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#include "mru.h"
27#include "list.h"
28
29#ifndef O_NOATIME
30#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
31#define O_NOATIME 01000000
32#else
33#define O_NOATIME 0
34#endif
35#endif
36
37#define SZ_FMT PRIuMAX
38static inline uintmax_t sz_fmt(size_t s) { return s; }
39
40const unsigned char null_sha1[20];
41const struct object_id null_oid;
42
43/*
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
48 */
49static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54} *cached_objects;
55static int cached_object_nr, cached_object_alloc;
56
57static struct cached_object empty_tree = {
58 EMPTY_TREE_SHA1_BIN_LITERAL,
59 OBJ_TREE,
60 "",
61 0
62};
63
64static struct cached_object *find_cached_object(const unsigned char *sha1)
65{
66 int i;
67 struct cached_object *co = cached_objects;
68
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
72 }
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
76}
77
78int mkdir_in_gitdir(const char *path)
79{
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
84
85 if (errno != EEXIST)
86 return -1;
87 /*
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
93 */
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
101 }
102 strbuf_release(&sb);
103 }
104 return adjust_shared_perm(path);
105}
106
107enum scld_error safe_create_leading_directories(char *path)
108{
109 char *next_component = path + offset_1st_component(path);
110 enum scld_error ret = SCLD_OK;
111
112 while (ret == SCLD_OK && next_component) {
113 struct stat st;
114 char *slash = next_component, slash_character;
115
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
118
119 if (!*slash)
120 break;
121
122 next_component = slash + 1;
123 while (is_dir_sep(*next_component))
124 next_component++;
125 if (!*next_component)
126 break;
127
128 slash_character = *slash;
129 *slash = '\0';
130 if (!stat(path, &st)) {
131 /* path exists */
132 if (!S_ISDIR(st.st_mode))
133 ret = SCLD_EXISTS;
134 } else if (mkdir(path, 0777)) {
135 if (errno == EEXIST &&
136 !stat(path, &st) && S_ISDIR(st.st_mode))
137 ; /* somebody created it since we checked */
138 else if (errno == ENOENT)
139 /*
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
147 */
148 ret = SCLD_VANISHED;
149 else
150 ret = SCLD_FAILED;
151 } else if (adjust_shared_perm(path)) {
152 ret = SCLD_PERMS;
153 }
154 *slash = slash_character;
155 }
156 return ret;
157}
158
159enum scld_error safe_create_leading_directories_const(const char *path)
160{
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf = xstrdup(path);
163 enum scld_error result = safe_create_leading_directories(buf);
164 free(buf);
165 return result;
166}
167
168static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
169{
170 int i;
171 for (i = 0; i < 20; i++) {
172 static char hex[] = "0123456789abcdef";
173 unsigned int val = sha1[i];
174 char *pos = pathbuf + i*2 + (i > 0);
175 *pos++ = hex[val >> 4];
176 *pos = hex[val & 0xf];
177 }
178}
179
180const char *sha1_file_name(const unsigned char *sha1)
181{
182 static char buf[PATH_MAX];
183 const char *objdir;
184 int len;
185
186 objdir = get_object_directory();
187 len = strlen(objdir);
188
189 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
190 if (len + 43 > PATH_MAX)
191 die("insanely long object directory %s", objdir);
192 memcpy(buf, objdir, len);
193 buf[len] = '/';
194 buf[len+3] = '/';
195 buf[len+42] = '\0';
196 fill_sha1_path(buf + len + 1, sha1);
197 return buf;
198}
199
200/*
201 * Return the name of the pack or index file with the specified sha1
202 * in its filename. *base and *name are scratch space that must be
203 * provided by the caller. which should be "pack" or "idx".
204 */
205static char *sha1_get_pack_name(const unsigned char *sha1,
206 struct strbuf *buf,
207 const char *which)
208{
209 strbuf_reset(buf);
210 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
211 sha1_to_hex(sha1), which);
212 return buf->buf;
213}
214
215char *sha1_pack_name(const unsigned char *sha1)
216{
217 static struct strbuf buf = STRBUF_INIT;
218 return sha1_get_pack_name(sha1, &buf, "pack");
219}
220
221char *sha1_pack_index_name(const unsigned char *sha1)
222{
223 static struct strbuf buf = STRBUF_INIT;
224 return sha1_get_pack_name(sha1, &buf, "idx");
225}
226
227struct alternate_object_database *alt_odb_list;
228static struct alternate_object_database **alt_odb_tail;
229
230/*
231 * Prepare alternate object database registry.
232 *
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
244 */
245static int link_alt_odb_entry(const char *entry, const char *relative_base,
246 int depth, const char *normalized_objdir)
247{
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 size_t pfxlen, entlen;
251 struct strbuf pathbuf = STRBUF_INIT;
252
253 if (!is_absolute_path(entry) && relative_base) {
254 strbuf_addstr(&pathbuf, real_path(relative_base));
255 strbuf_addch(&pathbuf, '/');
256 }
257 strbuf_addstr(&pathbuf, entry);
258
259 normalize_path_copy(pathbuf.buf, pathbuf.buf);
260
261 pfxlen = strlen(pathbuf.buf);
262
263 /*
264 * The trailing slash after the directory name is given by
265 * this function at the end. Remove duplicates.
266 */
267 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
268 pfxlen -= 1;
269
270 entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
271 ent = xmalloc(st_add(sizeof(*ent), entlen));
272 memcpy(ent->base, pathbuf.buf, pfxlen);
273 strbuf_release(&pathbuf);
274
275 ent->name = ent->base + pfxlen + 1;
276 ent->base[pfxlen + 3] = '/';
277 ent->base[pfxlen] = ent->base[entlen-1] = 0;
278
279 /* Detect cases where alternate disappeared */
280 if (!is_directory(ent->base)) {
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
283 ent->base);
284 free(ent);
285 return -1;
286 }
287
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
290 */
291 for (alt = alt_odb_list; alt; alt = alt->next) {
292 if (pfxlen == alt->name - alt->base - 1 &&
293 !memcmp(ent->base, alt->base, pfxlen)) {
294 free(ent);
295 return -1;
296 }
297 }
298 if (!fspathcmp(ent->base, normalized_objdir)) {
299 free(ent);
300 return -1;
301 }
302
303 /* add the alternate entry */
304 *alt_odb_tail = ent;
305 alt_odb_tail = &(ent->next);
306 ent->next = NULL;
307
308 /* recursively add alternates */
309 read_info_alternates(ent->base, depth + 1);
310
311 ent->base[pfxlen] = '/';
312
313 return 0;
314}
315
316static void link_alt_odb_entries(const char *alt, int len, int sep,
317 const char *relative_base, int depth)
318{
319 struct string_list entries = STRING_LIST_INIT_NODUP;
320 char *alt_copy;
321 int i;
322 struct strbuf objdirbuf = STRBUF_INIT;
323
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
328 }
329
330 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
331 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
332
333 alt_copy = xmemdupz(alt, len);
334 string_list_split_in_place(&entries, alt_copy, sep, -1);
335 for (i = 0; i < entries.nr; i++) {
336 const char *entry = entries.items[i].string;
337 if (entry[0] == '\0' || entry[0] == '#')
338 continue;
339 if (!is_absolute_path(entry) && depth) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, entry);
342 } else {
343 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
344 }
345 }
346 string_list_clear(&entries, 0);
347 free(alt_copy);
348 strbuf_release(&objdirbuf);
349}
350
351void read_info_alternates(const char * relative_base, int depth)
352{
353 char *map;
354 size_t mapsz;
355 struct stat st;
356 char *path;
357 int fd;
358
359 path = xstrfmt("%s/info/alternates", relative_base);
360 fd = git_open_noatime(path);
361 free(path);
362 if (fd < 0)
363 return;
364 if (fstat(fd, &st) || (st.st_size == 0)) {
365 close(fd);
366 return;
367 }
368 mapsz = xsize_t(st.st_size);
369 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
370 close(fd);
371
372 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
373
374 munmap(map, mapsz);
375}
376
377void add_to_alternates_file(const char *reference)
378{
379 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
380 char *alts = git_pathdup("objects/info/alternates");
381 FILE *in, *out;
382
383 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
384 out = fdopen_lock_file(lock, "w");
385 if (!out)
386 die_errno("unable to fdopen alternates lockfile");
387
388 in = fopen(alts, "r");
389 if (in) {
390 struct strbuf line = STRBUF_INIT;
391 int found = 0;
392
393 while (strbuf_getline(&line, in) != EOF) {
394 if (!strcmp(reference, line.buf)) {
395 found = 1;
396 break;
397 }
398 fprintf_or_die(out, "%s\n", line.buf);
399 }
400
401 strbuf_release(&line);
402 fclose(in);
403
404 if (found) {
405 rollback_lock_file(lock);
406 lock = NULL;
407 }
408 }
409 else if (errno != ENOENT)
410 die_errno("unable to read alternates file");
411
412 if (lock) {
413 fprintf_or_die(out, "%s\n", reference);
414 if (commit_lock_file(lock))
415 die_errno("unable to move new alternates file into place");
416 if (alt_odb_tail)
417 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
418 }
419 free(alts);
420}
421
422int foreach_alt_odb(alt_odb_fn fn, void *cb)
423{
424 struct alternate_object_database *ent;
425 int r = 0;
426
427 prepare_alt_odb();
428 for (ent = alt_odb_list; ent; ent = ent->next) {
429 r = fn(ent, cb);
430 if (r)
431 break;
432 }
433 return r;
434}
435
436void prepare_alt_odb(void)
437{
438 const char *alt;
439
440 if (alt_odb_tail)
441 return;
442
443 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
444 if (!alt) alt = "";
445
446 alt_odb_tail = &alt_odb_list;
447 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
448
449 read_info_alternates(get_object_directory(), 0);
450}
451
452/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
453static int freshen_file(const char *fn)
454{
455 struct utimbuf t;
456 t.actime = t.modtime = time(NULL);
457 return !utime(fn, &t);
458}
459
460/*
461 * All of the check_and_freshen functions return 1 if the file exists and was
462 * freshened (if freshening was requested), 0 otherwise. If they return
463 * 0, you should not assume that it is safe to skip a write of the object (it
464 * either does not exist on disk, or has a stale mtime and may be subject to
465 * pruning).
