write_buffer(int fd, const void *buf, size_t len)
2869 {
2870 if (write_in_full(fd, buf, len) < 0)
2871 return error("file write error (%s)", strerror(errno));
2872 return 0;
2873 }
2874
2875 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2876 unsigned char *sha1)
2877 {
2878 char hdr[32];
2879 int hdrlen;
2880 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2881 return 0;
2882 }
2883
2884 /* Finalize a file on disk, and close it. */
2885 static void close_sha1_file(int fd)
2886 {
2887 if (fsync_object_files)
2888 fsync_or_die(fd, "sha1 file");
2889 if (close(fd) != 0)
2890 die_errno("error when closing sha1 file");
2891 }
2892
2893 /* Size of directory component, including the ending '/' */
2894 static inline int directory_size(const char *filename)
2895 {
2896 const char *s = strrchr(filename, '/');
2897 if (!s)
2898 return 0;
2899 return s - filename + 1;
2900 }
2901
2902 /*
2903 * This creates a temporary file in the same directory as the final
2904 * 'filename'
2905 *
2906 * We want to avoid cross-directory filename renames, because those
2907 * can have problems on various filesystems (FAT, NFS, Coda).
2908 */
2909 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2910 {
2911 int fd, dirlen = directory_size(filename);
2912
2913 if (dirlen + 20 > bufsiz) {
2914 errno = ENAMETOOLONG;
2915 return -1;
2916 }
2917 memcpy(buffer, filename, dirlen);
2918 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2919 fd = git_mkstemp_mode(buffer, 0444);
2920 if (fd < 0 && dirlen && errno == ENOENT) {
2921 /* Make sure the directory exists */
2922 memcpy(buffer, filename, dirlen);
2923 buffer[dirlen-1] = 0;
2924 if (mkdir(buffer, 0777) && errno != EEXIST)
2925 return -1;
2926 if (adjust_shared_perm(buffer))
2927 return -1;
2928
2929 /* Try again */
2930 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2931 fd = git_mkstemp_mode(buffer, 0444);
2932 }
2933 return fd;
2934 }
2935
2936 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2937 const void *buf, unsigned long len, time_t mtime)
2938 {
2939 int fd, ret;
2940 unsigned char compressed[4096];
2941 git_zstream stream;
2942 git_SHA_CTX c;
2943 unsigned char parano_sha1[20];
2944 static char tmp_file[PATH_MAX];
2945 const char *filename = sha1_file_name(sha1);
2946
2947 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2948 if (fd < 0) {
2949 if (errno == EACCES)
2950 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2951 else
2952 return error("unable to create temporary file: %s", strerror(errno));
2953 }
2954
2955 /* Set it up */
2956 git_deflate_init(&stream, zlib_compression_level);
2957 stream.next_out = compressed;
2958 stream.avail_out = sizeof(compressed);
2959 git_SHA1_Init(&c);
2960
2961 /* First header.. */
2962 stream.next_in = (unsigned char *)hdr;
2963 stream.avail_in = hdrlen;
2964 while (git_deflate(&stream, 0) == Z_OK)
2965 ; /* nothing */
2966 git_SHA1_Update(&c, hdr, hdrlen);
2967
2968 /* Then the data itself.. */
2969 stream.next_in = (void *)buf;
2970 stream.avail_in = len;
2971 do {
2972 unsigned char *in0 = stream.next_in;
2973 ret = git_deflate(&stream, Z_FINISH);
2974 git_SHA1_Update(&c, in0, stream.next_in - in0);
2975 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2976 die("unable to write sha1 file");
2977 stream.next_out = compressed;
2978 stream.avail_out = sizeof(compressed);
2979 } while (ret == Z_OK);
2980
2981 if (ret != Z_STREAM_END)
2982 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2983 ret = git_deflate_end_gently(&stream);
2984 if (ret != Z_OK)
2985 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2986 git_SHA1_Final(parano_sha1, &c);
2987 if (hashcmp(sha1, parano_sha1) != 0)
2988 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2989
2990 close_sha1_file(fd);
2991
2992 if (mtime) {
2993 struct utimbuf utb;
2994 utb.actime = mtime;
2995 utb.modtime = mtime;
2996 if (utime(tmp_file, &utb) < 0)
2997 warning("failed utime() on %s: %s",
2998 tmp_file, strerror(errno));
2999 }
3000
3001 return move_temp_to_file(tmp_file, filename);
3002 }
3003
3004 static int freshen_loose_object(const unsigned char *sha1)
3005 {
3006 return check_and_freshen(sha1, 1);
3007 }
3008
3009 static int freshen_packed_object(const unsigned char *sha1)
3010 {
3011 struct pack_entry e;
3012 if (!find_pack_entry(sha1, &e))
3013 return 0;
3014 if (e.p->freshened)
3015 return 1;
3016 if (!freshen_file(e.p->pack_name))
3017 return 0;
3018 e.p->freshened = 1;
3019 return 1;
3020 }
3021
3022 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3023 {
3024 char hdr[32];
3025 int hdrlen;
3026
3027 /* Normally if we have it in the pack then we do not bother writing
3028 * it out into .git/objects/??/?{38} file.
