Merge branch 'jk/index-pack-reduce-recheck' into maint
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
34
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38 const unsigned char null_sha1[20];
39
40 /*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
53
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59 };
60
61 /*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67 static struct packed_git *last_found_pack;
68
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 {
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81 }
82
83 int mkdir_in_gitdir(const char *path)
84 {
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110 }
111
112 enum scld_error safe_create_leading_directories(char *path)
113 {
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162 }
163
164 enum scld_error safe_create_leading_directories_const(const char *path)
165 {
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171 }
172
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 {
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183 }
184
185 const char *sha1_file_name(const unsigned char *sha1)
186 {
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203 }
204
205 /*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212 {
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235 }
236
237 char *sha1_pack_name(const unsigned char *sha1)
238 {
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242 }
243
244 char *sha1_pack_index_name(const unsigned char *sha1)
245 {
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249 }
250
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
253
254 /*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271 {
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338 }
339
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342 {
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373 }
374
375 void read_info_alternates(const char * relative_base, int depth)
376 {
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 const char alt_file_name[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path[PATH_MAX + 1 + sizeof alt_file_name];
385 int fd;
386
387 sprintf(path, "%s/%s", relative_base, alt_file_name);
388 fd = git_open_noatime(path);
389 if (fd < 0)
390 return;
391 if (fstat(fd, &st) || (st.st_size == 0)) {
392 close(fd);
393 return;
394 }
395 mapsz = xsize_t(st.st_size);
396 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
397 close(fd);
398
399 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
400
401 munmap(map, mapsz);
402 }
403
404 void add_to_alternates_file(const char *reference)
405 {
406 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
407 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
408 char *alt = mkpath("%s\n", reference);
409 write_or_die(fd, alt, strlen(alt));
410 if (commit_lock_file(lock))
411 die("could not close alternates file");
412 if (alt_odb_tail)
413 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
414 }
415
416 int foreach_alt_odb(alt_odb_fn fn, void *cb)
417 {
418 struct alternate_object_database *ent;
419 int r = 0;
420
421 prepare_alt_odb();
422 for (ent = alt_odb_list; ent; ent = ent->next) {
423 r = fn(ent, cb);
424 if (r)
425 break;
426 }
427 return r;
428 }
429
430 void prepare_alt_odb(void)
431 {
432 const char *alt;
433
434 if (alt_odb_tail)
435 return;
436
437 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
438 if (!alt) alt = "";
439
440 alt_odb_tail = &alt_odb_list;
441 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
442
443 read_info_alternates(get_object_directory(), 0);
444 }
445
446 static int freshen_file(const char *fn)
447 {
448 struct utimbuf t;
449 t.actime = t.modtime = time(NULL);
450 return !utime(fn, &t);
451 }
452
453 static int check_and_freshen_file(const char *fn, int freshen)
454 {
455 if (access(fn, F_OK))
456 return 0;
457 if (freshen && freshen_file(fn))
458 return 0;
459 return 1;
460 }
461
462 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
463 {
464 return check_and_freshen_file(sha1_file_name(sha1), freshen);
465 }
466
467 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
468 {
469 struct alternate_object_database *alt;
470 prepare_alt_odb();
471 for (alt = alt_odb_list; alt; alt = alt->next) {
472 fill_sha1_path(alt->name, sha1);
473 if (check_and_freshen_file(alt->base, freshen))
474 return 1;
475 }
476 return 0;
477 }
478
479 static int check_and_freshen(const unsigned char *sha1, int freshen)
480 {
481 return check_and_freshen_local(sha1, freshen) ||
482 check_and_freshen_nonlocal(sha1, freshen);
483 }
484
485 int has_loose_object_nonlocal(const unsigned char *sha1)
486 {
487 return check_and_freshen_nonlocal(sha1, 0);
488 }
489
490 static int has_loose_object(const unsigned char *sha1)
491 {
492 return check_and_freshen(sha1, 0);
493 }
494
495 static unsigned int pack_used_ctr;
496 static unsigned int pack_mmap_calls;
497 static unsigned int peak_pack_open_windows;
498 static unsigned int pack_open_windows;
499 static unsigned int pack_open_fds;
500 static unsigned int pack_max_fds;
501 static size_t peak_pack_mapped;
502 static size_t pack_mapped;
503 struct packed_git *packed_git;
504
505 void pack_report(void)
506 {
507 fprintf(stderr,
508 "pack_report: getpagesize() = %10" SZ_FMT "\n"
509 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
510 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
511 sz_fmt(getpagesize()),
512 sz_fmt(packed_git_window_size),
513 sz_fmt(packed_git_limit));
514 fprintf(stderr,
515 "pack_report: pack_used_ctr = %10u\n"
516 "pack_report: pack_mmap_calls = %10u\n"
517 "pack_report: pack_open_windows = %10u / %10u\n"
518 "pack_report: pack_mapped = "
519 "%10" SZ_FMT " / %10" SZ_FMT "\n",
520 pack_used_ctr,
521 pack_mmap_calls,
522 pack_open_windows, peak_pack_open_windows,
523 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
524 }
525
526 /*
527 * Open and mmap the index file at path, perform a couple of
528 * consistency checks, then record its information to p. Return 0 on
529 * success.
530 */
531 static int check_packed_git_idx(const char *path, struct packed_git *p)
532 {
533 void *idx_map;
534 struct pack_idx_header *hdr;
535 size_t idx_size;
536 uint32_t version, nr, i, *index;
537 int fd = git_open_noatime(path);
538 struct stat st;
539
540 if (fd < 0)
541 return -1;
542 if (fstat(fd, &st)) {
543 close(fd);
544 return -1;
545 }
546 idx_size = xsize_t(st.st_size);
547 if (idx_size < 4 * 256 + 20 + 20) {
548 close(fd);
549 return error("index file %s is too small", path);
550 }
551 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
552 close(fd);
553
554 hdr = idx_map;
555 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
556 version = ntohl(hdr->idx_version);
557 if (version < 2 || version > 2) {
558 munmap(idx_map, idx_size);
559 return error("index file %s is version %"PRIu32
560 " and is not supported by this binary"
561 " (try upgrading GIT to a newer version)",
562 path, version);
563 }
564 } else
565 version = 1;
566
567 nr = 0;
568 index = idx_map;
569 if (version > 1)
570 index += 2; /* skip index header */
571 for (i = 0; i < 256; i++) {
572 uint32_t n = ntohl(index[i]);
573 if (n < nr) {
574 munmap(idx_map, idx_size);
575 return error("non-monotonic index %s", path);
576 }
577 nr = n;
578 }
579
580 if (version == 1) {
581 /*
582 * Total size:
583 * - 256 index entries 4 bytes each
584 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
585 * - 20-byte SHA1 of the packfile
586 * - 20-byte SHA1 file checksum
587 */
588 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
589 munmap(idx_map, idx_size);
590 return error("wrong index v1 file size in %s", path);
591 }
592 } else if (version == 2) {
593 /*
594 * Minimum size:
595 * - 8 bytes of header
596 * - 256 index entries 4 bytes each
597 * - 20-byte sha1 entry * nr
598 * - 4-byte crc entry * nr
599 * - 4-byte offset entry * nr
600 * - 20-byte SHA1 of the packfile
601 * - 20-byte SHA1 file checksum
602 * And after the 4-byte offset table might be a
603 * variable sized table containing 8-byte entries
604 * for offsets larger than 2^31.
605 */
606 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
607 unsigned long max_size = min_size;
608 if (nr)
609 max_size += (nr - 1)*8;
610 if (idx_size < min_size || idx_size > max_size) {
611 munmap(idx_map, idx_size);
612 return error("wrong index v2 file size in %s", path);
613 }
614 if (idx_size != min_size &&
615 /*
616 * make sure we can deal with large pack offsets.
617 * 31-bit signed offset won't be enough, neither
618 * 32-bit unsigned one will be.
619 */
620 (sizeof(off_t) <= 4)) {
621 munmap(idx_map, idx_size);
622 return error("pack too large for current definition of off_t in %s", path);
623 }
624 }
625
626 p->index_version = version;
627 p->index_data = idx_map;
628 p->index_size = idx_size;
629 p->num_objects = nr;
630 return 0;
631 }
632
633 int open_pack_index(struct packed_git *p)
634 {
635 char *idx_name;
636 int ret;
637
638 if (p->index_data)
639 return 0;
640
641 idx_name = xstrdup(p->pack_name);
642 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
643 ret = check_packed_git_idx(idx_name, p);
644 free(idx_name);
645 return ret;
646 }
647
648 static void scan_windows(struct packed_git *p,
649 struct packed_git **lru_p,
650 struct pack_window **lru_w,
651 struct pack_window **lru_l)
652 {
653 struct pack_window *w, *w_l;
654
655 for (w_l = NULL, w = p->windows; w; w = w->next) {
656 if (!w->inuse_cnt) {
657 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
658 *lru_p = p;
659 *lru_w = w;
660 *lru_l = w_l;
661 }
662 }
663 w_l = w;
664 }
665 }
666
667 static int unuse_one_window(struct packed_git *current)
668 {
669 struct packed_git *p, *lru_p = NULL;
670 struct pack_window *lru_w = NULL, *lru_l = NULL;
671
672 if (current)
673 scan_windows(current, &lru_p, &lru_w, &lru_l);
674 for (p = packed_git; p; p = p->next)
675 scan_windows(p, &lru_p, &lru_w, &lru_l);
676 if (lru_p) {
677 munmap(lru_w->base, lru_w->len);
678 pack_mapped -= lru_w->len;
679 if (lru_l)
680 lru_l->next = lru_w->next;
681 else
682 lru_p->windows = lru_w->next;
683 free(lru_w);
684 pack_open_windows--;
685 return 1;
686 }
687 return 0;
688 }
689
690 void release_pack_memory(size_t need)
691 {
692 size_t cur = pack_mapped;
693 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
694 ; /* nothing */
695 }
696
697 static void mmap_limit_check(size_t length)
698 {
699 static size_t limit = 0;
700 if (!limit) {
701 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
702 if (!limit)
703 limit = SIZE_MAX;
704 }
705 if (length > limit)
706 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
707 (uintmax_t)length, (uintmax_t)limit);
708 }
709
710 void *xmmap_gently(void *start, size_t length,
711 int prot, int flags, int fd, off_t offset)
712 {
713 void *ret;
714
715 mmap_limit_check(length);
716 ret = mmap(start, length, prot, flags, fd, offset);
717 if (ret == MAP_FAILED) {
718 if (!length)
719 return NULL;
720 release_pack_memory(length);
721 ret = mmap(start, length, prot, flags, fd, offset);
722 }
723 return ret;
724 }
725
726 void *xmmap(void *start, size_t length,
727 int prot, int flags, int fd, off_t offset)
728 {
729 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
730 if (ret == MAP_FAILED)
731 die_errno("mmap failed");
732 return ret;
733 }
734
735 void close_pack_windows(struct packed_git *p)
736 {
737 while (p->windows) {
738 struct pack_window *w = p->windows;
739
740 if (w->inuse_cnt)
741 die("pack '%s' still has open windows to it",
742 p->pack_name);
743 munmap(w->base, w->len);
744 pack_mapped -= w->len;
745 pack_open_windows--;
746 p->windows = w->next;
747 free(w);
748 }
749 }
750
751 /*
752 * The LRU pack is the one with the oldest MRU window, preferring packs
753 * with no used windows, or the oldest mtime if it has no windows allocated.
754 */
755 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
756 {
757 struct pack_window *w, *this_mru_w;
758 int has_windows_inuse = 0;
759
760 /*
761 * Reject this pack if it has windows and the previously selected
762 * one does not. If this pack does not have windows, reject
763 * it if the pack file is newer than the previously selected one.