466 */
467static int check_and_freshen_file(const char *fn, int freshen)
468{
469 if (access(fn, F_OK))
470 return 0;
471 if (freshen && !freshen_file(fn))
472 return 0;
473 return 1;
474}
475
476static int check_and_freshen_local(const unsigned char *sha1, int freshen)
477{
478 return check_and_freshen_file(sha1_file_name(sha1), freshen);
479}
480
481static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
482{
483 struct alternate_object_database *alt;
484 prepare_alt_odb();
485 for (alt = alt_odb_list; alt; alt = alt->next) {
486 fill_sha1_path(alt->name, sha1);
487 if (check_and_freshen_file(alt->base, freshen))
488 return 1;
489 }
490 return 0;
491}
492
493static int check_and_freshen(const unsigned char *sha1, int freshen)
494{
495 return check_and_freshen_local(sha1, freshen) ||
496 check_and_freshen_nonlocal(sha1, freshen);
497}
498
499int has_loose_object_nonlocal(const unsigned char *sha1)
500{
501 return check_and_freshen_nonlocal(sha1, 0);
502}
503
504static int has_loose_object(const unsigned char *sha1)
505{
506 return check_and_freshen(sha1, 0);
507}
508
509static unsigned int pack_used_ctr;
510static unsigned int pack_mmap_calls;
511static unsigned int peak_pack_open_windows;
512static unsigned int pack_open_windows;
513static unsigned int pack_open_fds;
514static unsigned int pack_max_fds;
515static size_t peak_pack_mapped;
516static size_t pack_mapped;
517struct packed_git *packed_git;
518
519static struct mru packed_git_mru_storage;
520struct mru *packed_git_mru = &packed_git_mru_storage;
521
522void pack_report(void)
523{
524 fprintf(stderr,
525 "pack_report: getpagesize() = %10" SZ_FMT "\n"
526 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
527 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
528 sz_fmt(getpagesize()),
529 sz_fmt(packed_git_window_size),
530 sz_fmt(packed_git_limit));
531 fprintf(stderr,
532 "pack_report: pack_used_ctr = %10u\n"
533 "pack_report: pack_mmap_calls = %10u\n"
534 "pack_report: pack_open_windows = %10u / %10u\n"
535 "pack_report: pack_mapped = "
536 "%10" SZ_FMT " / %10" SZ_FMT "\n",
537 pack_used_ctr,
538 pack_mmap_calls,
539 pack_open_windows, peak_pack_open_windows,
540 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
541}
542
543/*
544 * Open and mmap the index file at path, perform a couple of
545 * consistency checks, then record its information to p. Return 0 on
546 * success.
547 */
548static int check_packed_git_idx(const char *path, struct packed_git *p)
549{
550 void *idx_map;
551 struct pack_idx_header *hdr;
552 size_t idx_size;
553 uint32_t version, nr, i, *index;
554 int fd = git_open_noatime(path);
555 struct stat st;
556
557 if (fd < 0)
558 return -1;
559 if (fstat(fd, &st)) {
560 close(fd);
561 return -1;
562 }
563 idx_size = xsize_t(st.st_size);
564 if (idx_size < 4 * 256 + 20 + 20) {
565 close(fd);
566 return error("index file %s is too small", path);
567 }
568 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
569 close(fd);
570
571 hdr = idx_map;
572 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
573 version = ntohl(hdr->idx_version);
574 if (version < 2 || version > 2) {
575 munmap(idx_map, idx_size);
576 return error("index file %s is version %"PRIu32
577 " and is not supported by this binary"
578 " (try upgrading GIT to a newer version)",
579 path, version);
580 }
581 } else
582 version = 1;
583
584 nr = 0;
585 index = idx_map;
586 if (version > 1)
587 index += 2; /* skip index header */
588 for (i = 0; i < 256; i++) {
589 uint32_t n = ntohl(index[i]);
590 if (n < nr) {
591 munmap(idx_map, idx_size);
592 return error("non-monotonic index %s", path);
593 }
594 nr = n;
595 }
596
597 if (version == 1) {
598 /*
599 * Total size:
600 * - 256 index entries 4 bytes each
601 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
602 * - 20-byte SHA1 of the packfile
603 * - 20-byte SHA1 file checksum
604 */
605 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
606 munmap(idx_map, idx_size);
607 return error("wrong index v1 file size in %s", path);
608 }
609 } else if (version == 2) {
610 /*
611 * Minimum size:
612 * - 8 bytes of header
613 * - 256 index entries 4 bytes each
614 * - 20-byte sha1 entry * nr
615 * - 4-byte crc entry * nr
616 * - 4-byte offset entry * nr
617 * - 20-byte SHA1 of the packfile
618 * - 20-byte SHA1 file checksum
619 * And after the 4-byte offset table might be a
620 * variable sized table containing 8-byte entries
621 * for offsets larger than 2^31.
622 */
623 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
624 unsigned long max_size = min_size;
625 if (nr)
626 max_size += (nr - 1)*8;
627 if (idx_size < min_size || idx_size > max_size) {
628 munmap(idx_map, idx_size);
629 return error("wrong index v2 file size in %s", path);
630 }
631 if (idx_size != min_size &&
632 /*
633 * make sure we can deal with large pack offsets.
634 * 31-bit signed offset won't be enough, neither
635 * 32-bit unsigned one will be.
636 */
637 (sizeof(off_t) <= 4)) {
638 munmap(idx_map, idx_size);
639 return error("pack too large for current definition of off_t in %s", path);
640 }
641 }
642
643 p->index_version = version;
644 p->index_data = idx_map;
645 p->index_size = idx_size;
646 p->num_objects = nr;
647 return 0;
648}
649
650int open_pack_index(struct packed_git *p)
651{
652 char *idx_name;
653 size_t len;
654 int ret;
655
656 if (p->index_data)
657 return 0;
658
659 if (!strip_suffix(p->pack_name, ".pack", &len))
660 die("BUG: pack_name does not end in .pack");
661 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
662 ret = check_packed_git_idx(idx_name, p);
663 free(idx_name);
664 return ret;
665}
666
667static void scan_windows(struct packed_git *p,
668 struct packed_git **lru_p,
669 struct pack_window **lru_w,
670 struct pack_window **lru_l)
671{
672 struct pack_window *w, *w_l;
673
674 for (w_l = NULL, w = p->windows; w; w = w->next) {
675 if (!w->inuse_cnt) {
676 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
677 *lru_p = p;
678 *lru_w = w;
679 *lru_l = w_l;
680 }
681 }
682 w_l = w;
683 }
684}
685
686static int unuse_one_window(struct packed_git *current)
687{
688 struct packed_git *p, *lru_p = NULL;
689 struct pack_window *lru_w = NULL, *lru_l = NULL;
690
691 if (current)
692 scan_windows(current, &lru_p, &lru_w, &lru_l);
693 for (p = packed_git; p; p = p->next)
694 scan_windows(p, &lru_p, &lru_w, &lru_l);
695 if (lru_p) {
696 munmap(lru_w->base, lru_w->len);
697 pack_mapped -= lru_w->len;
698 if (lru_l)
699 lru_l->next = lru_w->next;
700 else
701 lru_p->windows = lru_w->next;
702 free(lru_w);
703 pack_open_windows--;
704 return 1;
705 }
706 return 0;
707}
708
709void release_pack_memory(size_t need)
710{
711 size_t cur = pack_mapped;
712 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
713 ; /* nothing */
714}
715
716static void mmap_limit_check(size_t length)
717{
718 static size_t limit = 0;
719 if (!limit) {
720 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
721 if (!limit)
722 limit = SIZE_MAX;
723 }
724 if (length > limit)
725 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
726 (uintmax_t)length, (uintmax_t)limit);
727}
728
729void *xmmap_gently(void *start, size_t length,
730 int prot, int flags, int fd, off_t offset)
731{
732 void *ret;
733
734 mmap_limit_check(length);
735 ret = mmap(start, length, prot, flags, fd, offset);
736 if (ret == MAP_FAILED) {
737 if (!length)
738 return NULL;
739 release_pack_memory(length);
740 ret = mmap(start, length, prot, flags, fd, offset);
741 }
742 return ret;
743}
744
745void *xmmap(void *start, size_t length,
746 int prot, int flags, int fd, off_t offset)
747{
748 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
749 if (ret == MAP_FAILED)
750 die_errno("mmap failed");
751 return ret;
752}
753
754void close_pack_windows(struct packed_git *p)
755{
756 while (p->windows) {
757 struct pack_window *w = p->windows;
758
759 if (w->inuse_cnt)
760 die("pack '%s' still has open windows to it",
761 p->pack_name);
762 munmap(w->base, w->len);
763 pack_mapped -= w->len;
764 pack_open_windows--;
765 p->windows = w->next;
766 free(w);
767 }
768}
769
770static int close_pack_fd(struct packed_git *p)
771{
772 if (p->pack_fd < 0)
773 return 0;
774
775 close(p->pack_fd);
776 pack_open_fds--;
777 p->pack_fd = -1;
778
779 return 1;
780}
781
782static void close_pack(struct packed_git *p)
783{
784 close_pack_windows(p);
785 close_pack_fd(p);
786 close_pack_index(p);
787}
788
789void close_all_packs(void)
790{
791 struct packed_git *p;
792
793 for (p = packed_git; p; p = p->next)
794 if (p->do_not_close)
795 die("BUG: want to close pack marked 'do-not-close'");
796 else
797 close_pack(p);
798}
799
800
801/*
802 * The LRU pack is the one with the oldest MRU window, preferring packs
803 * with no used windows, or the oldest mtime if it has no windows allocated.
804 */
805static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
806{
807 struct pack_window *w, *this_mru_w;
808 int has_windows_inuse = 0;
809
810 /*
811 * Reject this pack if it has windows and the previously selected
812 * one does not. If this pack does not have windows, reject
813 * it if the pack file is newer than the previously selected one.
814 */
815 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
816 return;
817
818 for (w = this_mru_w = p->windows; w; w = w->next) {
819 /*
820 * Reject this pack if any of its windows are in use,
821 * but the previously selected pack did not have any
822 * inuse windows. Otherwise, record that this pack
823 * has windows in use.
824 */
825 if (w->inuse_cnt) {
826 if (*accept_windows_inuse)
827 has_windows_inuse = 1;
828 else
829 return;
830 }
831
832 if (w->last_used > this_mru_w->last_used)
833 this_mru_w = w;
834
835 /*
836 * Reject this pack if it has windows that have been
837 * used more recently than the previously selected pack.
838 * If the previously selected pack had windows inuse and
839 * we have not encountered a window in this pack that is
840 * inuse, skip this check since we prefer a pack with no
841 * inuse windows to one that has inuse windows.
842 */
843 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
844 this_mru_w->last_used > (*mru_w)->last_used)
845 return;
846 }
847
848 /*
849 * Select this pack.
850 */
851 *mru_w = this_mru_w;
852 *lru_p = p;
853 *accept_windows_inuse = has_windows_inuse;
854}
855
856static int close_one_pack(void)
857{
858 struct packed_git *p, *lru_p = NULL;
859 struct pack_window *mru_w = NULL;
860 int accept_windows_inuse = 1;
861
862 for (p = packed_git; p; p = p->next) {
863 if (p->pack_fd == -1)
864 continue;
865 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
866 }
867
868 if (lru_p)
869 return close_pack_fd(lru_p);
870
871 return 0;
872}
873
874void unuse_pack(struct pack_window **w_cursor)
875{
876 struct pack_window *w = *w_cursor;
877 if (w) {
878 w->inuse_cnt--;
879 *w_cursor = NULL;
880 }
881}
882
883void close_pack_index(struct packed_git *p)
884{
885 if (p->index_data) {
886 munmap((void *)p->index_data, p->index_size);
887 p->index_data = NULL;
888 }
889}
890
891static unsigned int get_max_fd_limit(void)
892{
893#ifdef RLIMIT_NOFILE
894 {
895 struct rlimit lim;
896
897 if (!getrlimit(RLIMIT_NOFILE, &lim))
898 return lim.rlim_cur;
899 }
900#endif
901
902#ifdef _SC_OPEN_MAX
903 {
904 long open_max = sysconf(_SC_OPEN_MAX);
905 if (0 < open_max)
906 return open_max;
907 /*
908 * Otherwise, we got -1 for one of the two
909 * reasons:
910 *
911 * (1) sysconf() did not understand _SC_OPEN_MAX
912 * and signaled an error with -1; or
913 * (2) sysconf() said there is no limit.
914 *
915 * We _could_ clear errno before calling sysconf() to
916 * tell these two cases apart and return a huge number
917 * in the latter case to let the caller cap it to a
918 * value that is not so selfish, but letting the
919 * fallback OPEN_MAX codepath take care of these cases
920 * is a lot simpler.