3029 */
3030 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3031 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3032 return 0;
3033 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3034 }
3035
3036 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3037 unsigned char *sha1, unsigned flags)
3038 {
3039 char *header;
3040 int hdrlen, status = 0;
3041
3042 /* type string, SP, %lu of the length plus NUL must fit this */
3043 header = xmalloc(strlen(type) + 32);
3044 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3045
3046 if (!(flags & HASH_WRITE_OBJECT))
3047 goto cleanup;
3048 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3049 goto cleanup;
3050 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3051
3052 cleanup:
3053 free(header);
3054 return status;
3055 }
3056
3057 int force_object_loose(const unsigned char *sha1, time_t mtime)
3058 {
3059 void *buf;
3060 unsigned long len;
3061 enum object_type type;
3062 char hdr[32];
3063 int hdrlen;
3064 int ret;
3065
3066 if (has_loose_object(sha1))
3067 return 0;
3068 buf = read_packed_sha1(sha1, &type, &len);
3069 if (!buf)
3070 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3071 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3072 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3073 free(buf);
3074
3075 return ret;
3076 }
3077
3078 int has_pack_index(const unsigned char *sha1)
3079 {
3080 struct stat st;
3081 if (stat(sha1_pack_index_name(sha1), &st))
3082 return 0;
3083 return 1;
3084 }
3085
3086 int has_sha1_pack(const unsigned char *sha1)
3087 {
3088 struct pack_entry e;
3089 return find_pack_entry(sha1, &e);
3090 }
3091
3092 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3093 {
3094 struct pack_entry e;
3095
3096 if (find_pack_entry(sha1, &e))
3097 return 1;
3098 if (has_loose_object(sha1))
3099 return 1;
3100 if (flags & HAS_SHA1_QUICK)
3101 return 0;
3102 reprepare_packed_git();
3103 return find_pack_entry(sha1, &e);
3104 }
3105
3106 static void check_tree(const void *buf, size_t size)
3107 {
3108 struct tree_desc desc;
3109 struct name_entry entry;
3110
3111 init_tree_desc(&desc, buf, size);
3112 while (tree_entry(&desc, &entry))
3113 /* do nothing
3114 * tree_entry() will die() on malformed entries */
3115 ;
3116 }
3117
3118 static void check_commit(const void *buf, size_t size)
3119 {
3120 struct commit c;
3121 memset(&c, 0, sizeof(c));
3122 if (parse_commit_buffer(&c, buf, size))
3123 die("corrupt commit");
3124 }
3125
3126 static void check_tag(const void *buf, size_t size)
3127 {
3128 struct tag t;
3129 memset(&t, 0, sizeof(t));
3130 if (parse_tag_buffer(&t, buf, size))
3131 die("corrupt tag");
3132 }
3133
3134 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3135 enum object_type type,
3136 const char *path, unsigned flags)
3137 {
3138 int ret, re_allocated = 0;
3139 int write_object = flags & HASH_WRITE_OBJECT;
3140
3141 if (!type)
3142 type = OBJ_BLOB;
3143
3144 /*
3145 * Convert blobs to git internal format
3146 */
3147 if ((type == OBJ_BLOB) && path) {
3148 struct strbuf nbuf = STRBUF_INIT;
3149 if (convert_to_git(path, buf, size, &nbuf,
3150 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3151 buf = strbuf_detach(&nbuf, &size);
3152 re_allocated = 1;
3153 }
3154 }
3155 if (flags & HASH_FORMAT_CHECK) {
3156 if (type == OBJ_TREE)
3157 check_tree(buf, size);
3158 if (type == OBJ_COMMIT)
3159 check_commit(buf, size);
3160 if (type == OBJ_TAG)
3161 check_tag(buf, size);
3162 }
3163
3164 if (write_object)
3165 ret = write_sha1_file(buf, size, typename(type), sha1);
3166 else
3167 ret = hash_sha1_file(buf, size, typename(type), sha1);
3168 if (re_allocated)
3169 free(buf);
3170 return ret;
3171 }
3172
3173 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3174 const char *path, unsigned flags)
3175 {
3176 int ret;
3177 const int write_object = flags & HASH_WRITE_OBJECT;
3178 struct strbuf sbuf = STRBUF_INIT;
3179
3180 assert(path);
3181 assert(would_convert_to_git_filter_fd(path));
3182
3183 convert_to_git_filter_fd(path, fd, &sbuf,
3184 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3185
3186 if (write_object)
3187 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3188 sha1);
3189 else
3190 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3191 sha1);
3192 strbuf_release(&sbuf);
3193 return ret;
3194 }
3195