764 */
765 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
766 return;
767
768 for (w = this_mru_w = p->windows; w; w = w->next) {
769 /*
770 * Reject this pack if any of its windows are in use,
771 * but the previously selected pack did not have any
772 * inuse windows. Otherwise, record that this pack
773 * has windows in use.
774 */
775 if (w->inuse_cnt) {
776 if (*accept_windows_inuse)
777 has_windows_inuse = 1;
778 else
779 return;
780 }
781
782 if (w->last_used > this_mru_w->last_used)
783 this_mru_w = w;
784
785 /*
786 * Reject this pack if it has windows that have been
787 * used more recently than the previously selected pack.
788 * If the previously selected pack had windows inuse and
789 * we have not encountered a window in this pack that is
790 * inuse, skip this check since we prefer a pack with no
791 * inuse windows to one that has inuse windows.
792 */
793 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
794 this_mru_w->last_used > (*mru_w)->last_used)
795 return;
796 }
797
798 /*
799 * Select this pack.
800 */
801 *mru_w = this_mru_w;
802 *lru_p = p;
803 *accept_windows_inuse = has_windows_inuse;
804 }
805
806 static int close_one_pack(void)
807 {
808 struct packed_git *p, *lru_p = NULL;
809 struct pack_window *mru_w = NULL;
810 int accept_windows_inuse = 1;
811
812 for (p = packed_git; p; p = p->next) {
813 if (p->pack_fd == -1)
814 continue;
815 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
816 }
817
818 if (lru_p) {
819 close(lru_p->pack_fd);
820 pack_open_fds--;
821 lru_p->pack_fd = -1;
822 return 1;
823 }
824
825 return 0;
826 }
827
828 void unuse_pack(struct pack_window **w_cursor)
829 {
830 struct pack_window *w = *w_cursor;
831 if (w) {
832 w->inuse_cnt--;
833 *w_cursor = NULL;
834 }
835 }
836
837 void close_pack_index(struct packed_git *p)
838 {
839 if (p->index_data) {
840 munmap((void *)p->index_data, p->index_size);
841 p->index_data = NULL;
842 }
843 }
844
845 /*
846 * This is used by git-repack in case a newly created pack happens to
847 * contain the same set of objects as an existing one. In that case
848 * the resulting file might be different even if its name would be the
849 * same. It is best to close any reference to the old pack before it is
850 * replaced on disk. Of course no index pointers or windows for given pack
851 * must subsist at this point. If ever objects from this pack are requested
852 * again, the new version of the pack will be reinitialized through
853 * reprepare_packed_git().
854 */
855 void free_pack_by_name(const char *pack_name)
856 {
857 struct packed_git *p, **pp = &packed_git;
858
859 while (*pp) {
860 p = *pp;
861 if (strcmp(pack_name, p->pack_name) == 0) {
862 clear_delta_base_cache();
863 close_pack_windows(p);
864 if (p->pack_fd != -1) {
865 close(p->pack_fd);
866 pack_open_fds--;
867 }
868 close_pack_index(p);
869 free(p->bad_object_sha1);
870 *pp = p->next;
871 if (last_found_pack == p)
872 last_found_pack = NULL;
873 free(p);
874 return;
875 }
876 pp = &p->next;
877 }
878 }
879
880 static unsigned int get_max_fd_limit(void)
881 {
882 #ifdef RLIMIT_NOFILE
883 {
884 struct rlimit lim;
885
886 if (!getrlimit(RLIMIT_NOFILE, &lim))
887 return lim.rlim_cur;
888 }
889 #endif
890
891 #ifdef _SC_OPEN_MAX
892 {
893 long open_max = sysconf(_SC_OPEN_MAX);
894 if (0 < open_max)
895 return open_max;
896 /*
897 * Otherwise, we got -1 for one of the two
898 * reasons:
899 *
900 * (1) sysconf() did not understand _SC_OPEN_MAX
901 * and signaled an error with -1; or
902 * (2) sysconf() said there is no limit.
903 *
904 * We _could_ clear errno before calling sysconf() to
905 * tell these two cases apart and return a huge number
906 * in the latter case to let the caller cap it to a
907 * value that is not so selfish, but letting the
908 * fallback OPEN_MAX codepath take care of these cases
909 * is a lot simpler.
910 */
911 }
912 #endif
913
914 #ifdef OPEN_MAX
915 return OPEN_MAX;
916 #else
917 return 1; /* see the caller ;-) */
918 #endif
919 }
920
921 /*
922 * Do not call this directly as this leaks p->pack_fd on error return;
923 * call open_packed_git() instead.
924 */
925 static int open_packed_git_1(struct packed_git *p)
926 {
927 struct stat st;
928 struct pack_header hdr;
929 unsigned char sha1[20];
930 unsigned char *idx_sha1;
931 long fd_flag;
932
933 if (!p->index_data && open_pack_index(p))
934 return error("packfile %s index unavailable", p->pack_name);
935
936 if (!pack_max_fds) {
937 unsigned int max_fds = get_max_fd_limit();
938
939 /* Save 3 for stdin/stdout/stderr, 22 for work */
940 if (25 < max_fds)
941 pack_max_fds = max_fds - 25;
942 else
943 pack_max_fds = 1;
944 }
945
946 while (pack_max_fds <= pack_open_fds && close_one_pack())
947 ; /* nothing */
948
949 p->pack_fd = git_open_noatime(p->pack_name);
950 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
951 return -1;
952 pack_open_fds++;
953
954 /* If we created the struct before we had the pack we lack size. */
955 if (!p->pack_size) {
956 if (!S_ISREG(st.st_mode))
957 return error("packfile %s not a regular file", p->pack_name);
958 p->pack_size = st.st_size;
959 } else if (p->pack_size != st.st_size)
960 return error("packfile %s size changed", p->pack_name);
961
962 /* We leave these file descriptors open with sliding mmap;
963 * there is no point keeping them open across exec(), though.
964 */
965 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
966 if (fd_flag < 0)
967 return error("cannot determine file descriptor flags");
968 fd_flag |= FD_CLOEXEC;
969 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
970 return error("cannot set FD_CLOEXEC");
971
972 /* Verify we recognize this pack file format. */
973 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
974 return error("file %s is far too short to be a packfile", p->pack_name);
975 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
976 return error("file %s is not a GIT packfile", p->pack_name);
977 if (!pack_version_ok(hdr.hdr_version))
978 return error("packfile %s is version %"PRIu32" and not"
979 " supported (try upgrading GIT to a newer version)",
980 p->pack_name, ntohl(hdr.hdr_version));
981
982 /* Verify the pack matches its index. */
983 if (p->num_objects != ntohl(hdr.hdr_entries))
984 return error("packfile %s claims to have %"PRIu32" objects"
985 " while index indicates %"PRIu32" objects",
986 p->pack_name, ntohl(hdr.hdr_entries),
987 p->num_objects);
988 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
989 return error("end of packfile %s is unavailable", p->pack_name);
990 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
991 return error("packfile %s signature is unavailable", p->pack_name);
992 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
993 if (hashcmp(sha1, idx_sha1))
994 return error("packfile %s does not match index", p->pack_name);
995 return 0;
996 }
997
998 static int open_packed_git(struct packed_git *p)
999 {
1000 if (!open_packed_git_1(p))
1001 return 0;
1002 if (p->pack_fd != -1) {
1003 close(p->pack_fd);
1004 pack_open_fds--;
1005 p->pack_fd = -1;
1006 }
1007 return -1;
1008 }
1009
1010 static int in_window(struct pack_window *win, off_t offset)
1011 {
1012 /* We must promise at least 20 bytes (one hash) after the
1013 * offset is available from this window, otherwise the offset
1014 * is not actually in this window and a different window (which
1015 * has that one hash excess) must be used. This is to support
1016 * the object header and delta base parsing routines below.
1017 */
1018 off_t win_off = win->offset;
1019 return win_off <= offset
1020 && (offset + 20) <= (win_off + win->len);
1021 }
1022
1023 unsigned char *use_pack(struct packed_git *p,
1024 struct pack_window **w_cursor,
1025 off_t offset,
1026 unsigned long *left)
1027 {
1028 struct pack_window *win = *w_cursor;
1029
1030 /* Since packfiles end in a hash of their content and it's
1031 * pointless to ask for an offset into the middle of that
1032 * hash, and the in_window function above wouldn't match
1033 * don't allow an offset too close to the end of the file.
1034 */
1035 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1036 die("packfile %s cannot be accessed", p->pack_name);
1037 if (offset > (p->pack_size - 20))
1038 die("offset beyond end of packfile (truncated pack?)");
1039
1040 if (!win || !in_window(win, offset)) {
1041 if (win)
1042 win->inuse_cnt--;
1043 for (win = p->windows; win; win = win->next) {
1044 if (in_window(win, offset))
1045 break;
1046 }
1047 if (!win) {
1048 size_t window_align = packed_git_window_size / 2;
1049 off_t len;
1050
1051 if (p->pack_fd == -1 && open_packed_git(p))
1052 die("packfile %s cannot be accessed", p->pack_name);
1053
1054 win = xcalloc(1, sizeof(*win));
1055 win->offset = (offset / window_align) * window_align;
1056 len = p->pack_size - win->offset;
1057 if (len > packed_git_window_size)
1058 len = packed_git_window_size;
1059 win->len = (size_t)len;
1060 pack_mapped += win->len;
1061 while (packed_git_limit < pack_mapped
1062 && unuse_one_window(p))
1063 ; /* nothing */
1064 win->base = xmmap(NULL, win->len,
1065 PROT_READ, MAP_PRIVATE,
1066 p->pack_fd, win->offset);
1067 if (win->base == MAP_FAILED)
1068 die("packfile %s cannot be mapped: %s",
1069 p->pack_name,
1070 strerror(errno));
1071 if (!win->offset && win->len == p->pack_size
1072 && !p->do_not_close) {
1073 close(p->pack_fd);
1074 pack_open_fds--;
1075 p->pack_fd = -1;
1076 }
1077 pack_mmap_calls++;
1078 pack_open_windows++;
1079 if (pack_mapped > peak_pack_mapped)
1080 peak_pack_mapped = pack_mapped;
1081 if (pack_open_windows > peak_pack_open_windows)
1082 peak_pack_open_windows = pack_open_windows;
1083 win->next = p->windows;
1084 p->windows = win;
1085 }
1086 }
1087 if (win != *w_cursor) {
1088 win->last_used = pack_used_ctr++;
1089 win->inuse_cnt++;
1090 *w_cursor = win;
1091 }
1092 offset -= win->offset;
1093 if (left)
1094 *left = win->len - xsize_t(offset);
1095 return win->base + offset;
1096 }
1097
1098 static struct packed_git *alloc_packed_git(int extra)
1099 {
1100 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1101 memset(p, 0, sizeof(*p));
1102 p->pack_fd = -1;
1103 return p;
1104 }
1105
1106 static void try_to_free_pack_memory(size_t size)
1107 {
1108 release_pack_memory(size);
1109 }
1110
1111 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1112 {
1113 static int have_set_try_to_free_routine;
1114 struct stat st;
1115 struct packed_git *p = alloc_packed_git(path_len + 2);
1116
1117 if (!have_set_try_to_free_routine) {
1118 have_set_try_to_free_routine = 1;
1119 set_try_to_free_routine(try_to_free_pack_memory);
1120 }
1121
1122 /*
1123 * Make sure a corresponding .pack file exists and that
1124 * the index looks sane.