921 */
922 }
923#endif
924
925#ifdef OPEN_MAX
926 return OPEN_MAX;
927#else
928 return 1; /* see the caller ;-) */
929#endif
930}
931
932/*
933 * Do not call this directly as this leaks p->pack_fd on error return;
934 * call open_packed_git() instead.
935 */
936static int open_packed_git_1(struct packed_git *p)
937{
938 struct stat st;
939 struct pack_header hdr;
940 unsigned char sha1[20];
941 unsigned char *idx_sha1;
942 long fd_flag;
943
944 if (!p->index_data && open_pack_index(p))
945 return error("packfile %s index unavailable", p->pack_name);
946
947 if (!pack_max_fds) {
948 unsigned int max_fds = get_max_fd_limit();
949
950 /* Save 3 for stdin/stdout/stderr, 22 for work */
951 if (25 < max_fds)
952 pack_max_fds = max_fds - 25;
953 else
954 pack_max_fds = 1;
955 }
956
957 while (pack_max_fds <= pack_open_fds && close_one_pack())
958 ; /* nothing */
959
960 p->pack_fd = git_open_noatime(p->pack_name);
961 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
962 return -1;
963 pack_open_fds++;
964
965 /* If we created the struct before we had the pack we lack size. */
966 if (!p->pack_size) {
967 if (!S_ISREG(st.st_mode))
968 return error("packfile %s not a regular file", p->pack_name);
969 p->pack_size = st.st_size;
970 } else if (p->pack_size != st.st_size)
971 return error("packfile %s size changed", p->pack_name);
972
973 /* We leave these file descriptors open with sliding mmap;
974 * there is no point keeping them open across exec(), though.
975 */
976 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
977 if (fd_flag < 0)
978 return error("cannot determine file descriptor flags");
979 fd_flag |= FD_CLOEXEC;
980 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
981 return error("cannot set FD_CLOEXEC");
982
983 /* Verify we recognize this pack file format. */
984 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
985 return error("file %s is far too short to be a packfile", p->pack_name);
986 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
987 return error("file %s is not a GIT packfile", p->pack_name);
988 if (!pack_version_ok(hdr.hdr_version))
989 return error("packfile %s is version %"PRIu32" and not"
990 " supported (try upgrading GIT to a newer version)",
991 p->pack_name, ntohl(hdr.hdr_version));
992
993 /* Verify the pack matches its index. */
994 if (p->num_objects != ntohl(hdr.hdr_entries))
995 return error("packfile %s claims to have %"PRIu32" objects"
996 " while index indicates %"PRIu32" objects",
997 p->pack_name, ntohl(hdr.hdr_entries),
998 p->num_objects);
999 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1000 return error("end of packfile %s is unavailable", p->pack_name);
1001 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1002 return error("packfile %s signature is unavailable", p->pack_name);
1003 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1004 if (hashcmp(sha1, idx_sha1))
1005 return error("packfile %s does not match index", p->pack_name);
1006 return 0;
1007}
1008
1009static int open_packed_git(struct packed_git *p)
1010{
1011 if (!open_packed_git_1(p))
1012 return 0;
1013 close_pack_fd(p);
1014 return -1;
1015}
1016
1017static int in_window(struct pack_window *win, off_t offset)
1018{
1019 /* We must promise at least 20 bytes (one hash) after the
1020 * offset is available from this window, otherwise the offset
1021 * is not actually in this window and a different window (which
1022 * has that one hash excess) must be used. This is to support
1023 * the object header and delta base parsing routines below.
1024 */
1025 off_t win_off = win->offset;
1026 return win_off <= offset
1027 && (offset + 20) <= (win_off + win->len);
1028}
1029
1030unsigned char *use_pack(struct packed_git *p,
1031 struct pack_window **w_cursor,
1032 off_t offset,
1033 unsigned long *left)
1034{
1035 struct pack_window *win = *w_cursor;
1036
1037 /* Since packfiles end in a hash of their content and it's
1038 * pointless to ask for an offset into the middle of that
1039 * hash, and the in_window function above wouldn't match
1040 * don't allow an offset too close to the end of the file.
1041 */
1042 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1043 die("packfile %s cannot be accessed", p->pack_name);
1044 if (offset > (p->pack_size - 20))
1045 die("offset beyond end of packfile (truncated pack?)");
1046 if (offset < 0)
1047 die(_("offset before end of packfile (broken .idx?)"));
1048
1049 if (!win || !in_window(win, offset)) {
1050 if (win)
1051 win->inuse_cnt--;
1052 for (win = p->windows; win; win = win->next) {
1053 if (in_window(win, offset))
1054 break;
1055 }
1056 if (!win) {
1057 size_t window_align = packed_git_window_size / 2;
1058 off_t len;
1059
1060 if (p->pack_fd == -1 && open_packed_git(p))
1061 die("packfile %s cannot be accessed", p->pack_name);
1062
1063 win = xcalloc(1, sizeof(*win));
1064 win->offset = (offset / window_align) * window_align;
1065 len = p->pack_size - win->offset;
1066 if (len > packed_git_window_size)
1067 len = packed_git_window_size;
1068 win->len = (size_t)len;
1069 pack_mapped += win->len;
1070 while (packed_git_limit < pack_mapped
1071 && unuse_one_window(p))
1072 ; /* nothing */
1073 win->base = xmmap(NULL, win->len,
1074 PROT_READ, MAP_PRIVATE,
1075 p->pack_fd, win->offset);
1076 if (win->base == MAP_FAILED)
1077 die_errno("packfile %s cannot be mapped",
1078 p->pack_name);
1079 if (!win->offset && win->len == p->pack_size
1080 && !p->do_not_close)
1081 close_pack_fd(p);
1082 pack_mmap_calls++;
1083 pack_open_windows++;
1084 if (pack_mapped > peak_pack_mapped)
1085 peak_pack_mapped = pack_mapped;
1086 if (pack_open_windows > peak_pack_open_windows)
1087 peak_pack_open_windows = pack_open_windows;
1088 win->next = p->windows;
1089 p->windows = win;
1090 }
1091 }
1092 if (win != *w_cursor) {
1093 win->last_used = pack_used_ctr++;
1094 win->inuse_cnt++;
1095 *w_cursor = win;
1096 }
1097 offset -= win->offset;
1098 if (left)
1099 *left = win->len - xsize_t(offset);
1100 return win->base + offset;
1101}
1102
1103static struct packed_git *alloc_packed_git(int extra)
1104{
1105 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1106 memset(p, 0, sizeof(*p));
1107 p->pack_fd = -1;
1108 return p;
1109}
1110
1111static void try_to_free_pack_memory(size_t size)
1112{
1113 release_pack_memory(size);
1114}
1115
1116struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1117{
1118 static int have_set_try_to_free_routine;
1119 struct stat st;
1120 size_t alloc;
1121 struct packed_git *p;
1122
1123 if (!have_set_try_to_free_routine) {
1124 have_set_try_to_free_routine = 1;
1125 set_try_to_free_routine(try_to_free_pack_memory);
1126 }
1127
1128 /*
1129 * Make sure a corresponding .pack file exists and that
1130 * the index looks sane.
1131 */
1132 if (!strip_suffix_mem(path, &path_len, ".idx"))
1133 return NULL;
1134
1135 /*
1136 * ".pack" is long enough to hold any suffix we're adding (and
1137 * the use xsnprintf double-checks that)
1138 */
1139 alloc = st_add3(path_len, strlen(".pack"), 1);
1140 p = alloc_packed_git(alloc);
1141 memcpy(p->pack_name, path, path_len);
1142
1143 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1144 if (!access(p->pack_name, F_OK))
1145 p->pack_keep = 1;
1146
1147 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1148 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1149 free(p);
1150 return NULL;
1151 }
1152
1153 /* ok, it looks sane as far as we can check without
1154 * actually mapping the pack file.
1155 */
1156 p->pack_size = st.st_size;
1157 p->pack_local = local;
1158 p->mtime = st.st_mtime;
1159 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1160 hashclr(p->sha1);
1161 return p;
1162}
1163
1164struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1165{
1166 const char *path = sha1_pack_name(sha1);
1167 size_t alloc = st_add(strlen(path), 1);
1168 struct packed_git *p = alloc_packed_git(alloc);
1169
1170 memcpy(p->pack_name, path, alloc); /* includes NUL */
1171 hashcpy(p->sha1, sha1);
1172 if (check_packed_git_idx(idx_path, p)) {
1173 free(p);
1174 return NULL;
1175 }
1176
1177 return p;
1178}
1179
1180void install_packed_git(struct packed_git *pack)
1181{
1182 if (pack->pack_fd != -1)
1183 pack_open_fds++;
1184
1185 pack->next = packed_git;
1186 packed_git = pack;
1187}
1188
1189void (*report_garbage)(unsigned seen_bits, const char *path);
1190
1191static void report_helper(const struct string_list *list,
1192 int seen_bits, int first, int last)
1193{
1194 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1195 return;
1196
1197 for (; first < last; first++)
1198 report_garbage(seen_bits, list->items[first].string);
1199}
1200
1201static 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(PACKDIR_FILE_GARBAGE, 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
1235static 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_errno("unable to open object pack directory: %s",
1249 path.buf);
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(PACKDIR_FILE_GARBAGE, path.buf);
1294 }
1295 closedir(dir);
1296 report_pack_garbage(&garbage);
1297 string_list_clear(&garbage, 0);
1298 strbuf_release(&path);
1299}
1300
1301static 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
1329static 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
1355static void prepare_packed_git_mru(void)
1356{
1357 struct packed_git *p;
1358
1359 mru_clear(packed_git_mru);
1360 for (p = packed_git; p; p = p->next)
1361 mru_append(packed_git_mru, p);
1362}
1363
1364static int prepare_packed_git_run_once = 0;
1365void prepare_packed_git(void)
1366{
1367 struct alternate_object_database *alt;
1368
1369 if (prepare_packed_git_run_once)
1370 return;
1371 prepare_packed_git_one(get_object_directory(), 1);
1372 prepare_alt_odb();
1373 for (alt = alt_odb_list; alt; alt = alt->next) {
1374 alt->name[-1] = 0;
1375 prepare_packed_git_one(alt->base, 0);
1376 alt->name[-1] = '/';
1377 }
1378 rearrange_packed_git();
1379 prepare_packed_git_mru();
1380 prepare_packed_git_run_once = 1;
1381}
1382
1383void reprepare_packed_git(void)
1384{
1385 prepare_packed_git_run_once = 0;
1386 prepare_packed_git();
1387}
1388
1389static void mark_bad_packed_object(struct packed_git *p,
1390 const unsigned char *sha1)
1391{
1392 unsigned i;
1393 for (i = 0; i < p->num_bad_objects; i++)
1394 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1395 return;
1396 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1397 st_mult(GIT_SHA1_RAWSZ,
1398 st_add(p->num_bad_objects, 1)));
1399 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1400 p->num_bad_objects++;
1401}
1402
1403static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1404{
1405 struct packed_git *p;
1406 unsigned i;
1407
1408 for (p = packed_git; p; p = p->next)
1409 for (i = 0; i < p->num_bad_objects; i++)
1410 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1411 return p;
1412 return NULL;
1413}
1414
1415/*
1416 * With an in-core object data in "map", rehash it to make sure the
1417 * object name actually matches "sha1" to detect object corruption.
1418 * With "map" == NULL, try reading the object named with "sha1" using
1419 * the streaming interface and rehash it to do the same.