3196 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3197 const char *path, unsigned flags)
3198 {
3199 struct strbuf sbuf = STRBUF_INIT;
3200 int ret;
3201
3202 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3203 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3204 else
3205 ret = -1;
3206 strbuf_release(&sbuf);
3207 return ret;
3208 }
3209
3210 #define SMALL_FILE_SIZE (32*1024)
3211
3212 static int index_core(unsigned char *sha1, int fd, size_t size,
3213 enum object_type type, const char *path,
3214 unsigned flags)
3215 {
3216 int ret;
3217
3218 if (!size) {
3219 ret = index_mem(sha1, "", size, type, path, flags);
3220 } else if (size <= SMALL_FILE_SIZE) {
3221 char *buf = xmalloc(size);
3222 if (size == read_in_full(fd, buf, size))
3223 ret = index_mem(sha1, buf, size, type, path, flags);
3224 else
3225 ret = error("short read %s", strerror(errno));
3226 free(buf);
3227 } else {
3228 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3229 ret = index_mem(sha1, buf, size, type, path, flags);
3230 munmap(buf, size);
3231 }
3232 return ret;
3233 }
3234
3235 /*
3236 * This creates one packfile per large blob unless bulk-checkin
3237 * machinery is "plugged".
3238 *
3239 * This also bypasses the usual "convert-to-git" dance, and that is on
3240 * purpose. We could write a streaming version of the converting
3241 * functions and insert that before feeding the data to fast-import
3242 * (or equivalent in-core API described above). However, that is
3243 * somewhat complicated, as we do not know the size of the filter
3244 * result, which we need to know beforehand when writing a git object.
3245 * Since the primary motivation for trying to stream from the working
3246 * tree file and to avoid mmaping it in core is to deal with large
3247 * binary blobs, they generally do not want to get any conversion, and
3248 * callers should avoid this code path when filters are requested.
3249 */
3250 static int index_stream(unsigned char *sha1, int fd, size_t size,
3251 enum object_type type, const char *path,
3252 unsigned flags)
3253 {
3254 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3255 }
3256
3257 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3258 enum object_type type, const char *path, unsigned flags)
3259 {
3260 int ret;
3261
3262 /*
3263 * Call xsize_t() only when needed to avoid potentially unnecessary
3264 * die() for large files.
3265 */
3266 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3267 ret = index_stream_convert_blob(sha1, fd, path, flags);
3268 else if (!S_ISREG(st->st_mode))
3269 ret = index_pipe(sha1, fd, type, path, flags);
3270 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3271 (path && would_convert_to_git(path)))
3272 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3273 flags);
3274 else
3275 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3276 flags);
3277 close(fd);
3278 return ret;
3279 }
3280
3281 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3282 {
3283 int fd;
3284 struct strbuf sb = STRBUF_INIT;
3285
3286 switch (st->st_mode & S_IFMT) {
3287 case S_IFREG:
3288 fd = open(path, O_RDONLY);
3289 if (fd < 0)
3290 return error("open(\"%s\"): %s", path,
3291 strerror(errno));
3292 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3293 return error("%s: failed to insert into database",
3294 path);
3295 break;
3296 case S_IFLNK:
3297 if (strbuf_readlink(&sb, path, st->st_size)) {
3298 char *errstr = strerror(errno);
3299 return error("readlink(\"%s\"): %s", path,
3300 errstr);
3301 }
3302 if (!(flags & HASH_WRITE_OBJECT))
3303 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3304 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3305 return error("%s: failed to insert into database",
3306 path);
3307 strbuf_release(&sb);
3308 break;
3309 case S_IFDIR:
3310 return resolve_gitlink_ref(path, "HEAD", sha1);
3311 default:
3312 return error("%s: unsupported file type", path);
3313 }
3314 return 0;
3315 }
3316
3317 int read_pack_header(int fd, struct pack_header *header)
3318 {
3319 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3320 /* "eof before pack header was fully read" */
3321 return PH_ERROR_EOF;
3322
3323 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3324 /* "protocol error (pack signature mismatch detected)" */
3325 return PH_ERROR_PACK_SIGNATURE;
3326 if (!pack_version_ok(header->hdr_version))