1125 */
1126 path_len -= strlen(".idx");
1127 if (path_len < 1) {
1128 free(p);
1129 return NULL;
1130 }
1131 memcpy(p->pack_name, path, path_len);
1132
1133 strcpy(p->pack_name + path_len, ".keep");
1134 if (!access(p->pack_name, F_OK))
1135 p->pack_keep = 1;
1136
1137 strcpy(p->pack_name + path_len, ".pack");
1138 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1139 free(p);
1140 return NULL;
1141 }
1142
1143 /* ok, it looks sane as far as we can check without
1144 * actually mapping the pack file.
1145 */
1146 p->pack_size = st.st_size;
1147 p->pack_local = local;
1148 p->mtime = st.st_mtime;
1149 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1150 hashclr(p->sha1);
1151 return p;
1152 }
1153
1154 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1155 {
1156 const char *path = sha1_pack_name(sha1);
1157 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1158
1159 strcpy(p->pack_name, path);
1160 hashcpy(p->sha1, sha1);
1161 if (check_packed_git_idx(idx_path, p)) {
1162 free(p);
1163 return NULL;
1164 }
1165
1166 return p;
1167 }
1168
1169 void install_packed_git(struct packed_git *pack)
1170 {
1171 if (pack->pack_fd != -1)
1172 pack_open_fds++;
1173
1174 pack->next = packed_git;
1175 packed_git = pack;
1176 }
1177
1178 void (*report_garbage)(const char *desc, const char *path);
1179
1180 static void report_helper(const struct string_list *list,
1181 int seen_bits, int first, int last)
1182 {
1183 const char *msg;
1184 switch (seen_bits) {
1185 case 0:
1186 msg = "no corresponding .idx or .pack";
1187 break;
1188 case 1:
1189 msg = "no corresponding .idx";
1190 break;
1191 case 2:
1192 msg = "no corresponding .pack";
1193 break;
1194 default:
1195 return;
1196 }
1197 for (; first < last; first++)
1198 report_garbage(msg, list->items[first].string);
1199 }
1200
1201 static void report_pack_garbage(struct string_list *list)
1202 {
1203 int i, baselen = -1, first = 0, seen_bits = 0;
1204
1205 if (!report_garbage)
1206 return;
1207
1208 string_list_sort(list);
1209
1210 for (i = 0; i < list->nr; i++) {
1211 const char *path = list->items[i].string;
1212 if (baselen != -1 &&
1213 strncmp(path, list->items[first].string, baselen)) {
1214 report_helper(list, seen_bits, first, i);
1215 baselen = -1;
1216 seen_bits = 0;
1217 }
1218 if (baselen == -1) {
1219 const char *dot = strrchr(path, '.');
1220 if (!dot) {
1221 report_garbage("garbage found", path);
1222 continue;
1223 }
1224 baselen = dot - path + 1;
1225 first = i;
1226 }
1227 if (!strcmp(path + baselen, "pack"))
1228 seen_bits |= 1;
1229 else if (!strcmp(path + baselen, "idx"))
1230 seen_bits |= 2;
1231 }
1232 report_helper(list, seen_bits, first, list->nr);
1233 }
1234
1235 static void prepare_packed_git_one(char *objdir, int local)
1236 {
1237 struct strbuf path = STRBUF_INIT;
1238 size_t dirnamelen;
1239 DIR *dir;
1240 struct dirent *de;
1241 struct string_list garbage = STRING_LIST_INIT_DUP;
1242
1243 strbuf_addstr(&path, objdir);
1244 strbuf_addstr(&path, "/pack");
1245 dir = opendir(path.buf);
1246 if (!dir) {
1247 if (errno != ENOENT)
1248 error("unable to open object pack directory: %s: %s",
1249 path.buf, strerror(errno));
1250 strbuf_release(&path);
1251 return;
1252 }
1253 strbuf_addch(&path, '/');
1254 dirnamelen = path.len;
1255 while ((de = readdir(dir)) != NULL) {
1256 struct packed_git *p;
1257 size_t base_len;
1258
1259 if (is_dot_or_dotdot(de->d_name))
1260 continue;
1261
1262 strbuf_setlen(&path, dirnamelen);
1263 strbuf_addstr(&path, de->d_name);
1264
1265 base_len = path.len;
1266 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1267 /* Don't reopen a pack we already have. */
1268 for (p = packed_git; p; p = p->next) {
1269 size_t len;
1270 if (strip_suffix(p->pack_name, ".pack", &len) &&
1271 len == base_len &&
1272 !memcmp(p->pack_name, path.buf, len))
1273 break;
1274 }
1275 if (p == NULL &&
1276 /*
1277 * See if it really is a valid .idx file with
1278 * corresponding .pack file that we can map.
1279 */
1280 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1281 install_packed_git(p);
1282 }
1283
1284 if (!report_garbage)
1285 continue;
1286
1287 if (ends_with(de->d_name, ".idx") ||
1288 ends_with(de->d_name, ".pack") ||
1289 ends_with(de->d_name, ".bitmap") ||
1290 ends_with(de->d_name, ".keep"))
1291 string_list_append(&garbage, path.buf);
1292 else
1293 report_garbage("garbage found", path.buf);
1294 }
1295 closedir(dir);
1296 report_pack_garbage(&garbage);
1297 string_list_clear(&garbage, 0);
1298 strbuf_release(&path);
1299 }
1300
1301 static int sort_pack(const void *a_, const void *b_)
1302 {
1303 struct packed_git *a = *((struct packed_git **)a_);
1304 struct packed_git *b = *((struct packed_git **)b_);
1305 int st;
1306
1307 /*
1308 * Local packs tend to contain objects specific to our
1309 * variant of the project than remote ones. In addition,
1310 * remote ones could be on a network mounted filesystem.
1311 * Favor local ones for these reasons.
1312 */
1313 st = a->pack_local - b->pack_local;
1314 if (st)
1315 return -st;
1316
1317 /*
1318 * Younger packs tend to contain more recent objects,
1319 * and more recent objects tend to get accessed more
1320 * often.
1321 */
1322 if (a->mtime < b->mtime)
1323 return 1;
1324 else if (a->mtime == b->mtime)
1325 return 0;
1326 return -1;
1327 }
1328
1329 static void rearrange_packed_git(void)
1330 {
1331 struct packed_git **ary, *p;
1332 int i, n;
1333
1334 for (n = 0, p = packed_git; p; p = p->next)
1335 n++;
1336 if (n < 2)
1337 return;
1338
1339 /* prepare an array of packed_git for easier sorting */
1340 ary = xcalloc(n, sizeof(struct packed_git *));
1341 for (n = 0, p = packed_git; p; p = p->next)
1342 ary[n++] = p;
1343
1344 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1345
1346 /* link them back again */
1347 for (i = 0; i < n - 1; i++)
1348 ary[i]->next = ary[i + 1];
1349 ary[n - 1]->next = NULL;
1350 packed_git = ary[0];
1351
1352 free(ary);
1353 }
1354
1355 static int prepare_packed_git_run_once = 0;
1356 void prepare_packed_git(void)
1357 {
1358 struct alternate_object_database *alt;
1359
1360 if (prepare_packed_git_run_once)
1361 return;
1362 prepare_packed_git_one(get_object_directory(), 1);
1363 prepare_alt_odb();
1364 for (alt = alt_odb_list; alt; alt = alt->next) {
1365 alt->name[-1] = 0;
1366 prepare_packed_git_one(alt->base, 0);
1367 alt->name[-1] = '/';
1368 }
1369 rearrange_packed_git();
1370 prepare_packed_git_run_once = 1;
1371 }
1372
1373 void reprepare_packed_git(void)
1374 {
1375 prepare_packed_git_run_once = 0;
1376 prepare_packed_git();
1377 }
1378
1379 static void mark_bad_packed_object(struct packed_git *p,
1380 const unsigned char *sha1)
1381 {
1382 unsigned i;
1383 for (i = 0; i < p->num_bad_objects; i++)
1384 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1385 return;
1386 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1387 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1388 p->num_bad_objects++;
1389 }
1390
1391 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1392 {
1393 struct packed_git *p;
1394 unsigned i;
1395
1396 for (p = packed_git; p; p = p->next)
1397 for (i = 0; i < p->num_bad_objects; i++)
1398 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1399 return p;
1400 return NULL;
1401 }
1402
1403 /*
1404 * With an in-core object data in "map", rehash it to make sure the
1405 * object name actually matches "sha1" to detect object corruption.
1406 * With "map" == NULL, try reading the object named with "sha1" using
1407 * the streaming interface and rehash it to do the same.
1408 */
1409 int check_sha1_signature(const unsigned char *sha1, void *map,
1410 unsigned long size, const char *type)
1411 {
1412 unsigned char real_sha1[20];
1413 enum object_type obj_type;
1414 struct git_istream *st;
1415 git_SHA_CTX c;
1416 char hdr[32];
1417 int hdrlen;
1418
1419 if (map) {
1420 hash_sha1_file(map, size, type, real_sha1);
1421 return hashcmp(sha1, real_sha1) ? -1 : 0;
1422 }
1423
1424 st = open_istream(sha1, &obj_type, &size, NULL);
1425 if (!st)
1426 return -1;
1427
1428 /* Generate the header */
1429 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1430
1431 /* Sha1.. */
1432 git_SHA1_Init(&c);
1433 git_SHA1_Update(&c, hdr, hdrlen);
1434 for (;;) {
1435 char buf[1024 * 16];
1436 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1437
1438 if (readlen < 0) {
1439 close_istream(st);
1440 return -1;
1441 }
1442 if (!readlen)
1443 break;
1444 git_SHA1_Update(&c, buf, readlen);
1445 }
1446 git_SHA1_Final(real_sha1, &c);
1447 close_istream(st);
1448 return hashcmp(sha1, real_sha1) ? -1 : 0;
1449 }
1450
1451 int git_open_noatime(const char *name)
1452 {
1453 static int sha1_file_open_flag = O_NOATIME;
1454
1455 for (;;) {
1456 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1457 if (fd >= 0)
1458 return fd;
1459
1460 /* Might the failure be due to O_NOATIME? */
1461 if (errno != ENOENT && sha1_file_open_flag) {
1462 sha1_file_open_flag = 0;
1463 continue;
1464 }
1465
1466 return -1;
1467 }
1468 }
1469
1470 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1471 {
1472 struct alternate_object_database *alt;
1473
1474 if (!lstat(sha1_file_name(sha1), st))
1475 return 0;
1476
1477 prepare_alt_odb();
1478 errno = ENOENT;
1479 for (alt = alt_odb_list; alt; alt = alt->next) {
1480 fill_sha1_path(alt->name, sha1);
1481 if (!lstat(alt->base, st))
1482 return 0;
1483 }
1484
1485 return -1;
1486 }
1487
1488 static int open_sha1_file(const unsigned char *sha1)
1489 {
1490 int fd;
1491 struct alternate_object_database *alt;
1492 int most_interesting_errno;
1493
1494 fd = git_open_noatime(sha1_file_name(sha1));
1495 if (fd >= 0)
1496 return fd;
1497 most_interesting_errno = errno;
1498
1499 prepare_alt_odb();
1500 for (alt = alt_odb_list; alt; alt = alt->next) {
1501 fill_sha1_path(alt->name, sha1);
1502 fd = git_open_noatime(alt->base);
1503 if (fd >= 0)
1504 return fd;
1505 if (most_interesting_errno == ENOENT)
1506 most_interesting_errno = errno;
1507 }
1508 errno = most_interesting_errno;
1509 return -1;
1510 }
1511
1512 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1513 {
1514 void *map;
1515 int fd;
1516
1517 fd = open_sha1_file(sha1);
1518 map = NULL;
1519 if (fd >= 0) {
1520 struct stat st;
1521
1522 if (!fstat(fd, &st)) {
1523 *size = xsize_t(st.st_size);
1524 if (!*size) {
1525 /* mmap() is forbidden on empty files */
1526 error("object file %s is empty", sha1_file_name(sha1));
1527 return NULL;
1528 }
1529 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1530 }
1531 close(fd);
1532 }
1533 return map;
1534 }
1535
1536 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1537 unsigned long len, enum object_type *type, unsigned long *sizep)
1538 {
1539 unsigned shift;
1540 unsigned long size, c;
1541 unsigned long used = 0;
1542
1543 c = buf[used++];
1544 *type = (c >> 4) & 7;
1545 size = c & 15;
1546 shift = 4;
1547 while (c & 0x80) {
1548 if (len <= used || bitsizeof(long) <= shift) {
1549 error("bad object header");
1550 size = used = 0;
1551 break;
1552 }
1553 c = buf[used++];
1554 size += (c & 0x7f) << shift;
1555 shift += 7;
1556 }
1557 *sizep = size;
1558 return used;
1559 }
1560
1561 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1562 {
1563 /* Get the data stream */
1564 memset(stream, 0, sizeof(*stream));
1565 stream->next_in = map;
1566 stream->avail_in = mapsize;
1567 stream->next_out = buffer;
1568 stream->avail_out = bufsiz;
1569
1570 git_inflate_init(stream);
1571 return git_inflate(stream, 0);
1572 }
1573
1574 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1575 {
1576 int bytes = strlen(buffer) + 1;
1577 unsigned char *buf = xmallocz(size);
1578 unsigned long n;
1579 int status = Z_OK;
1580
1581 n = stream->total_out - bytes;
1582 if (n > size)
1583 n = size;
1584 memcpy(buf, (char *) buffer + bytes, n);
1585 bytes = n;
1586 if (bytes <= size) {
1587 /*
1588 * The above condition must be (bytes <= size), not
1589 * (bytes < size). In other words, even though we
1590 * expect no more output and set avail_out to zero,
1591 * the input zlib stream may have bytes that express
1592 * "this concludes the stream", and we *do* want to
1593 * eat that input.