1420 */
1421int check_sha1_signature(const unsigned char *sha1, void *map,
1422 unsigned long size, const char *type)
1423{
1424 unsigned char real_sha1[20];
1425 enum object_type obj_type;
1426 struct git_istream *st;
1427 git_SHA_CTX c;
1428 char hdr[32];
1429 int hdrlen;
1430
1431 if (map) {
1432 hash_sha1_file(map, size, type, real_sha1);
1433 return hashcmp(sha1, real_sha1) ? -1 : 0;
1434 }
1435
1436 st = open_istream(sha1, &obj_type, &size, NULL);
1437 if (!st)
1438 return -1;
1439
1440 /* Generate the header */
1441 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1442
1443 /* Sha1.. */
1444 git_SHA1_Init(&c);
1445 git_SHA1_Update(&c, hdr, hdrlen);
1446 for (;;) {
1447 char buf[1024 * 16];
1448 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1449
1450 if (readlen < 0) {
1451 close_istream(st);
1452 return -1;
1453 }
1454 if (!readlen)
1455 break;
1456 git_SHA1_Update(&c, buf, readlen);
1457 }
1458 git_SHA1_Final(real_sha1, &c);
1459 close_istream(st);
1460 return hashcmp(sha1, real_sha1) ? -1 : 0;
1461}
1462
1463int git_open_noatime(const char *name)
1464{
1465 static int sha1_file_open_flag = O_NOATIME;
1466
1467 for (;;) {
1468 int fd;
1469
1470 errno = 0;
1471 fd = open(name, O_RDONLY | sha1_file_open_flag);
1472 if (fd >= 0)
1473 return fd;
1474
1475 /* Might the failure be due to O_NOATIME? */
1476 if (errno != ENOENT && sha1_file_open_flag) {
1477 sha1_file_open_flag = 0;
1478 continue;
1479 }
1480
1481 return -1;
1482 }
1483}
1484
1485static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1486{
1487 struct alternate_object_database *alt;
1488
1489 if (!lstat(sha1_file_name(sha1), st))
1490 return 0;
1491
1492 prepare_alt_odb();
1493 errno = ENOENT;
1494 for (alt = alt_odb_list; alt; alt = alt->next) {
1495 fill_sha1_path(alt->name, sha1);
1496 if (!lstat(alt->base, st))
1497 return 0;
1498 }
1499
1500 return -1;
1501}
1502
1503static int open_sha1_file(const unsigned char *sha1)
1504{
1505 int fd;
1506 struct alternate_object_database *alt;
1507 int most_interesting_errno;
1508
1509 fd = git_open_noatime(sha1_file_name(sha1));
1510 if (fd >= 0)
1511 return fd;
1512 most_interesting_errno = errno;
1513
1514 prepare_alt_odb();
1515 for (alt = alt_odb_list; alt; alt = alt->next) {
1516 fill_sha1_path(alt->name, sha1);
1517 fd = git_open_noatime(alt->base);
1518 if (fd >= 0)
1519 return fd;
1520 if (most_interesting_errno == ENOENT)
1521 most_interesting_errno = errno;
1522 }
1523 errno = most_interesting_errno;
1524 return -1;
1525}
1526
1527void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1528{
1529 void *map;
1530 int fd;
1531
1532 fd = open_sha1_file(sha1);
1533 map = NULL;
1534 if (fd >= 0) {
1535 struct stat st;
1536
1537 if (!fstat(fd, &st)) {
1538 *size = xsize_t(st.st_size);
1539 if (!*size) {
1540 /* mmap() is forbidden on empty files */
1541 error("object file %s is empty", sha1_file_name(sha1));
1542 return NULL;
1543 }
1544 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1545 }
1546 close(fd);
1547 }
1548 return map;
1549}
1550
1551unsigned long unpack_object_header_buffer(const unsigned char *buf,
1552 unsigned long len, enum object_type *type, unsigned long *sizep)
1553{
1554 unsigned shift;
1555 unsigned long size, c;
1556 unsigned long used = 0;
1557
1558 c = buf[used++];
1559 *type = (c >> 4) & 7;
1560 size = c & 15;
1561 shift = 4;
1562 while (c & 0x80) {
1563 if (len <= used || bitsizeof(long) <= shift) {
1564 error("bad object header");
1565 size = used = 0;
1566 break;
1567 }
1568 c = buf[used++];
1569 size += (c & 0x7f) << shift;
1570 shift += 7;
1571 }
1572 *sizep = size;
1573 return used;
1574}
1575
1576int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1577{
1578 /* Get the data stream */
1579 memset(stream, 0, sizeof(*stream));
1580 stream->next_in = map;
1581 stream->avail_in = mapsize;
1582 stream->next_out = buffer;
1583 stream->avail_out = bufsiz;
1584
1585 git_inflate_init(stream);
1586 return git_inflate(stream, 0);
1587}
1588
1589static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1590 unsigned long mapsize, void *buffer,
1591 unsigned long bufsiz, struct strbuf *header)
1592{
1593 int status;
1594
1595 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1596
1597 /*
1598 * Check if entire header is unpacked in the first iteration.
1599 */
1600 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1601 return 0;
1602
1603 /*
1604 * buffer[0..bufsiz] was not large enough. Copy the partial
1605 * result out to header, and then append the result of further
1606 * reading the stream.
1607 */
1608 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1609 stream->next_out = buffer;
1610 stream->avail_out = bufsiz;
1611
1612 do {
1613 status = git_inflate(stream, 0);
1614 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1615 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1616 return 0;
1617 stream->next_out = buffer;
1618 stream->avail_out = bufsiz;
1619 } while (status != Z_STREAM_END);
1620 return -1;
1621}
1622
1623static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1624{
1625 int bytes = strlen(buffer) + 1;
1626 unsigned char *buf = xmallocz(size);
1627 unsigned long n;
1628 int status = Z_OK;
1629
1630 n = stream->total_out - bytes;
1631 if (n > size)
1632 n = size;
1633 memcpy(buf, (char *) buffer + bytes, n);
1634 bytes = n;
1635 if (bytes <= size) {
1636 /*
1637 * The above condition must be (bytes <= size), not
1638 * (bytes < size). In other words, even though we
1639 * expect no more output and set avail_out to zero,
1640 * the input zlib stream may have bytes that express
1641 * "this concludes the stream", and we *do* want to
1642 * eat that input.
1643 *
1644 * Otherwise we would not be able to test that we
1645 * consumed all the input to reach the expected size;
1646 * we also want to check that zlib tells us that all
1647 * went well with status == Z_STREAM_END at the end.
1648 */
1649 stream->next_out = buf + bytes;
1650 stream->avail_out = size - bytes;
1651 while (status == Z_OK)
1652 status = git_inflate(stream, Z_FINISH);
1653 }
1654 if (status == Z_STREAM_END && !stream->avail_in) {
1655 git_inflate_end(stream);
1656 return buf;
1657 }
1658
1659 if (status < 0)
1660 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1661 else if (stream->avail_in)
1662 error("garbage at end of loose object '%s'",
1663 sha1_to_hex(sha1));
1664 free(buf);
1665 return NULL;
1666}
1667
1668/*
1669 * We used to just use "sscanf()", but that's actually way
1670 * too permissive for what we want to check. So do an anal
1671 * object header parse by hand.
1672 */
1673static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1674 unsigned int flags)
1675{
1676 const char *type_buf = hdr;
1677 unsigned long size;
1678 int type, type_len = 0;
1679
1680 /*
1681 * The type can be of any size but is followed by
1682 * a space.
1683 */
1684 for (;;) {
1685 char c = *hdr++;
1686 if (c == ' ')
1687 break;
1688 type_len++;
1689 }
1690
1691 type = type_from_string_gently(type_buf, type_len, 1);
1692 if (oi->typename)
1693 strbuf_add(oi->typename, type_buf, type_len);
1694 /*
1695 * Set type to 0 if its an unknown object and
1696 * we're obtaining the type using '--allow-unknown-type'
1697 * option.
1698 */
1699 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1700 type = 0;
1701 else if (type < 0)
1702 die("invalid object type");
1703 if (oi->typep)
1704 *oi->typep = type;
1705
1706 /*
1707 * The length must follow immediately, and be in canonical
1708 * decimal format (ie "010" is not valid).
1709 */
1710 size = *hdr++ - '0';
1711 if (size > 9)
1712 return -1;
1713 if (size) {
1714 for (;;) {
1715 unsigned long c = *hdr - '0';
1716 if (c > 9)
1717 break;
1718 hdr++;
1719 size = size * 10 + c;
1720 }
1721 }
1722
1723 if (oi->sizep)
1724 *oi->sizep = size;
1725
1726 /*
1727 * The length must be followed by a zero byte
1728 */
1729 return *hdr ? -1 : type;
1730}
1731
1732int parse_sha1_header(const char *hdr, unsigned long *sizep)
1733{
1734 struct object_info oi;
1735
1736 oi.sizep = sizep;
1737 oi.typename = NULL;
1738 oi.typep = NULL;
1739 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1740}
1741
1742static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1743{
1744 int ret;
1745 git_zstream stream;
1746 char hdr[8192];
1747
1748 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1749 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1750 return NULL;
1751
1752 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1753}
1754
1755unsigned long get_size_from_delta(struct packed_git *p,
1756 struct pack_window **w_curs,
1757 off_t curpos)
1758{
1759 const unsigned char *data;
1760 unsigned char delta_head[20], *in;
1761 git_zstream stream;
1762 int st;
1763
1764 memset(&stream, 0, sizeof(stream));
1765 stream.next_out = delta_head;
1766 stream.avail_out = sizeof(delta_head);
1767
1768 git_inflate_init(&stream);
1769 do {
1770 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1771 stream.next_in = in;
1772 st = git_inflate(&stream, Z_FINISH);
1773 curpos += stream.next_in - in;
1774 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1775 stream.total_out < sizeof(delta_head));
1776 git_inflate_end(&stream);
1777 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1778 error("delta data unpack-initial failed");
1779 return 0;
1780 }
1781
1782 /* Examine the initial part of the delta to figure out
1783 * the result size.
1784 */
1785 data = delta_head;
1786
1787 /* ignore base size */
1788 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1789
1790 /* Read the result size */
1791 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1792}
1793
1794static off_t get_delta_base(struct packed_git *p,
1795 struct pack_window **w_curs,
1796 off_t *curpos,
1797 enum object_type type,
1798 off_t delta_obj_offset)
1799{
1800 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1801 off_t base_offset;
1802
1803 /* use_pack() assured us we have [base_info, base_info + 20)
1804 * as a range that we can look at without walking off the
1805 * end of the mapped window. Its actually the hash size
1806 * that is assured. An OFS_DELTA longer than the hash size
1807 * is stupid, as then a REF_DELTA would be smaller to store.
1808 */
1809 if (type == OBJ_OFS_DELTA) {
1810 unsigned used = 0;
1811 unsigned char c = base_info[used++];
1812 base_offset = c & 127;
1813 while (c & 128) {
1814 base_offset += 1;
1815 if (!base_offset || MSB(base_offset, 7))
1816 return 0; /* overflow */
1817 c = base_info[used++];
1818 base_offset = (base_offset << 7) + (c & 127);
1819 }
1820 base_offset = delta_obj_offset - base_offset;
1821 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1822 return 0; /* out of bound */
1823 *curpos += used;
1824 } else if (type == OBJ_REF_DELTA) {
1825 /* The base entry _must_ be in the same pack */
1826 base_offset = find_pack_entry_one(base_info, p);
1827 *curpos += 20;
1828 } else
1829 die("I am totally screwed");
1830 return base_offset;
1831}
1832
1833/*
1834 * Like get_delta_base above, but we return the sha1 instead of the pack
1835 * offset. This means it is cheaper for REF deltas (we do not have to do
1836 * the final object lookup), but more expensive for OFS deltas (we
1837 * have to load the revidx to convert the offset back into a sha1).