3327 /* "protocol error (pack version unsupported)" */
3328 return PH_ERROR_PROTOCOL;
3329 return 0;
3330 }
3331
3332 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3333 {
3334 enum object_type type = sha1_object_info(sha1, NULL);
3335 if (type < 0)
3336 die("%s is not a valid object", sha1_to_hex(sha1));
3337 if (type != expect)
3338 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3339 typename(expect));
3340 }
3341
3342 static int for_each_file_in_obj_subdir(int subdir_nr,
3343 struct strbuf *path,
3344 each_loose_object_fn obj_cb,
3345 each_loose_cruft_fn cruft_cb,
3346 each_loose_subdir_fn subdir_cb,
3347 void *data)
3348 {
3349 size_t baselen = path->len;
3350 DIR *dir = opendir(path->buf);
3351 struct dirent *de;
3352 int r = 0;
3353
3354 if (!dir) {
3355 if (errno == ENOENT)
3356 return 0;
3357 return error("unable to open %s: %s", path->buf, strerror(errno));
3358 }
3359
3360 while ((de = readdir(dir))) {
3361 if (is_dot_or_dotdot(de->d_name))
3362 continue;
3363
3364 strbuf_setlen(path, baselen);
3365 strbuf_addf(path, "/%s", de->d_name);
3366
3367 if (strlen(de->d_name) == 38) {
3368 char hex[41];
3369 unsigned char sha1[20];
3370
3371 snprintf(hex, sizeof(hex), "%02x%s",
3372 subdir_nr, de->d_name);
3373 if (!get_sha1_hex(hex, sha1)) {
3374 if (obj_cb) {
3375 r = obj_cb(sha1, path->buf, data);
3376 if (r)
3377 break;
3378 }
3379 continue;
3380 }
3381 }
3382
3383 if (cruft_cb) {
3384 r = cruft_cb(de->d_name, path->buf, data);
3385 if (r)
3386 break;
3387 }
3388 }
3389 strbuf_setlen(path, baselen);
3390
3391 if (!r && subdir_cb)
3392 r = subdir_cb(subdir_nr, path->buf, data);
3393
3394 closedir(dir);
3395 return r;
3396 }
3397
3398 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3399 each_loose_object_fn obj_cb,
3400 each_loose_cruft_fn cruft_cb,
3401 each_loose_subdir_fn subdir_cb,
3402 void *data)
3403 {
3404 size_t baselen = path->len;
3405 int r = 0;
3406 int i;
3407
3408 for (i = 0; i < 256; i++) {
3409 strbuf_addf(path, "/%02x", i);
3410 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3411 subdir_cb, data);
3412 strbuf_setlen(path, baselen);
3413 if (r)
3414 break;
3415 }
3416
3417 return r;
3418 }
3419
3420 int for_each_loose_file_in_objdir(const char *path,
3421 each_loose_object_fn obj_cb,
3422 each_loose_cruft_fn cruft_cb,
3423 each_loose_subdir_fn subdir_cb,
3424 void *data)
3425 {
3426 struct strbuf buf = STRBUF_INIT;
3427 int r;
3428
3429 strbuf_addstr(&buf, path);
3430 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3431 subdir_cb, data);
3432 strbuf_release(&buf);
3433
3434 return r;
3435 }
3436
3437 struct loose_alt_odb_data {
3438 each_loose_object_fn *cb;
3439 void *data;
3440 };
3441
3442 static int loose_from_alt_odb(struct alternate_object_database *alt,
3443 void *vdata)
3444 {
3445 struct loose_alt_odb_data *data = vdata;
3446 struct strbuf buf = STRBUF_INIT;
3447 int r;
3448
3449 /* copy base not including trailing '/' */
3450 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3451 r = for_each_loose_file_in_objdir_buf(&buf,
3452 data->cb, NULL, NULL,
3453 data->data);
3454 strbuf_release(&buf);
3455 return r;
3456 }
3457
3458 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3459 {
3460 struct loose_alt_odb_data alt;
3461 int r;
3462
3463 r = for_each_loose_file_in_objdir(get_object_directory(),
3464 cb, NULL, NULL, data);
3465 if (r)
3466 return r;
3467
3468 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3469 return 0;
3470
3471 alt.cb = cb;
3472 alt.data = data;
3473 return foreach_alt_odb(loose_from_alt_odb, &alt);
3474 }
3475
3476 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3477 {
3478 uint32_t i;
3479 int r = 0;
3480
3481 for (i = 0; i < p->num_objects; i++) {
3482 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3483
3484 if (!sha1)
3485 return error("unable to get sha1 of object %u in %s",
3486 i, p->pack_name);
3487
3488 r = cb(sha1, p, i, data);
3489 if (r)
3490 break;
3491 }
3492 return r;
3493 }
3494
3495 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3496 {
3497 struct packed_git *p;
3498 int r = 0;
3499
3500 prepare_packed_git();
3501 for (p = packed_git; p; p = p->next) {
3502 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3503 continue;
3504 r = for_each_object_in_pack(p, cb, data);
3505 if (r)
3506 break;
3507 }
3508 return r;
3509 }