1594 *
1595 * Otherwise we would not be able to test that we
1596 * consumed all the input to reach the expected size;
1597 * we also want to check that zlib tells us that all
1598 * went well with status == Z_STREAM_END at the end.
1599 */
1600 stream->next_out = buf + bytes;
1601 stream->avail_out = size - bytes;
1602 while (status == Z_OK)
1603 status = git_inflate(stream, Z_FINISH);
1604 }
1605 if (status == Z_STREAM_END && !stream->avail_in) {
1606 git_inflate_end(stream);
1607 return buf;
1608 }
1609
1610 if (status < 0)
1611 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1612 else if (stream->avail_in)
1613 error("garbage at end of loose object '%s'",
1614 sha1_to_hex(sha1));
1615 free(buf);
1616 return NULL;
1617 }
1618
1619 /*
1620 * We used to just use "sscanf()", but that's actually way
1621 * too permissive for what we want to check. So do an anal
1622 * object header parse by hand.
1623 */
1624 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1625 {
1626 char type[10];
1627 int i;
1628 unsigned long size;
1629
1630 /*
1631 * The type can be at most ten bytes (including the
1632 * terminating '\0' that we add), and is followed by
1633 * a space.
1634 */
1635 i = 0;
1636 for (;;) {
1637 char c = *hdr++;
1638 if (c == ' ')
1639 break;
1640 type[i++] = c;
1641 if (i >= sizeof(type))
1642 return -1;
1643 }
1644 type[i] = 0;
1645
1646 /*
1647 * The length must follow immediately, and be in canonical
1648 * decimal format (ie "010" is not valid).
1649 */
1650 size = *hdr++ - '0';
1651 if (size > 9)
1652 return -1;
1653 if (size) {
1654 for (;;) {
1655 unsigned long c = *hdr - '0';
1656 if (c > 9)
1657 break;
1658 hdr++;
1659 size = size * 10 + c;
1660 }
1661 }
1662 *sizep = size;
1663
1664 /*
1665 * The length must be followed by a zero byte
1666 */
1667 return *hdr ? -1 : type_from_string(type);
1668 }
1669
1670 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1671 {
1672 int ret;
1673 git_zstream stream;
1674 char hdr[8192];
1675
1676 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1677 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1678 return NULL;
1679
1680 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1681 }
1682
1683 unsigned long get_size_from_delta(struct packed_git *p,
1684 struct pack_window **w_curs,
1685 off_t curpos)
1686 {
1687 const unsigned char *data;
1688 unsigned char delta_head[20], *in;
1689 git_zstream stream;
1690 int st;
1691
1692 memset(&stream, 0, sizeof(stream));
1693 stream.next_out = delta_head;
1694 stream.avail_out = sizeof(delta_head);
1695
1696 git_inflate_init(&stream);
1697 do {
1698 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1699 stream.next_in = in;
1700 st = git_inflate(&stream, Z_FINISH);
1701 curpos += stream.next_in - in;
1702 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1703 stream.total_out < sizeof(delta_head));
1704 git_inflate_end(&stream);
1705 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1706 error("delta data unpack-initial failed");
1707 return 0;
1708 }
1709
1710 /* Examine the initial part of the delta to figure out
1711 * the result size.
1712 */
1713 data = delta_head;
1714
1715 /* ignore base size */
1716 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1717
1718 /* Read the result size */
1719 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1720 }
1721
1722 static off_t get_delta_base(struct packed_git *p,
1723 struct pack_window **w_curs,
1724 off_t *curpos,
1725 enum object_type type,
1726 off_t delta_obj_offset)
1727 {
1728 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1729 off_t base_offset;
1730
1731 /* use_pack() assured us we have [base_info, base_info + 20)
1732 * as a range that we can look at without walking off the
1733 * end of the mapped window. Its actually the hash size
1734 * that is assured. An OFS_DELTA longer than the hash size
1735 * is stupid, as then a REF_DELTA would be smaller to store.
1736 */
1737 if (type == OBJ_OFS_DELTA) {
1738 unsigned used = 0;
1739 unsigned char c = base_info[used++];
1740 base_offset = c & 127;
1741 while (c & 128) {
1742 base_offset += 1;
1743 if (!base_offset || MSB(base_offset, 7))
1744 return 0; /* overflow */
1745 c = base_info[used++];
1746 base_offset = (base_offset << 7) + (c & 127);
1747 }
1748 base_offset = delta_obj_offset - base_offset;
1749 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1750 return 0; /* out of bound */
1751 *curpos += used;
1752 } else if (type == OBJ_REF_DELTA) {
1753 /* The base entry _must_ be in the same pack */
1754 base_offset = find_pack_entry_one(base_info, p);
1755 *curpos += 20;
1756 } else
1757 die("I am totally screwed");
1758 return base_offset;
1759 }
1760
1761 /*
1762 * Like get_delta_base above, but we return the sha1 instead of the pack
1763 * offset. This means it is cheaper for REF deltas (we do not have to do
1764 * the final object lookup), but more expensive for OFS deltas (we
1765 * have to load the revidx to convert the offset back into a sha1).
1766 */
1767 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1768 struct pack_window **w_curs,
1769 off_t curpos,
1770 enum object_type type,
1771 off_t delta_obj_offset)
1772 {
1773 if (type == OBJ_REF_DELTA) {
1774 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1775 return base;
1776 } else if (type == OBJ_OFS_DELTA) {
1777 struct revindex_entry *revidx;
1778 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1779 type, delta_obj_offset);
1780
1781 if (!base_offset)
1782 return NULL;
1783
1784 revidx = find_pack_revindex(p, base_offset);
1785 if (!revidx)
1786 return NULL;
1787
1788 return nth_packed_object_sha1(p, revidx->nr);
1789 } else
1790 return NULL;
1791 }
1792
1793 int unpack_object_header(struct packed_git *p,
1794 struct pack_window **w_curs,
1795 off_t *curpos,
1796 unsigned long *sizep)
1797 {
1798 unsigned char *base;
1799 unsigned long left;
1800 unsigned long used;
1801 enum object_type type;
1802
1803 /* use_pack() assures us we have [base, base + 20) available
1804 * as a range that we can look at. (Its actually the hash
1805 * size that is assured.) With our object header encoding
1806 * the maximum deflated object size is 2^137, which is just
1807 * insane, so we know won't exceed what we have been given.
1808 */
1809 base = use_pack(p, w_curs, *curpos, &left);
1810 used = unpack_object_header_buffer(base, left, &type, sizep);
1811 if (!used) {
1812 type = OBJ_BAD;
1813 } else
1814 *curpos += used;
1815
1816 return type;
1817 }
1818
1819 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1820 {
1821 int type;
1822 struct revindex_entry *revidx;
1823 const unsigned char *sha1;
1824 revidx = find_pack_revindex(p, obj_offset);
1825 if (!revidx)
1826 return OBJ_BAD;
1827 sha1 = nth_packed_object_sha1(p, revidx->nr);
1828 mark_bad_packed_object(p, sha1);
1829 type = sha1_object_info(sha1, NULL);
1830 if (type <= OBJ_NONE)
1831 return OBJ_BAD;
1832 return type;
1833 }
1834
1835 #define POI_STACK_PREALLOC 64
1836
1837 static enum object_type packed_to_object_type(struct packed_git *p,
1838 off_t obj_offset,
1839 enum object_type type,
1840 struct pack_window **w_curs,
1841 off_t curpos)
1842 {
1843 off_t small_poi_stack[POI_STACK_PREALLOC];
1844 off_t *poi_stack = small_poi_stack;
1845 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1846
1847 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1848 off_t base_offset;
1849 unsigned long size;
1850 /* Push the object we're going to leave behind */
1851 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1852 poi_stack_alloc = alloc_nr(poi_stack_nr);
1853 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1854 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1855 } else {
1856 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1857 }
1858 poi_stack[poi_stack_nr++] = obj_offset;
1859 /* If parsing the base offset fails, just unwind */
1860 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1861 if (!base_offset)
1862 goto unwind;
1863 curpos = obj_offset = base_offset;
1864 type = unpack_object_header(p, w_curs, &curpos, &size);
1865 if (type <= OBJ_NONE) {
1866 /* If getting the base itself fails, we first
1867 * retry the base, otherwise unwind */
1868 type = retry_bad_packed_offset(p, base_offset);
1869 if (type > OBJ_NONE)
1870 goto out;
1871 goto unwind;
1872 }
1873 }
1874
1875 switch (type) {
1876 case OBJ_BAD:
1877 case OBJ_COMMIT:
1878 case OBJ_TREE:
1879 case OBJ_BLOB:
1880 case OBJ_TAG:
1881 break;
1882 default:
1883 error("unknown object type %i at offset %"PRIuMAX" in %s",
1884 type, (uintmax_t)obj_offset, p->pack_name);
1885 type = OBJ_BAD;
1886 }
1887
1888 out:
1889 if (poi_stack != small_poi_stack)
1890 free(poi_stack);
1891 return type;
1892
1893 unwind:
1894 while (poi_stack_nr) {
1895 obj_offset = poi_stack[--poi_stack_nr];
1896 type = retry_bad_packed_offset(p, obj_offset);
1897 if (type > OBJ_NONE)
1898 goto out;
1899 }
1900 type = OBJ_BAD;
1901 goto out;
1902 }
1903
1904 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1905 struct object_info *oi)
1906 {
1907 struct pack_window *w_curs = NULL;
1908 unsigned long size;
1909 off_t curpos = obj_offset;
1910 enum object_type type;
1911
1912 /*
1913 * We always get the representation type, but only convert it to
1914 * a "real" type later if the caller is interested.