1838 */
1839static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1840 struct pack_window **w_curs,
1841 off_t curpos,
1842 enum object_type type,
1843 off_t delta_obj_offset)
1844{
1845 if (type == OBJ_REF_DELTA) {
1846 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1847 return base;
1848 } else if (type == OBJ_OFS_DELTA) {
1849 struct revindex_entry *revidx;
1850 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1851 type, delta_obj_offset);
1852
1853 if (!base_offset)
1854 return NULL;
1855
1856 revidx = find_pack_revindex(p, base_offset);
1857 if (!revidx)
1858 return NULL;
1859
1860 return nth_packed_object_sha1(p, revidx->nr);
1861 } else
1862 return NULL;
1863}
1864
1865int unpack_object_header(struct packed_git *p,
1866 struct pack_window **w_curs,
1867 off_t *curpos,
1868 unsigned long *sizep)
1869{
1870 unsigned char *base;
1871 unsigned long left;
1872 unsigned long used;
1873 enum object_type type;
1874
1875 /* use_pack() assures us we have [base, base + 20) available
1876 * as a range that we can look at. (Its actually the hash
1877 * size that is assured.) With our object header encoding
1878 * the maximum deflated object size is 2^137, which is just
1879 * insane, so we know won't exceed what we have been given.
1880 */
1881 base = use_pack(p, w_curs, *curpos, &left);
1882 used = unpack_object_header_buffer(base, left, &type, sizep);
1883 if (!used) {
1884 type = OBJ_BAD;
1885 } else
1886 *curpos += used;
1887
1888 return type;
1889}
1890
1891static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1892{
1893 int type;
1894 struct revindex_entry *revidx;
1895 const unsigned char *sha1;
1896 revidx = find_pack_revindex(p, obj_offset);
1897 if (!revidx)
1898 return OBJ_BAD;
1899 sha1 = nth_packed_object_sha1(p, revidx->nr);
1900 mark_bad_packed_object(p, sha1);
1901 type = sha1_object_info(sha1, NULL);
1902 if (type <= OBJ_NONE)
1903 return OBJ_BAD;
1904 return type;
1905}
1906
1907#define POI_STACK_PREALLOC 64
1908
1909static enum object_type packed_to_object_type(struct packed_git *p,
1910 off_t obj_offset,
1911 enum object_type type,
1912 struct pack_window **w_curs,
1913 off_t curpos)
1914{
1915 off_t small_poi_stack[POI_STACK_PREALLOC];
1916 off_t *poi_stack = small_poi_stack;
1917 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1918
1919 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1920 off_t base_offset;
1921 unsigned long size;
1922 /* Push the object we're going to leave behind */
1923 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1924 poi_stack_alloc = alloc_nr(poi_stack_nr);
1925 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1926 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1927 } else {
1928 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1929 }
1930 poi_stack[poi_stack_nr++] = obj_offset;
1931 /* If parsing the base offset fails, just unwind */
1932 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1933 if (!base_offset)
1934 goto unwind;
1935 curpos = obj_offset = base_offset;
1936 type = unpack_object_header(p, w_curs, &curpos, &size);
1937 if (type <= OBJ_NONE) {
1938 /* If getting the base itself fails, we first
1939 * retry the base, otherwise unwind */
1940 type = retry_bad_packed_offset(p, base_offset);
1941 if (type > OBJ_NONE)
1942 goto out;
1943 goto unwind;
1944 }
1945 }
1946
1947 switch (type) {
1948 case OBJ_BAD:
1949 case OBJ_COMMIT:
1950 case OBJ_TREE:
1951 case OBJ_BLOB:
1952 case OBJ_TAG:
1953 break;
1954 default:
1955 error("unknown object type %i at offset %"PRIuMAX" in %s",
1956 type, (uintmax_t)obj_offset, p->pack_name);
1957 type = OBJ_BAD;
1958 }
1959
1960out:
1961 if (poi_stack != small_poi_stack)
1962 free(poi_stack);
1963 return type;
1964
1965unwind:
1966 while (poi_stack_nr) {
1967 obj_offset = poi_stack[--poi_stack_nr];
1968 type = retry_bad_packed_offset(p, obj_offset);
1969 if (type > OBJ_NONE)
1970 goto out;
1971 }
1972 type = OBJ_BAD;
1973 goto out;
1974}
1975
1976static int packed_object_info(struct packed_git *p, off_t obj_offset,
1977 struct object_info *oi)
1978{
1979 struct pack_window *w_curs = NULL;
1980 unsigned long size;
1981 off_t curpos = obj_offset;
1982 enum object_type type;
1983
1984 /*
1985 * We always get the representation type, but only convert it to
1986 * a "real" type later if the caller is interested.
1987 */
1988 type = unpack_object_header(p, &w_curs, &curpos, &size);
1989
1990 if (oi->sizep) {
1991 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1992 off_t tmp_pos = curpos;
1993 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1994 type, obj_offset);
1995 if (!base_offset) {
1996 type = OBJ_BAD;
1997 goto out;
1998 }
1999 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2000 if (*oi->sizep == 0) {
2001 type = OBJ_BAD;
2002 goto out;
2003 }
2004 } else {
2005 *oi->sizep = size;
2006 }
2007 }
2008
2009 if (oi->disk_sizep) {
2010 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2011 *oi->disk_sizep = revidx[1].offset - obj_offset;
2012 }
2013
2014 if (oi->typep) {
2015 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2016 if (*oi->typep < 0) {
2017 type = OBJ_BAD;
2018 goto out;
2019 }
2020 }
2021
2022 if (oi->delta_base_sha1) {
2023 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2024 const unsigned char *base;
2025
2026 base = get_delta_base_sha1(p, &w_curs, curpos,
2027 type, obj_offset);
2028 if (!base) {
2029 type = OBJ_BAD;
2030 goto out;
2031 }
2032
2033 hashcpy(oi->delta_base_sha1, base);
2034 } else
2035 hashclr(oi->delta_base_sha1);
2036 }
2037
2038out:
2039 unuse_pack(&w_curs);
2040 return type;
2041}
2042
2043static void *unpack_compressed_entry(struct packed_git *p,
2044 struct pack_window **w_curs,
2045 off_t curpos,
2046 unsigned long size)
2047{
2048 int st;
2049 git_zstream stream;
2050 unsigned char *buffer, *in;
2051
2052 buffer = xmallocz_gently(size);
2053 if (!buffer)
2054 return NULL;
2055 memset(&stream, 0, sizeof(stream));
2056 stream.next_out = buffer;
2057 stream.avail_out = size + 1;
2058
2059 git_inflate_init(&stream);
2060 do {
2061 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2062 stream.next_in = in;
2063 st = git_inflate(&stream, Z_FINISH);
2064 if (!stream.avail_out)
2065 break; /* the payload is larger than it should be */
2066 curpos += stream.next_in - in;
2067 } while (st == Z_OK || st == Z_BUF_ERROR);
2068 git_inflate_end(&stream);
2069 if ((st != Z_STREAM_END) || stream.total_out != size) {
2070 free(buffer);
2071 return NULL;
2072 }
2073
2074 return buffer;
2075}
2076
2077static struct hashmap delta_base_cache;
2078static size_t delta_base_cached;
2079
2080static LIST_HEAD(delta_base_cache_lru);
2081
2082struct delta_base_cache_key {
2083 struct packed_git *p;
2084 off_t base_offset;
2085};
2086
2087struct delta_base_cache_entry {
2088 struct hashmap hash;
2089 struct delta_base_cache_key key;
2090 struct list_head lru;
2091 void *data;
2092 unsigned long size;
2093 enum object_type type;
2094};
2095
2096static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2097{
2098 unsigned int hash;
2099
2100 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2101 hash += (hash >> 8) + (hash >> 16);
2102 return hash;
2103}
2104
2105static struct delta_base_cache_entry *
2106get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2107{
2108 struct hashmap_entry entry;
2109 struct delta_base_cache_key key;
2110
2111 if (!delta_base_cache.cmpfn)
2112 return NULL;
2113
2114 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2115 key.p = p;
2116 key.base_offset = base_offset;
2117 return hashmap_get(&delta_base_cache, &entry, &key);
2118}
2119
2120static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2121 const struct delta_base_cache_key *b)
2122{
2123 return a->p == b->p && a->base_offset == b->base_offset;
2124}
2125
2126static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2127 const void *vkey)
2128{
2129 const struct delta_base_cache_entry *a = va, *b = vb;
2130 const struct delta_base_cache_key *key = vkey;
2131 if (key)
2132 return !delta_base_cache_key_eq(&a->key, key);
2133 else
2134 return !delta_base_cache_key_eq(&a->key, &b->key);
2135}
2136
2137static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2138{
2139 return !!get_delta_base_cache_entry(p, base_offset);
2140}
2141
2142/*
2143 * Remove the entry from the cache, but do _not_ free the associated
2144 * entry data. The caller takes ownership of the "data" buffer, and
2145 * should copy out any fields it wants before detaching.