1915 */
1916 type = unpack_object_header(p, &w_curs, &curpos, &size);
1917
1918 if (oi->sizep) {
1919 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1920 off_t tmp_pos = curpos;
1921 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1922 type, obj_offset);
1923 if (!base_offset) {
1924 type = OBJ_BAD;
1925 goto out;
1926 }
1927 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1928 if (*oi->sizep == 0) {
1929 type = OBJ_BAD;
1930 goto out;
1931 }
1932 } else {
1933 *oi->sizep = size;
1934 }
1935 }
1936
1937 if (oi->disk_sizep) {
1938 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1939 *oi->disk_sizep = revidx[1].offset - obj_offset;
1940 }
1941
1942 if (oi->typep) {
1943 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1944 if (*oi->typep < 0) {
1945 type = OBJ_BAD;
1946 goto out;
1947 }
1948 }
1949
1950 if (oi->delta_base_sha1) {
1951 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1952 const unsigned char *base;
1953
1954 base = get_delta_base_sha1(p, &w_curs, curpos,
1955 type, obj_offset);
1956 if (!base) {
1957 type = OBJ_BAD;
1958 goto out;
1959 }
1960
1961 hashcpy(oi->delta_base_sha1, base);
1962 } else
1963 hashclr(oi->delta_base_sha1);
1964 }
1965
1966 out:
1967 unuse_pack(&w_curs);
1968 return type;
1969 }
1970
1971 static void *unpack_compressed_entry(struct packed_git *p,
1972 struct pack_window **w_curs,
1973 off_t curpos,
1974 unsigned long size)
1975 {
1976 int st;
1977 git_zstream stream;
1978 unsigned char *buffer, *in;
1979
1980 buffer = xmallocz_gently(size);
1981 if (!buffer)
1982 return NULL;
1983 memset(&stream, 0, sizeof(stream));
1984 stream.next_out = buffer;
1985 stream.avail_out = size + 1;
1986
1987 git_inflate_init(&stream);
1988 do {
1989 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1990 stream.next_in = in;
1991 st = git_inflate(&stream, Z_FINISH);
1992 if (!stream.avail_out)
1993 break; /* the payload is larger than it should be */
1994 curpos += stream.next_in - in;
1995 } while (st == Z_OK || st == Z_BUF_ERROR);
1996 git_inflate_end(&stream);
1997 if ((st != Z_STREAM_END) || stream.total_out != size) {
1998 free(buffer);
1999 return NULL;
2000 }
2001
2002 return buffer;
2003 }
2004
2005 #define MAX_DELTA_CACHE (256)
2006
2007 static size_t delta_base_cached;
2008
2009 static struct delta_base_cache_lru_list {
2010 struct delta_base_cache_lru_list *prev;
2011 struct delta_base_cache_lru_list *next;
2012 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2013
2014 static struct delta_base_cache_entry {
2015 struct delta_base_cache_lru_list lru;
2016 void *data;
2017 struct packed_git *p;
2018 off_t base_offset;
2019 unsigned long size;
2020 enum object_type type;
2021 } delta_base_cache[MAX_DELTA_CACHE];
2022
2023 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2024 {
2025 unsigned long hash;
2026
2027 hash = (unsigned long)p + (unsigned long)base_offset;
2028 hash += (hash >> 8) + (hash >> 16);
2029 return hash % MAX_DELTA_CACHE;
2030 }
2031
2032 static struct delta_base_cache_entry *
2033 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2034 {
2035 unsigned long hash = pack_entry_hash(p, base_offset);
2036 return delta_base_cache + hash;
2037 }
2038
2039 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2040 struct packed_git *p, off_t base_offset)
2041 {
2042 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2043 }
2044
2045 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2046 {
2047 struct delta_base_cache_entry *ent;
2048 ent = get_delta_base_cache_entry(p, base_offset);
2049 return eq_delta_base_cache_entry(ent, p, base_offset);
2050 }
2051
2052 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2053 {
2054 ent->data = NULL;
2055 ent->lru.next->prev = ent->lru.prev;
2056 ent->lru.prev->next = ent->lru.next;
2057 delta_base_cached -= ent->size;
2058 }
2059
2060 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2061 unsigned long *base_size, enum object_type *type, int keep_cache)
2062 {
2063 struct delta_base_cache_entry *ent;
2064 void *ret;
2065
2066 ent = get_delta_base_cache_entry(p, base_offset);
2067
2068 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2069 return unpack_entry(p, base_offset, type, base_size);
2070
2071 ret = ent->data;
2072
2073 if (!keep_cache)
2074 clear_delta_base_cache_entry(ent);
2075 else
2076 ret = xmemdupz(ent->data, ent->size);
2077 *type = ent->type;
2078 *base_size = ent->size;
2079 return ret;
2080 }
2081
2082 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2083 {
2084 if (ent->data) {
2085 free(ent->data);
2086 ent->data = NULL;
2087 ent->lru.next->prev = ent->lru.prev;
2088 ent->lru.prev->next = ent->lru.next;
2089 delta_base_cached -= ent->size;
2090 }
2091 }
2092
2093 void clear_delta_base_cache(void)
2094 {
2095 unsigned long p;
2096 for (p = 0; p < MAX_DELTA_CACHE; p++)
2097 release_delta_base_cache(&delta_base_cache[p]);
2098 }
2099
2100 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2101 void *base, unsigned long base_size, enum object_type type)
2102 {
2103 unsigned long hash = pack_entry_hash(p, base_offset);
2104 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2105 struct delta_base_cache_lru_list *lru;
2106
2107 release_delta_base_cache(ent);
2108 delta_base_cached += base_size;
2109
2110 for (lru = delta_base_cache_lru.next;
2111 delta_base_cached > delta_base_cache_limit
2112 && lru != &delta_base_cache_lru;
2113 lru = lru->next) {
2114 struct delta_base_cache_entry *f = (void *)lru;
2115 if (f->type == OBJ_BLOB)
2116 release_delta_base_cache(f);
2117 }
2118 for (lru = delta_base_cache_lru.next;
2119 delta_base_cached > delta_base_cache_limit
2120 && lru != &delta_base_cache_lru;
2121 lru = lru->next) {
2122 struct delta_base_cache_entry *f = (void *)lru;
2123 release_delta_base_cache(f);
2124 }
2125
2126 ent->p = p;
2127 ent->base_offset = base_offset;
2128 ent->type = type;
2129 ent->data = base;
2130 ent->size = base_size;
2131 ent->lru.next = &delta_base_cache_lru;
2132 ent->lru.prev = delta_base_cache_lru.prev;
2133 delta_base_cache_lru.prev->next = &ent->lru;
2134 delta_base_cache_lru.prev = &ent->lru;
2135 }
2136
2137 static void *read_object(const unsigned char *sha1, enum object_type *type,
2138 unsigned long *size);
2139
2140 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2141 {
2142 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2143 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2144 p->pack_name, (uintmax_t)obj_offset);
2145 }
2146
2147 int do_check_packed_object_crc;
2148
2149 #define UNPACK_ENTRY_STACK_PREALLOC 64
2150 struct unpack_entry_stack_ent {
2151 off_t obj_offset;
2152 off_t curpos;
2153 unsigned long size;
2154 };
2155
2156 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2157 enum object_type *final_type, unsigned long *final_size)
2158 {
2159 struct pack_window *w_curs = NULL;
2160 off_t curpos = obj_offset;
2161 void *data = NULL;
2162 unsigned long size;
2163 enum object_type type;
2164 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2165 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2166 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2167 int base_from_cache = 0;
2168
2169 write_pack_access_log(p, obj_offset);
2170
2171 /* PHASE 1: drill down to the innermost base object */
2172 for (;;) {
2173 off_t base_offset;
2174 int i;
2175 struct delta_base_cache_entry *ent;
2176
2177 ent = get_delta_base_cache_entry(p, curpos);
2178 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2179 type = ent->type;
2180 data = ent->data;
2181 size = ent->size;
2182 clear_delta_base_cache_entry(ent);
2183 base_from_cache = 1;
2184 break;
2185 }
2186
2187 if (do_check_packed_object_crc && p->index_version > 1) {
2188 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2189 unsigned long len = revidx[1].offset - obj_offset;
2190 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2191 const unsigned char *sha1 =
2192 nth_packed_object_sha1(p, revidx->nr);
2193 error("bad packed object CRC for %s",
2194 sha1_to_hex(sha1));
2195 mark_bad_packed_object(p, sha1);
2196 unuse_pack(&w_curs);
2197 return NULL;
2198 }
2199 }
2200
2201 type = unpack_object_header(p, &w_curs, &curpos, &size);
2202 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2203 break;
2204
2205 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2206 if (!base_offset) {
2207 error("failed to validate delta base reference "
2208 "at offset %"PRIuMAX" from %s",
2209 (uintmax_t)curpos, p->pack_name);
2210 /* bail to phase 2, in hopes of recovery */
2211 data = NULL;
2212 break;
2213 }
2214
2215 /* push object, proceed to base */
2216 if (delta_stack_nr >= delta_stack_alloc
2217 && delta_stack == small_delta_stack) {
2218 delta_stack_alloc = alloc_nr(delta_stack_nr);
2219 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2220 memcpy(delta_stack, small_delta_stack,
2221 sizeof(*delta_stack)*delta_stack_nr);
2222 } else {
2223 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2224 }
2225 i = delta_stack_nr++;
2226 delta_stack[i].obj_offset = obj_offset;
2227 delta_stack[i].curpos = curpos;
2228 delta_stack[i].size = size;
2229
2230 curpos = obj_offset = base_offset;
2231 }
2232
2233 /* PHASE 2: handle the base */
2234 switch (type) {
2235 case OBJ_OFS_DELTA:
2236 case OBJ_REF_DELTA:
2237 if (data)
2238 die("BUG in unpack_entry: left loop at a valid delta");
2239 break;
2240 case OBJ_COMMIT:
2241 case OBJ_TREE:
2242 case OBJ_BLOB:
2243 case OBJ_TAG:
2244 if (!base_from_cache)
2245 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2246 break;
2247 default:
2248 data = NULL;
2249 error("unknown object type %i at offset %"PRIuMAX" in %s",
2250 type, (uintmax_t)obj_offset, p->pack_name);
2251 }
2252
2253 /* PHASE 3: apply deltas in order */
2254
2255 /* invariants:
2256 * 'data' holds the base data, or NULL if there was corruption
2257 */
2258 while (delta_stack_nr) {
2259 void *delta_data;
2260 void *base = data;
2261 unsigned long delta_size, base_size = size;
2262 int i;
2263
2264 data = NULL;
2265
2266 if (base)
2267 add_delta_base_cache(p, obj_offset, base, base_size, type);
2268
2269 if (!base) {
2270 /*
2271 * We're probably in deep shit, but let's try to fetch
2272 * the required base anyway from another pack or loose.
2273 * This is costly but should happen only in the presence
2274 * of a corrupted pack, and is better than failing outright.
2275 */
2276 struct revindex_entry *revidx;
2277 const unsigned char *base_sha1;
2278 revidx = find_pack_revindex(p, obj_offset);
2279 if (revidx) {
2280 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2281 error("failed to read delta base object %s"
2282 " at offset %"PRIuMAX" from %s",
2283 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2284 p->pack_name);
2285 mark_bad_packed_object(p, base_sha1);
2286 base = read_object(base_sha1, &type, &base_size);
2287 }
2288 }
2289
2290 i = --delta_stack_nr;
2291 obj_offset = delta_stack[i].obj_offset;
2292 curpos = delta_stack[i].curpos;
2293 delta_size = delta_stack[i].size;
2294
2295 if (!base)
2296 continue;
2297
2298 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2299
2300 if (!delta_data) {
2301 error("failed to unpack compressed delta "
2302 "at offset %"PRIuMAX" from %s",
2303 (uintmax_t)curpos, p->pack_name);
2304 data = NULL;
2305 continue;
2306 }
2307
2308 data = patch_delta(base, base_size,
2309 delta_data, delta_size,
2310 &size);
2311
2312 /*
2313 * We could not apply the delta; warn the user, but keep going.