2146 */
2147static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2148{
2149 hashmap_remove(&delta_base_cache, ent, &ent->key);
2150 list_del(&ent->lru);
2151 delta_base_cached -= ent->size;
2152 free(ent);
2153}
2154
2155static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2156 unsigned long *base_size, enum object_type *type)
2157{
2158 struct delta_base_cache_entry *ent;
2159
2160 ent = get_delta_base_cache_entry(p, base_offset);
2161 if (!ent)
2162 return unpack_entry(p, base_offset, type, base_size);
2163
2164 *type = ent->type;
2165 *base_size = ent->size;
2166 return xmemdupz(ent->data, ent->size);
2167}
2168
2169static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2170{
2171 free(ent->data);
2172 detach_delta_base_cache_entry(ent);
2173}
2174
2175void clear_delta_base_cache(void)
2176{
2177 struct hashmap_iter iter;
2178 struct delta_base_cache_entry *entry;
2179 for (entry = hashmap_iter_first(&delta_base_cache, &iter);
2180 entry;
2181 entry = hashmap_iter_next(&iter)) {
2182 release_delta_base_cache(entry);
2183 }
2184}
2185
2186static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2187 void *base, unsigned long base_size, enum object_type type)
2188{
2189 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2190 struct list_head *lru;
2191
2192 delta_base_cached += base_size;
2193
2194 list_for_each(lru, &delta_base_cache_lru) {
2195 struct delta_base_cache_entry *f =
2196 list_entry(lru, struct delta_base_cache_entry, lru);
2197 if (delta_base_cached <= delta_base_cache_limit)
2198 break;
2199 release_delta_base_cache(f);
2200 }
2201
2202 ent->key.p = p;
2203 ent->key.base_offset = base_offset;
2204 ent->type = type;
2205 ent->data = base;
2206 ent->size = base_size;
2207 list_add_tail(&ent->lru, &delta_base_cache_lru);
2208
2209 if (!delta_base_cache.cmpfn)
2210 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2211 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2212 hashmap_add(&delta_base_cache, ent);
2213}
2214
2215static void *read_object(const unsigned char *sha1, enum object_type *type,
2216 unsigned long *size);
2217
2218static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2219{
2220 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2221 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2222 p->pack_name, (uintmax_t)obj_offset);
2223}
2224
2225int do_check_packed_object_crc;
2226
2227#define UNPACK_ENTRY_STACK_PREALLOC 64
2228struct unpack_entry_stack_ent {
2229 off_t obj_offset;
2230 off_t curpos;
2231 unsigned long size;
2232};
2233
2234void *unpack_entry(struct packed_git *p, off_t obj_offset,
2235 enum object_type *final_type, unsigned long *final_size)
2236{
2237 struct pack_window *w_curs = NULL;
2238 off_t curpos = obj_offset;
2239 void *data = NULL;
2240 unsigned long size;
2241 enum object_type type;
2242 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2243 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2244 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2245 int base_from_cache = 0;
2246
2247 write_pack_access_log(p, obj_offset);
2248
2249 /* PHASE 1: drill down to the innermost base object */
2250 for (;;) {
2251 off_t base_offset;
2252 int i;
2253 struct delta_base_cache_entry *ent;
2254
2255 ent = get_delta_base_cache_entry(p, curpos);
2256 if (ent) {
2257 type = ent->type;
2258 data = ent->data;
2259 size = ent->size;
2260 detach_delta_base_cache_entry(ent);
2261 base_from_cache = 1;
2262 break;
2263 }
2264
2265 if (do_check_packed_object_crc && p->index_version > 1) {
2266 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2267 off_t len = revidx[1].offset - obj_offset;
2268 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2269 const unsigned char *sha1 =
2270 nth_packed_object_sha1(p, revidx->nr);
2271 error("bad packed object CRC for %s",
2272 sha1_to_hex(sha1));
2273 mark_bad_packed_object(p, sha1);
2274 unuse_pack(&w_curs);
2275 return NULL;
2276 }
2277 }
2278
2279 type = unpack_object_header(p, &w_curs, &curpos, &size);
2280 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2281 break;
2282
2283 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2284 if (!base_offset) {
2285 error("failed to validate delta base reference "
2286 "at offset %"PRIuMAX" from %s",
2287 (uintmax_t)curpos, p->pack_name);
2288 /* bail to phase 2, in hopes of recovery */
2289 data = NULL;
2290 break;
2291 }
2292
2293 /* push object, proceed to base */
2294 if (delta_stack_nr >= delta_stack_alloc
2295 && delta_stack == small_delta_stack) {
2296 delta_stack_alloc = alloc_nr(delta_stack_nr);
2297 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2298 memcpy(delta_stack, small_delta_stack,
2299 sizeof(*delta_stack)*delta_stack_nr);
2300 } else {
2301 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2302 }
2303 i = delta_stack_nr++;
2304 delta_stack[i].obj_offset = obj_offset;
2305 delta_stack[i].curpos = curpos;
2306 delta_stack[i].size = size;
2307
2308 curpos = obj_offset = base_offset;
2309 }
2310
2311 /* PHASE 2: handle the base */
2312 switch (type) {
2313 case OBJ_OFS_DELTA:
2314 case OBJ_REF_DELTA:
2315 if (data)
2316 die("BUG: unpack_entry: left loop at a valid delta");
2317 break;
2318 case OBJ_COMMIT:
2319 case OBJ_TREE:
2320 case OBJ_BLOB:
2321 case OBJ_TAG:
2322 if (!base_from_cache)
2323 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2324 break;
2325 default:
2326 data = NULL;
2327 error("unknown object type %i at offset %"PRIuMAX" in %s",
2328 type, (uintmax_t)obj_offset, p->pack_name);
2329 }
2330
2331 /* PHASE 3: apply deltas in order */
2332
2333 /* invariants:
2334 * 'data' holds the base data, or NULL if there was corruption
2335 */
2336 while (delta_stack_nr) {
2337 void *delta_data;
2338 void *base = data;
2339 unsigned long delta_size, base_size = size;
2340 int i;
2341
2342 data = NULL;
2343
2344 if (base)
2345 add_delta_base_cache(p, obj_offset, base, base_size, type);
2346
2347 if (!base) {
2348 /*
2349 * We're probably in deep shit, but let's try to fetch
2350 * the required base anyway from another pack or loose.
2351 * This is costly but should happen only in the presence
2352 * of a corrupted pack, and is better than failing outright.
2353 */
2354 struct revindex_entry *revidx;
2355 const unsigned char *base_sha1;
2356 revidx = find_pack_revindex(p, obj_offset);
2357 if (revidx) {
2358 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2359 error("failed to read delta base object %s"
2360 " at offset %"PRIuMAX" from %s",
2361 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2362 p->pack_name);
2363 mark_bad_packed_object(p, base_sha1);
2364 base = read_object(base_sha1, &type, &base_size);
2365 }
2366 }
2367
2368 i = --delta_stack_nr;
2369 obj_offset = delta_stack[i].obj_offset;
2370 curpos = delta_stack[i].curpos;
2371 delta_size = delta_stack[i].size;
2372
2373 if (!base)
2374 continue;
2375
2376 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2377
2378 if (!delta_data) {
2379 error("failed to unpack compressed delta "
2380 "at offset %"PRIuMAX" from %s",
2381 (uintmax_t)curpos, p->pack_name);
2382 data = NULL;
2383 continue;
2384 }
2385
2386 data = patch_delta(base, base_size,
2387 delta_data, delta_size,
2388 &size);
2389
2390 /*
2391 * We could not apply the delta; warn the user, but keep going.
2392 * Our failure will be noticed either in the next iteration of
2393 * the loop, or if this is the final delta, in the caller when
2394 * we return NULL. Those code paths will take care of making
2395 * a more explicit warning and retrying with another copy of
2396 * the object.
2397 */
2398 if (!data)
2399 error("failed to apply delta");
2400
2401 free(delta_data);
2402 }
2403
2404 *final_type = type;
2405 *final_size = size;
2406
2407 unuse_pack(&w_curs);
2408
2409 if (delta_stack != small_delta_stack)
2410 free(delta_stack);
2411
2412 return data;
2413}
2414
2415const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2416 uint32_t n)
2417{
2418 const unsigned char *index = p->index_data;
2419 if (!index) {
2420 if (open_pack_index(p))
2421 return NULL;
2422 index = p->index_data;
2423 }
2424 if (n >= p->num_objects)
2425 return NULL;
2426 index += 4 * 256;
2427 if (p->index_version == 1) {
2428 return index + 24 * n + 4;
2429 } else {
2430 index += 8;
2431 return index + 20 * n;
2432 }
2433}
2434
2435void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2436{
2437 const unsigned char *ptr = vptr;
2438 const unsigned char *start = p->index_data;
2439 const unsigned char *end = start + p->index_size;
2440 if (ptr < start)
2441 die(_("offset before start of pack index for %s (corrupt index?)"),
2442 p->pack_name);
2443 /* No need to check for underflow; .idx files must be at least 8 bytes */
2444 if (ptr >= end - 8)
2445 die(_("offset beyond end of pack index for %s (truncated index?)"),
2446 p->pack_name);
2447}
2448
2449off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2450{
2451 const unsigned char *index = p->index_data;
2452 index += 4 * 256;
2453 if (p->index_version == 1) {
2454 return ntohl(*((uint32_t *)(index + 24 * n)));
2455 } else {
2456 uint32_t off;
2457 index += 8 + p->num_objects * (20 + 4);
2458 off = ntohl(*((uint32_t *)(index + 4 * n)));
2459 if (!(off & 0x80000000))
2460 return off;
2461 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2462 check_pack_index_ptr(p, index);
2463 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2464 ntohl(*((uint32_t *)(index + 4)));
2465 }
2466}
2467
2468off_t find_pack_entry_one(const unsigned char *sha1,
2469 struct packed_git *p)
2470{
2471 const uint32_t *level1_ofs = p->index_data;
2472 const unsigned char *index = p->index_data;
2473 unsigned hi, lo, stride;
2474 static int use_lookup = -1;
2475 static int debug_lookup = -1;
2476
2477 if (debug_lookup < 0)
2478 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2479
2480 if (!index) {
2481 if (open_pack_index(p))
2482 return 0;
2483 level1_ofs = p->index_data;
2484 index = p->index_data;
2485 }
2486 if (p->index_version > 1) {
2487 level1_ofs += 2;
2488 index += 8;
2489 }
2490 index += 4 * 256;
2491 hi = ntohl(level1_ofs[*sha1]);
2492 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2493 if (p->index_version > 1) {
2494 stride = 20;
2495 } else {
2496 stride = 24;
2497 index += 4;
2498 }
2499
2500 if (debug_lookup)
2501 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2502 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2503
2504 if (use_lookup < 0)
2505 use_lookup = !!getenv("GIT_USE_LOOKUP");
2506 if (use_lookup) {
2507 int pos = sha1_entry_pos(index, stride, 0,
2508 lo, hi, p->num_objects, sha1);
2509 if (pos < 0)
2510 return 0;
2511 return nth_packed_object_offset(p, pos);
2512 }
2513
2514 do {
2515 unsigned mi = (lo + hi) / 2;
2516 int cmp = hashcmp(index + mi * stride, sha1);
2517
2518 if (debug_lookup)
2519 printf("lo %u hi %u rg %u mi %u\n",
2520 lo, hi, hi - lo, mi);
2521 if (!cmp)
2522 return nth_packed_object_offset(p, mi);
2523 if (cmp > 0)
2524 hi = mi;
2525 else
2526 lo = mi+1;
2527 } while (lo < hi);
2528 return 0;
2529}
2530
2531int is_pack_valid(struct packed_git *p)
2532{
2533 /* An already open pack is known to be valid. */
2534 if (p->pack_fd != -1)
2535 return 1;
2536
2537 /* If the pack has one window completely covering the
2538 * file size, the pack is known to be valid even if
2539 * the descriptor is not currently open.
2540 */
2541 if (p->windows) {
2542 struct pack_window *w = p->windows;
2543
2544 if (!w->offset && w->len == p->pack_size)
2545 return 1;
2546 }
2547
2548 /* Force the pack to open to prove its valid. */
2549 return !open_packed_git(p);
2550}
2551
2552static int fill_pack_entry(const unsigned char *sha1,
2553 struct pack_entry *e,
2554 struct packed_git *p)
2555{
2556 off_t offset;
2557
2558 if (p->num_bad_objects) {
2559 unsigned i;
2560 for (i = 0; i < p->num_bad_objects; i++)
2561 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2562 return 0;
2563 }
2564
2565 offset = find_pack_entry_one(sha1, p);
2566 if (!offset)
2567 return 0;
2568
2569 /*
2570 * We are about to tell the caller where they can locate the
2571 * requested object. We better make sure the packfile is
2572 * still here and can be accessed before supplying that
2573 * answer, as it may have been deleted since the index was
2574 * loaded!
2575 */
2576 if (!is_pack_valid(p))
2577 return 0;
2578 e->offset = offset;
2579 e->p = p;
2580 hashcpy(e->sha1, sha1);
2581 return 1;
2582}
2583
2584/*
2585 * Iff a pack file contains the object named by sha1, return true and
2586 * store its location to e.