2314 * Our failure will be noticed either in the next iteration of
2315 * the loop, or if this is the final delta, in the caller when
2316 * we return NULL. Those code paths will take care of making
2317 * a more explicit warning and retrying with another copy of
2318 * the object.
2319 */
2320 if (!data)
2321 error("failed to apply delta");
2322
2323 free(delta_data);
2324 }
2325
2326 *final_type = type;
2327 *final_size = size;
2328
2329 unuse_pack(&w_curs);
2330
2331 if (delta_stack != small_delta_stack)
2332 free(delta_stack);
2333
2334 return data;
2335 }
2336
2337 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2338 uint32_t n)
2339 {
2340 const unsigned char *index = p->index_data;
2341 if (!index) {
2342 if (open_pack_index(p))
2343 return NULL;
2344 index = p->index_data;
2345 }
2346 if (n >= p->num_objects)
2347 return NULL;
2348 index += 4 * 256;
2349 if (p->index_version == 1) {
2350 return index + 24 * n + 4;
2351 } else {
2352 index += 8;
2353 return index + 20 * n;
2354 }
2355 }
2356
2357 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2358 {
2359 const unsigned char *index = p->index_data;
2360 index += 4 * 256;
2361 if (p->index_version == 1) {
2362 return ntohl(*((uint32_t *)(index + 24 * n)));
2363 } else {
2364 uint32_t off;
2365 index += 8 + p->num_objects * (20 + 4);
2366 off = ntohl(*((uint32_t *)(index + 4 * n)));
2367 if (!(off & 0x80000000))
2368 return off;
2369 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2370 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2371 ntohl(*((uint32_t *)(index + 4)));
2372 }
2373 }
2374
2375 off_t find_pack_entry_one(const unsigned char *sha1,
2376 struct packed_git *p)
2377 {
2378 const uint32_t *level1_ofs = p->index_data;
2379 const unsigned char *index = p->index_data;
2380 unsigned hi, lo, stride;
2381 static int use_lookup = -1;
2382 static int debug_lookup = -1;
2383
2384 if (debug_lookup < 0)
2385 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2386
2387 if (!index) {
2388 if (open_pack_index(p))
2389 return 0;
2390 level1_ofs = p->index_data;
2391 index = p->index_data;
2392 }
2393 if (p->index_version > 1) {
2394 level1_ofs += 2;
2395 index += 8;
2396 }
2397 index += 4 * 256;
2398 hi = ntohl(level1_ofs[*sha1]);
2399 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2400 if (p->index_version > 1) {
2401 stride = 20;
2402 } else {
2403 stride = 24;
2404 index += 4;
2405 }
2406
2407 if (debug_lookup)
2408 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2409 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2410
2411 if (use_lookup < 0)
2412 use_lookup = !!getenv("GIT_USE_LOOKUP");
2413 if (use_lookup) {
2414 int pos = sha1_entry_pos(index, stride, 0,
2415 lo, hi, p->num_objects, sha1);
2416 if (pos < 0)
2417 return 0;
2418 return nth_packed_object_offset(p, pos);
2419 }
2420
2421 do {
2422 unsigned mi = (lo + hi) / 2;
2423 int cmp = hashcmp(index + mi * stride, sha1);
2424
2425 if (debug_lookup)
2426 printf("lo %u hi %u rg %u mi %u\n",
2427 lo, hi, hi - lo, mi);
2428 if (!cmp)
2429 return nth_packed_object_offset(p, mi);
2430 if (cmp > 0)
2431 hi = mi;
2432 else
2433 lo = mi+1;
2434 } while (lo < hi);
2435 return 0;
2436 }
2437
2438 int is_pack_valid(struct packed_git *p)
2439 {
2440 /* An already open pack is known to be valid. */
2441 if (p->pack_fd != -1)
2442 return 1;
2443
2444 /* If the pack has one window completely covering the
2445 * file size, the pack is known to be valid even if
2446 * the descriptor is not currently open.
2447 */
2448 if (p->windows) {
2449 struct pack_window *w = p->windows;
2450
2451 if (!w->offset && w->len == p->pack_size)
2452 return 1;
2453 }
2454
2455 /* Force the pack to open to prove its valid. */
2456 return !open_packed_git(p);
2457 }
2458
2459 static int fill_pack_entry(const unsigned char *sha1,
2460 struct pack_entry *e,
2461 struct packed_git *p)
2462 {
2463 off_t offset;
2464
2465 if (p->num_bad_objects) {
2466 unsigned i;
2467 for (i = 0; i < p->num_bad_objects; i++)
2468 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2469 return 0;
2470 }
2471
2472 offset = find_pack_entry_one(sha1, p);
2473 if (!offset)
2474 return 0;
2475
2476 /*
2477 * We are about to tell the caller where they can locate the
2478 * requested object. We better make sure the packfile is
2479 * still here and can be accessed before supplying that
2480 * answer, as it may have been deleted since the index was
2481 * loaded!
2482 */
2483 if (!is_pack_valid(p))
2484 return 0;
2485 e->offset = offset;
2486 e->p = p;
2487 hashcpy(e->sha1, sha1);
2488 return 1;
2489 }
2490
2491 /*
2492 * Iff a pack file contains the object named by sha1, return true and
2493 * store its location to e.
2494 */
2495 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2496 {
2497 struct packed_git *p;
2498
2499 prepare_packed_git();
2500 if (!packed_git)
2501 return 0;
2502
2503 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2504 return 1;
2505
2506 for (p = packed_git; p; p = p->next) {
2507 if (p == last_found_pack)
2508 continue; /* we already checked this one */
2509
2510 if (fill_pack_entry(sha1, e, p)) {
2511 last_found_pack = p;
2512 return 1;
2513 }
2514 }
2515 return 0;
2516 }
2517
2518 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2519 struct packed_git *packs)
2520 {
2521 struct packed_git *p;
2522
2523 for (p = packs; p; p = p->next) {
2524 if (find_pack_entry_one(sha1, p))
2525 return p;
2526 }
2527 return NULL;
2528
2529 }
2530
2531 static int sha1_loose_object_info(const unsigned char *sha1,
2532 struct object_info *oi)
2533 {
2534 int status;
2535 unsigned long mapsize, size;
2536 void *map;
2537 git_zstream stream;
2538 char hdr[32];
2539
2540 if (oi->delta_base_sha1)
2541 hashclr(oi->delta_base_sha1);
2542
2543 /*
2544 * If we don't care about type or size, then we don't
2545 * need to look inside the object at all. Note that we
2546 * do not optimize out the stat call, even if the
2547 * caller doesn't care about the disk-size, since our
2548 * return value implicitly indicates whether the
2549 * object even exists.
2550 */
2551 if (!oi->typep && !oi->sizep) {
2552 struct stat st;
2553 if (stat_sha1_file(sha1, &st) < 0)
2554 return -1;
2555 if (oi->disk_sizep)
2556 *oi->disk_sizep = st.st_size;
2557 return 0;
2558 }
2559
2560 map = map_sha1_file(sha1, &mapsize);
2561 if (!map)
2562 return -1;
2563 if (oi->disk_sizep)
2564 *oi->disk_sizep = mapsize;
2565 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2566 status = error("unable to unpack %s header",
2567 sha1_to_hex(sha1));
2568 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2569 status = error("unable to parse %s header", sha1_to_hex(sha1));
2570 else if (oi->sizep)
2571 *oi->sizep = size;
2572 git_inflate_end(&stream);
2573 munmap(map, mapsize);
2574 if (oi->typep)
2575 *oi->typep = status;
2576 return 0;
2577 }
2578
2579 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2580 {
2581 struct cached_object *co;
2582 struct pack_entry e;
2583 int rtype;
2584 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2585
2586 co = find_cached_object(real);
2587 if (co) {
2588 if (oi->typep)
2589 *(oi->typep) = co->type;
2590 if (oi->sizep)
2591 *(oi->sizep) = co->size;
2592 if (oi->disk_sizep)
2593 *(oi->disk_sizep) = 0;
2594 if (oi->delta_base_sha1)
2595 hashclr(oi->delta_base_sha1);
2596 oi->whence = OI_CACHED;
2597 return 0;
2598 }
2599
2600 if (!find_pack_entry(real, &e)) {
2601 /* Most likely it's a loose object. */
2602 if (!sha1_loose_object_info(real, oi)) {
2603 oi->whence = OI_LOOSE;
2604 return 0;
2605 }
2606
2607 /* Not a loose object; someone else may have just packed it. */
2608 reprepare_packed_git();
2609 if (!find_pack_entry(real, &e))
2610 return -1;
2611 }
2612
2613 rtype = packed_object_info(e.p, e.offset, oi);
2614 if (rtype < 0) {
2615 mark_bad_packed_object(e.p, real);
2616 return sha1_object_info_extended(real, oi, 0);
2617 } else if (in_delta_base_cache(e.p, e.offset)) {
2618 oi->whence = OI_DBCACHED;
2619 } else {
2620 oi->whence = OI_PACKED;
2621 oi->u.packed.offset = e.offset;
2622 oi->u.packed.pack = e.p;
2623 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2624 rtype == OBJ_OFS_DELTA);
2625 }
2626
2627 return 0;
2628 }
2629
2630 /* returns enum object_type or negative */
2631 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2632 {
2633 enum object_type type;
2634 struct object_info oi = {NULL};
2635
2636 oi.typep = &type;
2637 oi.sizep = sizep;
2638 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2639 return -1;
2640 return type;
2641 }
2642
2643 static void *read_packed_sha1(const unsigned char *sha1,
2644 enum object_type *type, unsigned long *size)
2645 {
2646 struct pack_entry e;
2647 void *data;
2648
2649 if (!find_pack_entry(sha1, &e))
2650 return NULL;
2651 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2652 if (!data) {
2653 /*
2654 * We're probably in deep shit, but let's try to fetch
2655 * the required object anyway from another pack or loose.
2656 * This should happen only in the presence of a corrupted
2657 * pack, and is better than failing outright.
2658 */
2659 error("failed to read object %s at offset %"PRIuMAX" from %s",
2660 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2661 mark_bad_packed_object(e.p, sha1);
2662 data = read_object(sha1, type, size);
2663 }
2664 return data;
2665 }
2666
2667 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2668 unsigned char *sha1)
2669 {
2670 struct cached_object *co;
2671
2672 hash_sha1_file(buf, len, typename(type), sha1);
2673 if (has_sha1_file(sha1) || find_cached_object(sha1))
2674 return 0;
2675 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2676 co = &cached_objects[cached_object_nr++];
2677 co->size = len;
2678 co->type = type;
2679 co->buf = xmalloc(len);
2680 memcpy(co->buf, buf, len);
2681 hashcpy(co->sha1, sha1);
2682 return 0;
2683 }
2684
2685 static void *read_object(const unsigned char *sha1, enum object_type *type,
2686 unsigned long *size)
2687 {
2688 unsigned long mapsize;
2689 void *map, *buf;
2690 struct cached_object *co;
2691
2692 co = find_cached_object(sha1);
2693 if (co) {
2694 *type = co->type;
2695 *size = co->size;
2696 return xmemdupz(co->buf, co->size);
2697 }
2698
2699 buf = read_packed_sha1(sha1, type, size);
2700 if (buf)
2701 return buf;
2702 map = map_sha1_file(sha1, &mapsize);
2703 if (map) {
2704 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2705 munmap(map, mapsize);
2706 return buf;
2707 }
2708 reprepare_packed_git();
2709 return read_packed_sha1(sha1, type, size);
2710 }
2711
2712 /*
2713 * This function dies on corrupt objects; the callers who want to
2714 * deal with them should arrange to call read_object() and give error
2715 * messages themselves.
2716 */
2717 void *read_sha1_file_extended(const unsigned char *sha1,
2718 enum object_type *type,
2719 unsigned long *size,
2720 unsigned flag)
2721 {
2722 void *data;
2723 const struct packed_git *p;
2724 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2725
2726 errno = 0;
2727 data = read_object(repl, type, size);
2728 if (data)
2729 return data;
2730
2731 if (errno && errno != ENOENT)
2732 die_errno("failed to read object %s", sha1_to_hex(sha1));
2733
2734 /* die if we replaced an object with one that does not exist */
2735 if (repl != sha1)
2736 die("replacement %s not found for %s",
2737 sha1_to_hex(repl), sha1_to_hex(sha1));
2738
2739 if (has_loose_object(repl)) {
2740 const char *path = sha1_file_name(sha1);
2741
2742 die("loose object %s (stored in %s) is corrupt",
2743 sha1_to_hex(repl), path);
2744 }
2745
2746 if ((p = has_packed_and_bad(repl)) != NULL)
2747 die("packed object %s (stored in %s) is corrupt",
2748 sha1_to_hex(repl), p->pack_name);
2749
2750 return NULL;
2751 }
2752
2753 void *read_object_with_reference(const unsigned char *sha1,
2754 const char *required_type_name,
2755 unsigned long *size,
2756 unsigned char *actual_sha1_return)
2757 {
2758 enum object_type type, required_type;
2759 void *buffer;
2760 unsigned long isize;
2761 unsigned char actual_sha1[20];
2762
2763 required_type = type_from_string(required_type_name);
2764 hashcpy(actual_sha1, sha1);
2765 while (1) {
2766 int ref_length = -1;
2767 const char *ref_type = NULL;
2768
2769 buffer = read_sha1_file(actual_sha1, &type, &isize);
2770 if (!buffer)
2771 return NULL;
2772 if (type == required_type) {
2773 *size = isize;
2774 if (actual_sha1_return)
2775 hashcpy(actual_sha1_return, actual_sha1);
2776 return buffer;
2777 }
2778 /* Handle references */
2779 else if (type == OBJ_COMMIT)
2780 ref_type = "tree ";
2781 else if (type == OBJ_TAG)
2782 ref_type = "object ";
2783 else {
2784 free(buffer);
2785 return NULL;
2786 }
2787 ref_length = strlen(ref_type);
2788
2789 if (ref_length + 40 > isize ||
2790 memcmp(buffer, ref_type, ref_length) ||
2791 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2792 free(buffer);
2793 return NULL;
2794 }
2795 free(buffer);
2796 /* Now we have the ID of the referred-to object in
2797 * actual_sha1. Check again. */
2798 }
2799 }
2800
2801 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2802 const char *type, unsigned char *sha1,
2803 char *hdr, int *hdrlen)
2804 {
2805 git_SHA_CTX c;
2806
2807 /* Generate the header */
2808 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2809
2810 /* Sha1.. */
2811 git_SHA1_Init(&c);
2812 git_SHA1_Update(&c, hdr, *hdrlen);
2813 git_SHA1_Update(&c, buf, len);
2814 git_SHA1_Final(sha1, &c);
2815 }
2816
2817 /*
2818 * Move the just written object into its final resting place.
2819 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2820 * "moving" is only a part of what it does, when no patch between
2821 * master to pu changes the call sites of this function.
2822 */
2823 int move_temp_to_file(const char *tmpfile, const char *filename)
2824 {
2825 int ret = 0;
2826
2827 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2828 goto try_rename;
2829 else if (link(tmpfile, filename))
2830 ret = errno;
2831
2832 /*
2833 * Coda hack - coda doesn't like cross-directory links,
2834 * so we fall back to a rename, which will mean that it
2835 * won't be able to check collisions, but that's not a
2836 * big deal.
2837 *
2838 * The same holds for FAT formatted media.
2839 *
2840 * When this succeeds, we just return. We have nothing
2841 * left to unlink.
2842 */
2843 if (ret && ret != EEXIST) {
2844 try_rename:
2845 if (!rename(tmpfile, filename))
2846 goto out;
2847 ret = errno;
2848 }
2849 unlink_or_warn(tmpfile);
2850 if (ret) {
2851 if (ret != EEXIST) {
2852 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2853 }
2854 /* FIXME!!! Collision check here ? */
2855 }
2856
2857 out:
2858 if (adjust_shared_perm(filename))
2859 return error("unable to set permission to '%s'", filename);
2860 return 0;
2861 }
2862
2863 static int write_buffer(int fd, const void *buf, size_t len)
2864 {
2865 if (write_in_full(fd, buf, len) < 0)
2866 return error("file write error (%s)", strerror(errno));
2867 return 0;
2868 }
2869
2870 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2871 unsigned char *sha1)
2872 {
2873 char hdr[32];
2874 int hdrlen;
2875 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2876 return 0;
2877 }
2878
2879 /* Finalize a file on disk, and close it. */
2880 static void close_sha1_file(int fd)
2881 {
2882 if (fsync_object_files)
2883 fsync_or_die(fd, "sha1 file");
2884 if (close(fd) != 0)
2885 die_errno("error when closing sha1 file");
2886 }
2887
2888 /* Size of directory component, including the ending '/' */
2889 static inline int directory_size(const char *filename)
2890 {
2891 const char *s = strrchr(filename, '/');
2892 if (!s)
2893 return 0;
2894 return s - filename + 1;
2895 }
2896
2897 /*
2898 * This creates a temporary file in the same directory as the final
2899 * 'filename'
2900 *
2901 * We want to avoid cross-directory filename renames, because those
2902 * can have problems on various filesystems (FAT, NFS, Coda).
2903 */
2904 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2905 {
2906 int fd, dirlen = directory_size(filename);
2907
2908 if (dirlen + 20 > bufsiz) {
2909 errno = ENAMETOOLONG;
2910 return -1;
2911 }
2912 memcpy(buffer, filename, dirlen);
2913 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2914 fd = git_mkstemp_mode(buffer, 0444);
2915 if (fd < 0 && dirlen && errno == ENOENT) {
2916 /* Make sure the directory exists */
2917 memcpy(buffer, filename, dirlen);
2918 buffer[dirlen-1] = 0;
2919 if (mkdir(buffer, 0777) && errno != EEXIST)
2920 return -1;
2921 if (adjust_shared_perm(buffer))
2922 return -1;
2923
2924 /* Try again */
2925 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2926 fd = git_mkstemp_mode(buffer, 0444);
2927 }
2928 return fd;
2929 }
2930
2931 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2932 const void *buf, unsigned long len, time_t mtime)
2933 {
2934 int fd, ret;
2935 unsigned char compressed[4096];
2936 git_zstream stream;
2937 git_SHA_CTX c;
2938 unsigned char parano_sha1[20];
2939 static char tmp_file[PATH_MAX];
2940 const char *filename = sha1_file_name(sha1);
2941
2942 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2943 if (fd < 0) {
2944 if (errno == EACCES)
2945 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2946 else
2947 return error("unable to create temporary file: %s", strerror(errno));
2948 }
2949
2950 /* Set it up */
2951 git_deflate_init(&stream, zlib_compression_level);
2952 stream.next_out = compressed;
2953 stream.avail_out = sizeof(compressed);
2954 git_SHA1_Init(&c);
2955
2956 /* First header.. */
2957 stream.next_in = (unsigned char *)hdr;
2958 stream.avail_in = hdrlen;
2959 while (git_deflate(&stream, 0) == Z_OK)
2960 ; /* nothing */
2961 git_SHA1_Update(&c, hdr, hdrlen);
2962
2963 /* Then the data itself.. */
2964 stream.next_in = (void *)buf;
2965 stream.avail_in = len;
2966 do {
2967 unsigned char *in0 = stream.next_in;
2968 ret = git_deflate(&stream, Z_FINISH);
2969 git_SHA1_Update(&c, in0, stream.next_in - in0);
2970 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2971 die("unable to write sha1 file");
2972 stream.next_out = compressed;
2973 stream.avail_out = sizeof(compressed);
2974 } while (ret == Z_OK);
2975
2976 if (ret != Z_STREAM_END)
2977 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2978 ret = git_deflate_end_gently(&stream);
2979 if (ret != Z_OK)
2980 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2981 git_SHA1_Final(parano_sha1, &c);
2982 if (hashcmp(sha1, parano_sha1) != 0)
2983 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2984
2985 close_sha1_file(fd);
2986
2987 if (mtime) {
2988 struct utimbuf utb;
2989 utb.actime = mtime;
2990 utb.modtime = mtime;
2991 if (utime(tmp_file, &utb) < 0)
2992 warning("failed utime() on %s: %s",
2993 tmp_file, strerror(errno));
2994 }
2995
2996 return move_temp_to_file(tmp_file, filename);
2997 }
2998
2999 static int freshen_loose_object(const unsigned char *sha1)
3000 {
3001 return check_and_freshen(sha1, 1);
3002 }
3003
3004 static int freshen_packed_object(const unsigned char *sha1)
3005 {
3006 struct pack_entry e;
3007 if (!find_pack_entry(sha1, &e))
3008 return 0;
3009 if (e.p->freshened)
3010 return 1;
3011 if (!freshen_file(e.p->pack_name))
3012 return 0;
3013 e.p->freshened = 1;
3014 return 1;
3015 }
3016
3017 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3018 {
3019 char hdr[32];
3020 int hdrlen;
3021
3022 /* Normally if we have it in the pack then we do not bother writing
3023 * it out into .git/objects/??/?{38} file.