2587 */
2588static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2589{
2590 struct mru_entry *p;
2591
2592 prepare_packed_git();
2593 if (!packed_git)
2594 return 0;
2595
2596 for (p = packed_git_mru->head; p; p = p->next) {
2597 if (fill_pack_entry(sha1, e, p->item)) {
2598 mru_mark(packed_git_mru, p);
2599 return 1;
2600 }
2601 }
2602 return 0;
2603}
2604
2605struct packed_git *find_sha1_pack(const unsigned char *sha1,
2606 struct packed_git *packs)
2607{
2608 struct packed_git *p;
2609
2610 for (p = packs; p; p = p->next) {
2611 if (find_pack_entry_one(sha1, p))
2612 return p;
2613 }
2614 return NULL;
2615
2616}
2617
2618static int sha1_loose_object_info(const unsigned char *sha1,
2619 struct object_info *oi,
2620 int flags)
2621{
2622 int status = 0;
2623 unsigned long mapsize;
2624 void *map;
2625 git_zstream stream;
2626 char hdr[32];
2627 struct strbuf hdrbuf = STRBUF_INIT;
2628
2629 if (oi->delta_base_sha1)
2630 hashclr(oi->delta_base_sha1);
2631
2632 /*
2633 * If we don't care about type or size, then we don't
2634 * need to look inside the object at all. Note that we
2635 * do not optimize out the stat call, even if the
2636 * caller doesn't care about the disk-size, since our
2637 * return value implicitly indicates whether the
2638 * object even exists.
2639 */
2640 if (!oi->typep && !oi->typename && !oi->sizep) {
2641 struct stat st;
2642 if (stat_sha1_file(sha1, &st) < 0)
2643 return -1;
2644 if (oi->disk_sizep)
2645 *oi->disk_sizep = st.st_size;
2646 return 0;
2647 }
2648
2649 map = map_sha1_file(sha1, &mapsize);
2650 if (!map)
2651 return -1;
2652 if (oi->disk_sizep)
2653 *oi->disk_sizep = mapsize;
2654 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2655 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2656 status = error("unable to unpack %s header with --allow-unknown-type",
2657 sha1_to_hex(sha1));
2658 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2659 status = error("unable to unpack %s header",
2660 sha1_to_hex(sha1));
2661 if (status < 0)
2662 ; /* Do nothing */
2663 else if (hdrbuf.len) {
2664 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2665 status = error("unable to parse %s header with --allow-unknown-type",
2666 sha1_to_hex(sha1));
2667 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2668 status = error("unable to parse %s header", sha1_to_hex(sha1));
2669 git_inflate_end(&stream);
2670 munmap(map, mapsize);
2671 if (status && oi->typep)
2672 *oi->typep = status;
2673 strbuf_release(&hdrbuf);
2674 return 0;
2675}
2676
2677int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2678{
2679 struct cached_object *co;
2680 struct pack_entry e;
2681 int rtype;
2682 enum object_type real_type;
2683 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2684
2685 co = find_cached_object(real);
2686 if (co) {
2687 if (oi->typep)
2688 *(oi->typep) = co->type;
2689 if (oi->sizep)
2690 *(oi->sizep) = co->size;
2691 if (oi->disk_sizep)
2692 *(oi->disk_sizep) = 0;
2693 if (oi->delta_base_sha1)
2694 hashclr(oi->delta_base_sha1);
2695 if (oi->typename)
2696 strbuf_addstr(oi->typename, typename(co->type));
2697 oi->whence = OI_CACHED;
2698 return 0;
2699 }
2700
2701 if (!find_pack_entry(real, &e)) {
2702 /* Most likely it's a loose object. */
2703 if (!sha1_loose_object_info(real, oi, flags)) {
2704 oi->whence = OI_LOOSE;
2705 return 0;
2706 }
2707
2708 /* Not a loose object; someone else may have just packed it. */
2709 reprepare_packed_git();
2710 if (!find_pack_entry(real, &e))
2711 return -1;
2712 }
2713
2714 /*
2715 * packed_object_info() does not follow the delta chain to
2716 * find out the real type, unless it is given oi->typep.
2717 */
2718 if (oi->typename && !oi->typep)
2719 oi->typep = &real_type;
2720
2721 rtype = packed_object_info(e.p, e.offset, oi);
2722 if (rtype < 0) {
2723 mark_bad_packed_object(e.p, real);
2724 if (oi->typep == &real_type)
2725 oi->typep = NULL;
2726 return sha1_object_info_extended(real, oi, 0);
2727 } else if (in_delta_base_cache(e.p, e.offset)) {
2728 oi->whence = OI_DBCACHED;
2729 } else {
2730 oi->whence = OI_PACKED;
2731 oi->u.packed.offset = e.offset;
2732 oi->u.packed.pack = e.p;
2733 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2734 rtype == OBJ_OFS_DELTA);
2735 }
2736 if (oi->typename)
2737 strbuf_addstr(oi->typename, typename(*oi->typep));
2738 if (oi->typep == &real_type)
2739 oi->typep = NULL;
2740
2741 return 0;
2742}
2743
2744/* returns enum object_type or negative */
2745int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2746{
2747 enum object_type type;
2748 struct object_info oi = {NULL};
2749
2750 oi.typep = &type;
2751 oi.sizep = sizep;
2752 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2753 return -1;
2754 return type;
2755}
2756
2757static void *read_packed_sha1(const unsigned char *sha1,
2758 enum object_type *type, unsigned long *size)
2759{
2760 struct pack_entry e;
2761 void *data;
2762
2763 if (!find_pack_entry(sha1, &e))
2764 return NULL;
2765 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2766 if (!data) {
2767 /*
2768 * We're probably in deep shit, but let's try to fetch
2769 * the required object anyway from another pack or loose.
2770 * This should happen only in the presence of a corrupted
2771 * pack, and is better than failing outright.
2772 */
2773 error("failed to read object %s at offset %"PRIuMAX" from %s",
2774 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2775 mark_bad_packed_object(e.p, sha1);
2776 data = read_object(sha1, type, size);
2777 }
2778 return data;
2779}
2780
2781int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2782 unsigned char *sha1)
2783{
2784 struct cached_object *co;
2785
2786 hash_sha1_file(buf, len, typename(type), sha1);
2787 if (has_sha1_file(sha1) || find_cached_object(sha1))
2788 return 0;
2789 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2790 co = &cached_objects[cached_object_nr++];
2791 co->size = len;
2792 co->type = type;
2793 co->buf = xmalloc(len);
2794 memcpy(co->buf, buf, len);
2795 hashcpy(co->sha1, sha1);
2796 return 0;
2797}
2798
2799static void *read_object(const unsigned char *sha1, enum object_type *type,
2800 unsigned long *size)
2801{
2802 unsigned long mapsize;
2803 void *map, *buf;
2804 struct cached_object *co;
2805
2806 co = find_cached_object(sha1);
2807 if (co) {
2808 *type = co->type;
2809 *size = co->size;
2810 return xmemdupz(co->buf, co->size);
2811 }
2812
2813 buf = read_packed_sha1(sha1, type, size);
2814 if (buf)
2815 return buf;
2816 map = map_sha1_file(sha1, &mapsize);
2817 if (map) {
2818 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2819 munmap(map, mapsize);
2820 return buf;
2821 }
2822 reprepare_packed_git();
2823 return read_packed_sha1(sha1, type, size);
2824}
2825
2826/*
2827 * This function dies on corrupt objects; the callers who want to
2828 * deal with them should arrange to call read_object() and give error
2829 * messages themselves.
2830 */
2831void *read_sha1_file_extended(const unsigned char *sha1,
2832 enum object_type *type,
2833 unsigned long *size,
2834 unsigned flag)
2835{
2836 void *data;
2837 const struct packed_git *p;
2838 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2839
2840 errno = 0;
2841 data = read_object(repl, type, size);
2842 if (data)
2843 return data;
2844
2845 if (errno && errno != ENOENT)
2846 die_errno("failed to read object %s", sha1_to_hex(sha1));
2847
2848 /* die if we replaced an object with one that does not exist */
2849 if (repl != sha1)
2850 die("replacement %s not found for %s",
2851 sha1_to_hex(repl), sha1_to_hex(sha1));
2852
2853 if (has_loose_object(repl)) {
2854 const char *path = sha1_file_name(sha1);
2855
2856 die("loose object %s (stored in %s) is corrupt",
2857 sha1_to_hex(repl), path);
2858 }
2859
2860 if ((p = has_packed_and_bad(repl)) != NULL)
2861 die("packed object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl), p->pack_name);
2863
2864 return NULL;
2865}
2866
2867void *read_object_with_reference(const unsigned char *sha1,
2868 const char *required_type_name,
2869 unsigned long *size,
2870 unsigned char *actual_sha1_return)
2871{
2872 enum object_type type, required_type;
2873 void *buffer;
2874 unsigned long isize;
2875 unsigned char actual_sha1[20];
2876
2877 required_type = type_from_string(required_type_name);
2878 hashcpy(actual_sha1, sha1);
2879 while (1) {
2880 int ref_length = -1;
2881 const char *ref_type = NULL;
2882
2883 buffer = read_sha1_file(actual_sha1, &type, &isize);
2884 if (!buffer)
2885 return NULL;
2886 if (type == required_type) {
2887 *size = isize;
2888 if (actual_sha1_return)
2889 hashcpy(actual_sha1_return, actual_sha1);
2890 return buffer;
2891 }
2892 /* Handle references */
2893 else if (type == OBJ_COMMIT)
2894 ref_type = "tree ";
2895 else if (type == OBJ_TAG)
2896 ref_type = "object ";
2897 else {
2898 free(buffer);
2899 return NULL;
2900 }
2901 ref_length = strlen(ref_type);
2902
2903 if (ref_length + 40 > isize ||
2904 memcmp(buffer, ref_type, ref_length) ||
2905 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2906 free(buffer);
2907 return NULL;
2908 }
2909 free(buffer);
2910 /* Now we have the ID of the referred-to object in
2911 * actual_sha1. Check again. */
2912 }
2913}
2914
2915static void write_sha1_file_prepare(const void *buf, unsigned long len,
2916 const char *type, unsigned char *sha1,
2917 char *hdr, int *hdrlen)
2918{
2919 git_SHA_CTX c;
2920
2921 /* Generate the header */
2922 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2923
2924 /* Sha1.. */
2925 git_SHA1_Init(&c);
2926 git_SHA1_Update(&c, hdr, *hdrlen);
2927 git_SHA1_Update(&c, buf, len);
2928 git_SHA1_Final(sha1, &c);
2929}
2930
2931/*
2932 * Move the just written object into its final resting place.
2933 */
2934int finalize_object_file(const char *tmpfile, const char *filename)
2935{
2936 int ret = 0;
2937
2938 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2939 goto try_rename;
2940 else if (link(tmpfile, filename))
2941 ret = errno;
2942
2943 /*
2944 * Coda hack - coda doesn't like cross-directory links,
2945 * so we fall back to a rename, which will mean that it
2946 * won't be able to check collisions, but that's not a
2947 * big deal.
2948 *
2949 * The same holds for FAT formatted media.
2950 *
2951 * When this succeeds, we just return. We have nothing
2952 * left to unlink.