3024 */
3025 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3026 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3027 return 0;
3028 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3029 }
3030
3031 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3032 unsigned char *sha1, unsigned flags)
3033 {
3034 char *header;
3035 int hdrlen, status = 0;
3036
3037 /* type string, SP, %lu of the length plus NUL must fit this */
3038 header = xmalloc(strlen(type) + 32);
3039 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3040
3041 if (!(flags & HASH_WRITE_OBJECT))
3042 goto cleanup;
3043 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3044 goto cleanup;
3045 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3046
3047 cleanup:
3048 free(header);
3049 return status;
3050 }
3051
3052 int force_object_loose(const unsigned char *sha1, time_t mtime)
3053 {
3054 void *buf;
3055 unsigned long len;
3056 enum object_type type;
3057 char hdr[32];
3058 int hdrlen;
3059 int ret;
3060
3061 if (has_loose_object(sha1))
3062 return 0;
3063 buf = read_packed_sha1(sha1, &type, &len);
3064 if (!buf)
3065 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3066 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3067 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3068 free(buf);
3069
3070 return ret;
3071 }
3072
3073 int has_pack_index(const unsigned char *sha1)
3074 {
3075 struct stat st;
3076 if (stat(sha1_pack_index_name(sha1), &st))
3077 return 0;
3078 return 1;
3079 }
3080
3081 int has_sha1_pack(const unsigned char *sha1)
3082 {
3083 struct pack_entry e;
3084 return find_pack_entry(sha1, &e);
3085 }
3086
3087 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3088 {
3089 struct pack_entry e;
3090
3091 if (find_pack_entry(sha1, &e))
3092 return 1;
3093 if (has_loose_object(sha1))
3094 return 1;
3095 if (flags & HAS_SHA1_QUICK)
3096 return 0;
3097 reprepare_packed_git();
3098 return find_pack_entry(sha1, &e);
3099 }
3100
3101 static void check_tree(const void *buf, size_t size)
3102 {
3103 struct tree_desc desc;
3104 struct name_entry entry;
3105
3106 init_tree_desc(&desc, buf, size);
3107 while (tree_entry(&desc, &entry))
3108 /* do nothing
3109 * tree_entry() will die() on malformed entries */
3110 ;
3111 }
3112
3113 static void check_commit(const void *buf, size_t size)
3114 {
3115 struct commit c;
3116 memset(&c, 0, sizeof(c));
3117 if (parse_commit_buffer(&c, buf, size))
3118 die("corrupt commit");
3119 }
3120
3121 static void check_tag(const void *buf, size_t size)
3122 {
3123 struct tag t;
3124 memset(&t, 0, sizeof(t));
3125 if (parse_tag_buffer(&t, buf, size))
3126 die("corrupt tag");
3127 }
3128
3129 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3130 enum object_type type,
3131 const char *path, unsigned flags)
3132 {
3133 int ret, re_allocated = 0;
3134 int write_object = flags & HASH_WRITE_OBJECT;
3135
3136 if (!type)
3137 type = OBJ_BLOB;
3138
3139 /*
3140 * Convert blobs to git internal format
3141 */
3142 if ((type == OBJ_BLOB) && path) {
3143 struct strbuf nbuf = STRBUF_INIT;
3144 if (convert_to_git(path, buf, size, &nbuf,
3145 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3146 buf = strbuf_detach(&nbuf, &size);
3147 re_allocated = 1;
3148 }
3149 }
3150 if (flags & HASH_FORMAT_CHECK) {
3151 if (type == OBJ_TREE)
3152 check_tree(buf, size);
3153 if (type == OBJ_COMMIT)
3154 check_commit(buf, size);
3155 if (type == OBJ_TAG)
3156 check_tag(buf, size);
3157 }
3158
3159 if (write_object)
3160 ret = write_sha1_file(buf, size, typename(type), sha1);
3161 else
3162 ret = hash_sha1_file(buf, size, typename(type), sha1);
3163 if (re_allocated)
3164 free(buf);
3165 return ret;
3166 }
3167
3168 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3169 const char *path, unsigned flags)
3170 {
3171 int ret;
3172 const int write_object = flags & HASH_WRITE_OBJECT;
3173 struct strbuf sbuf = STRBUF_INIT;
3174
3175 assert(path);
3176 assert(would_convert_to_git_filter_fd(path));
3177
3178 convert_to_git_filter_fd(path, fd, &sbuf,
3179 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3180
3181 if (write_object)
3182 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3183 sha1);
3184 else
3185 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3186 sha1);
3187 strbuf_release(&sbuf);
3188 return ret;
3189 }
3190
3191 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3192 const char *path, unsigned flags)
3193 {
3194 struct strbuf sbuf = STRBUF_INIT;
3195 int ret;
3196
3197 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3198 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3199 else
3200 ret = -1;
3201 strbuf_release(&sbuf);
3202 return ret;
3203 }
3204
3205 #define SMALL_FILE_SIZE (32*1024)
3206
3207 static int index_core(unsigned char *sha1, int fd, size_t size,
3208 enum object_type type, const char *path,
3209 unsigned flags)
3210 {
3211 int ret;
3212
3213 if (!size) {
3214 ret = index_mem(sha1, "", size, type, path, flags);
3215 } else if (size <= SMALL_FILE_SIZE) {
3216 char *buf = xmalloc(size);
3217 if (size == read_in_full(fd, buf, size))
3218 ret = index_mem(sha1, buf, size, type, path, flags);
3219 else
3220 ret = error("short read %s", strerror(errno));
3221 free(buf);
3222 } else {
3223 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3224 ret = index_mem(sha1, buf, size, type, path, flags);
3225 munmap(buf, size);
3226 }
3227 return ret;
3228 }
3229
3230 /*
3231 * This creates one packfile per large blob unless bulk-checkin
3232 * machinery is "plugged".
3233 *
3234 * This also bypasses the usual "convert-to-git" dance, and that is on
3235 * purpose. We could write a streaming version of the converting
3236 * functions and insert that before feeding the data to fast-import
3237 * (or equivalent in-core API described above). However, that is
3238 * somewhat complicated, as we do not know the size of the filter
3239 * result, which we need to know beforehand when writing a git object.
3240 * Since the primary motivation for trying to stream from the working
3241 * tree file and to avoid mmaping it in core is to deal with large
3242 * binary blobs, they generally do not want to get any conversion, and
3243 * callers should avoid this code path when filters are requested.
3244 */
3245 static int index_stream(unsigned char *sha1, int fd, size_t size,
3246 enum object_type type, const char *path,
3247 unsigned flags)
3248 {
3249 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3250 }
3251
3252 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3253 enum object_type type, const char *path, unsigned flags)
3254 {
3255 int ret;
3256
3257 /*
3258 * Call xsize_t() only when needed to avoid potentially unnecessary
3259 * die() for large files.
3260 */
3261 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3262 ret = index_stream_convert_blob(sha1, fd, path, flags);
3263 else if (!S_ISREG(st->st_mode))
3264 ret = index_pipe(sha1, fd, type, path, flags);
3265 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3266 (path && would_convert_to_git(path)))
3267 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3268 flags);
3269 else
3270 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3271 flags);
3272 close(fd);
3273 return ret;
3274 }
3275
3276 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3277 {
3278 int fd;
3279 struct strbuf sb = STRBUF_INIT;
3280
3281 switch (st->st_mode & S_IFMT) {
3282 case S_IFREG:
3283 fd = open(path, O_RDONLY);
3284 if (fd < 0)
3285 return error("open(\"%s\"): %s", path,
3286 strerror(errno));
3287 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3288 return error("%s: failed to insert into database",
3289 path);
3290 break;
3291 case S_IFLNK:
3292 if (strbuf_readlink(&sb, path, st->st_size)) {
3293 char *errstr = strerror(errno);
3294 return error("readlink(\"%s\"): %s", path,
3295 errstr);
3296 }
3297 if (!(flags & HASH_WRITE_OBJECT))
3298 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3299 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3300 return error("%s: failed to insert into database",
3301 path);
3302 strbuf_release(&sb);
3303 break;
3304 case S_IFDIR:
3305 return resolve_gitlink_ref(path, "HEAD", sha1);
3306 default:
3307 return error("%s: unsupported file type", path);
3308 }
3309 return 0;
3310 }
3311
3312 int read_pack_header(int fd, struct pack_header *header)
3313 {
3314 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3315 /* "eof before pack header was fully read" */
3316 return PH_ERROR_EOF;
3317
3318 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3319 /* "protocol error (pack signature mismatch detected)" */
3320 return PH_ERROR_PACK_SIGNATURE;
3321 if (!pack_version_ok(header->hdr_version))
3322 /* "protocol error (pack version unsupported)" */
3323 return PH_ERROR_PROTOCOL;
3324 return 0;
3325 }
3326
3327 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3328 {
3329 enum object_type type = sha1_object_info(sha1, NULL);
3330 if (type < 0)
3331 die("%s is not a valid object", sha1_to_hex(sha1));
3332 if (type != expect)
3333 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3334 typename(expect));
3335 }
3336
3337 static int for_each_file_in_obj_subdir(int subdir_nr,
3338 struct strbuf *path,
3339 each_loose_object_fn obj_cb,
3340 each_loose_cruft_fn cruft_cb,
3341 each_loose_subdir_fn subdir_cb,
3342 void *data)
3343 {
3344 size_t baselen = path->len;
3345 DIR *dir = opendir(path->buf);
3346 struct dirent *de;
3347 int r = 0;
3348
3349 if (!dir) {
3350 if (errno == ENOENT)
3351 return 0;
3352 return error("unable to open %s: %s", path->buf, strerror(errno));
3353 }
3354
3355 while ((de = readdir(dir))) {
3356 if (is_dot_or_dotdot(de->d_name))
3357 continue;
3358
3359 strbuf_setlen(path, baselen);
3360 strbuf_addf(path, "/%s", de->d_name);
3361
3362 if (strlen(de->d_name) == 38) {
3363 char hex[41];
3364 unsigned char sha1[20];
3365
3366 snprintf(hex, sizeof(hex), "%02x%s",
3367 subdir_nr, de->d_name);
3368 if (!get_sha1_hex(hex, sha1)) {
3369 if (obj_cb) {
3370 r = obj_cb(sha1, path->buf, data);
3371 if (r)
3372 break;
3373 }
3374 continue;
3375 }
3376 }
3377
3378 if (cruft_cb) {
3379 r = cruft_cb(de->d_name, path->buf, data);
3380 if (r)
3381 break;
3382 }
3383 }
3384 strbuf_setlen(path, baselen);
3385
3386 if (!r && subdir_cb)
3387 r = subdir_cb(subdir_nr, path->buf, data);
3388
3389 closedir(dir);
3390 return r;
3391 }
3392
3393 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3394 each_loose_object_fn obj_cb,
3395 each_loose_cruft_fn cruft_cb,
3396 each_loose_subdir_fn subdir_cb,
3397 void *data)
3398 {
3399 size_t baselen = path->len;
3400 int r = 0;
3401 int i;
3402
3403 for (i = 0; i < 256; i++) {
3404 strbuf_addf(path, "/%02x", i);
3405 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3406 subdir_cb, data);
3407 strbuf_setlen(path, baselen);
3408 if (r)
3409 break;
3410 }
3411
3412 return r;
3413 }
3414
3415 int for_each_loose_file_in_objdir(const char *path,
3416 each_loose_object_fn obj_cb,
3417 each_loose_cruft_fn cruft_cb,
3418 each_loose_subdir_fn subdir_cb,
3419 void *data)
3420 {
3421 struct strbuf buf = STRBUF_INIT;
3422 int r;
3423
3424 strbuf_addstr(&buf, path);
3425 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3426 subdir_cb, data);
3427 strbuf_release(&buf);
3428
3429 return r;
3430 }
3431
3432 struct loose_alt_odb_data {
3433 each_loose_object_fn *cb;
3434 void *data;
3435 };
3436
3437 static int loose_from_alt_odb(struct alternate_object_database *alt,
3438 void *vdata)
3439 {
3440 struct loose_alt_odb_data *data = vdata;
3441 struct strbuf buf = STRBUF_INIT;
3442 int r;
3443
3444 /* copy base not including trailing '/' */
3445 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3446 r = for_each_loose_file_in_objdir_buf(&buf,
3447 data->cb, NULL, NULL,
3448 data->data);
3449 strbuf_release(&buf);
3450 return r;
3451 }
3452
3453 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3454 {
3455 struct loose_alt_odb_data alt;
3456 int r;
3457
3458 r = for_each_loose_file_in_objdir(get_object_directory(),
3459 cb, NULL, NULL, data);
3460 if (r)
3461 return r;
3462
3463 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3464 return 0;
3465
3466 alt.cb = cb;
3467 alt.data = data;
3468 return foreach_alt_odb(loose_from_alt_odb, &alt);
3469 }
3470
3471 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3472 {
3473 uint32_t i;
3474 int r = 0;
3475
3476 for (i = 0; i < p->num_objects; i++) {
3477 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3478
3479 if (!sha1)
3480 return error("unable to get sha1 of object %u in %s",
3481 i, p->pack_name);
3482
3483 r = cb(sha1, p, i, data);
3484 if (r)
3485 break;
3486 }
3487 return r;
3488 }
3489
3490 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3491 {
3492 struct packed_git *p;
3493 int r = 0;
3494
3495 prepare_packed_git();
3496 for (p = packed_git; p; p = p->next) {
3497 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3498 continue;
3499 r = for_each_object_in_pack(p, cb, data);
3500 if (r)
3501 break;
3502 }
3503 return r;
3504 }