2953 */
2954 if (ret && ret != EEXIST) {
2955 try_rename:
2956 if (!rename(tmpfile, filename))
2957 goto out;
2958 ret = errno;
2959 }
2960 unlink_or_warn(tmpfile);
2961 if (ret) {
2962 if (ret != EEXIST) {
2963 return error_errno("unable to write sha1 filename %s", filename);
2964 }
2965 /* FIXME!!! Collision check here ? */
2966 }
2967
2968out:
2969 if (adjust_shared_perm(filename))
2970 return error("unable to set permission to '%s'", filename);
2971 return 0;
2972}
2973
2974static int write_buffer(int fd, const void *buf, size_t len)
2975{
2976 if (write_in_full(fd, buf, len) < 0)
2977 return error_errno("file write error");
2978 return 0;
2979}
2980
2981int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2982 unsigned char *sha1)
2983{
2984 char hdr[32];
2985 int hdrlen = sizeof(hdr);
2986 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2987 return 0;
2988}
2989
2990/* Finalize a file on disk, and close it. */
2991static void close_sha1_file(int fd)
2992{
2993 if (fsync_object_files)
2994 fsync_or_die(fd, "sha1 file");
2995 if (close(fd) != 0)
2996 die_errno("error when closing sha1 file");
2997}
2998
2999/* Size of directory component, including the ending '/' */
3000static inline int directory_size(const char *filename)
3001{
3002 const char *s = strrchr(filename, '/');
3003 if (!s)
3004 return 0;
3005 return s - filename + 1;
3006}
3007
3008/*
3009 * This creates a temporary file in the same directory as the final
3010 * 'filename'
3011 *
3012 * We want to avoid cross-directory filename renames, because those
3013 * can have problems on various filesystems (FAT, NFS, Coda).
3014 */
3015static int create_tmpfile(struct strbuf *tmp, const char *filename)
3016{
3017 int fd, dirlen = directory_size(filename);
3018
3019 strbuf_reset(tmp);
3020 strbuf_add(tmp, filename, dirlen);
3021 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3022 fd = git_mkstemp_mode(tmp->buf, 0444);
3023 if (fd < 0 && dirlen && errno == ENOENT) {
3024 /*
3025 * Make sure the directory exists; note that the contents
3026 * of the buffer are undefined after mkstemp returns an
3027 * error, so we have to rewrite the whole buffer from
3028 * scratch.
3029 */
3030 strbuf_reset(tmp);
3031 strbuf_add(tmp, filename, dirlen - 1);
3032 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3033 return -1;
3034 if (adjust_shared_perm(tmp->buf))
3035 return -1;
3036
3037 /* Try again */
3038 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3039 fd = git_mkstemp_mode(tmp->buf, 0444);
3040 }
3041 return fd;
3042}
3043
3044static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3045 const void *buf, unsigned long len, time_t mtime)
3046{
3047 int fd, ret;
3048 unsigned char compressed[4096];
3049 git_zstream stream;
3050 git_SHA_CTX c;
3051 unsigned char parano_sha1[20];
3052 static struct strbuf tmp_file = STRBUF_INIT;
3053 const char *filename = sha1_file_name(sha1);
3054
3055 fd = create_tmpfile(&tmp_file, filename);
3056 if (fd < 0) {
3057 if (errno == EACCES)
3058 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3059 else
3060 return error_errno("unable to create temporary file");
3061 }
3062
3063 /* Set it up */
3064 git_deflate_init(&stream, zlib_compression_level);
3065 stream.next_out = compressed;
3066 stream.avail_out = sizeof(compressed);
3067 git_SHA1_Init(&c);
3068
3069 /* First header.. */
3070 stream.next_in = (unsigned char *)hdr;
3071 stream.avail_in = hdrlen;
3072 while (git_deflate(&stream, 0) == Z_OK)
3073 ; /* nothing */
3074 git_SHA1_Update(&c, hdr, hdrlen);
3075
3076 /* Then the data itself.. */
3077 stream.next_in = (void *)buf;
3078 stream.avail_in = len;
3079 do {
3080 unsigned char *in0 = stream.next_in;
3081 ret = git_deflate(&stream, Z_FINISH);
3082 git_SHA1_Update(&c, in0, stream.next_in - in0);
3083 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3084 die("unable to write sha1 file");
3085 stream.next_out = compressed;
3086 stream.avail_out = sizeof(compressed);
3087 } while (ret == Z_OK);
3088
3089 if (ret != Z_STREAM_END)
3090 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3091 ret = git_deflate_end_gently(&stream);
3092 if (ret != Z_OK)
3093 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3094 git_SHA1_Final(parano_sha1, &c);
3095 if (hashcmp(sha1, parano_sha1) != 0)
3096 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3097
3098 close_sha1_file(fd);
3099
3100 if (mtime) {
3101 struct utimbuf utb;
3102 utb.actime = mtime;
3103 utb.modtime = mtime;
3104 if (utime(tmp_file.buf, &utb) < 0)
3105 warning_errno("failed utime() on %s", tmp_file.buf);
3106 }
3107
3108 return finalize_object_file(tmp_file.buf, filename);
3109}
3110
3111static int freshen_loose_object(const unsigned char *sha1)
3112{
3113 return check_and_freshen(sha1, 1);
3114}
3115
3116static int freshen_packed_object(const unsigned char *sha1)
3117{
3118 struct pack_entry e;
3119 if (!find_pack_entry(sha1, &e))
3120 return 0;
3121 if (e.p->freshened)
3122 return 1;
3123 if (!freshen_file(e.p->pack_name))
3124 return 0;
3125 e.p->freshened = 1;
3126 return 1;
3127}
3128
3129int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3130{
3131 char hdr[32];
3132 int hdrlen = sizeof(hdr);
3133
3134 /* Normally if we have it in the pack then we do not bother writing
3135 * it out into .git/objects/??/?{38} file.
3136 */
3137 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3138 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3139 return 0;
3140 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3141}
3142
3143int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3144 unsigned char *sha1, unsigned flags)
3145{
3146 char *header;
3147 int hdrlen, status = 0;
3148
3149 /* type string, SP, %lu of the length plus NUL must fit this */
3150 hdrlen = strlen(type) + 32;
3151 header = xmalloc(hdrlen);
3152 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3153
3154 if (!(flags & HASH_WRITE_OBJECT))
3155 goto cleanup;
3156 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3157 goto cleanup;
3158 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3159
3160cleanup:
3161 free(header);
3162 return status;
3163}
3164
3165int force_object_loose(const unsigned char *sha1, time_t mtime)
3166{
3167 void *buf;
3168 unsigned long len;
3169 enum object_type type;
3170 char hdr[32];
3171 int hdrlen;
3172 int ret;
3173
3174 if (has_loose_object(sha1))
3175 return 0;
3176 buf = read_packed_sha1(sha1, &type, &len);
3177 if (!buf)
3178 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3179 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3180 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3181 free(buf);
3182
3183 return ret;
3184}
3185
3186int has_pack_index(const unsigned char *sha1)
3187{
3188 struct stat st;
3189 if (stat(sha1_pack_index_name(sha1), &st))
3190 return 0;
3191 return 1;
3192}
3193
3194int has_sha1_pack(const unsigned char *sha1)
3195{
3196 struct pack_entry e;
3197 return find_pack_entry(sha1, &e);
3198}
3199
3200int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3201{
3202 struct pack_entry e;
3203
3204 if (find_pack_entry(sha1, &e))
3205 return 1;
3206 if (has_loose_object(sha1))
3207 return 1;
3208 if (flags & HAS_SHA1_QUICK)
3209 return 0;
3210 reprepare_packed_git();
3211 return find_pack_entry(sha1, &e);
3212}
3213
3214int has_object_file(const struct object_id *oid)
3215{
3216 return has_sha1_file(oid->hash);
3217}
3218
3219static void check_tree(const void *buf, size_t size)
3220{
3221 struct tree_desc desc;
3222 struct name_entry entry;
3223
3224 init_tree_desc(&desc, buf, size);
3225 while (tree_entry(&desc, &entry))
3226 /* do nothing
3227 * tree_entry() will die() on malformed entries */
3228 ;
3229}
3230
3231static void check_commit(const void *buf, size_t size)
3232{
3233 struct commit c;
3234 memset(&c, 0, sizeof(c));
3235 if (parse_commit_buffer(&c, buf, size))
3236 die("corrupt commit");
3237}
3238
3239static void check_tag(const void *buf, size_t size)
3240{
3241 struct tag t;
3242 memset(&t, 0, sizeof(t));
3243 if (parse_tag_buffer(&t, buf, size))
3244 die("corrupt tag");
3245}
3246
3247static int index_mem(unsigned char *sha1, void *buf, size_t size,
3248 enum object_type type,
3249 const char *path, unsigned flags)
3250{
3251 int ret, re_allocated = 0;
3252 int write_object = flags & HASH_WRITE_OBJECT;
3253
3254 if (!type)
3255 type = OBJ_BLOB;
3256
3257 /*
3258 * Convert blobs to git internal format
3259 */
3260 if ((type == OBJ_BLOB) && path) {
3261 struct strbuf nbuf = STRBUF_INIT;
3262 if (convert_to_git(path, buf, size, &nbuf,
3263 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3264 buf = strbuf_detach(&nbuf, &size);
3265 re_allocated = 1;
3266 }
3267 }
3268 if (flags & HASH_FORMAT_CHECK) {
3269 if (type == OBJ_TREE)
3270 check_tree(buf, size);
3271 if (type == OBJ_COMMIT)
3272 check_commit(buf, size);
3273 if (type == OBJ_TAG)
3274 check_tag(buf, size);
3275 }
3276
3277 if (write_object)
3278 ret = write_sha1_file(buf, size, typename(type), sha1);
3279 else
3280 ret = hash_sha1_file(buf, size, typename(type), sha1);
3281 if (re_allocated)
3282 free(buf);
3283 return ret;
3284}
3285
3286static int index_stream_convert_blob(unsigned char *sha1, int fd,
3287 const char *path, unsigned flags)
3288{
3289 int ret;
3290 const int write_object = flags & HASH_WRITE_OBJECT;
3291 struct strbuf sbuf = STRBUF_INIT;
3292
3293 assert(path);
3294 assert(would_convert_to_git_filter_fd(path));
3295
3296 convert_to_git_filter_fd(path, fd, &sbuf,
3297 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3298
3299 if (write_object)
3300 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3301 sha1);
3302 else
3303 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3304 sha1);
3305 strbuf_release(&sbuf);
3306 return ret;
3307}
3308
3309static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3310 const char *path, unsigned flags)
3311{
3312 struct strbuf sbuf = STRBUF_INIT;
3313 int ret;
3314
3315 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3316 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3317 else
3318 ret = -1;
3319 strbuf_release(&sbuf);
3320 return ret;
3321}
3322
3323#define SMALL_FILE_SIZE (32*1024)
3324
3325static int index_core(unsigned char *sha1, int fd, size_t size,
3326 enum object_type type, const char *path,
3327 unsigned flags)
3328{
3329 int ret;
3330
3331 if (!size) {
3332 ret = index_mem(sha1, "", size, type, path, flags);
3333 } else if (size <= SMALL_FILE_SIZE) {
3334 char *buf = xmalloc(size);
3335 if (size == read_in_full(fd, buf, size))
3336 ret = index_mem(sha1, buf, size, type, path, flags);
3337 else
3338 ret = error_errno("short read");
3339 free(buf);
3340 } else {
3341 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3342 ret = index_mem(sha1, buf, size, type, path, flags);
3343 munmap(buf, size);
3344 }
3345 return ret;
3346}
3347
3348/*
3349 * This creates one packfile per large blob unless bulk-checkin
3350 * machinery is "plugged".
3351 *
3352 * This also bypasses the usual "convert-to-git" dance, and that is on
3353 * purpose. We could write a streaming version of the converting
3354 * functions and insert that before feeding the data to fast-import
3355 * (or equivalent in-core API described above). However, that is
3356 * somewhat complicated, as we do not know the size of the filter
3357 * result, which we need to know beforehand when writing a git object.
3358 * Since the primary motivation for trying to stream from the working
3359 * tree file and to avoid mmaping it in core is to deal with large
3360 * binary blobs, they generally do not want to get any conversion, and
3361 * callers should avoid this code path when filters are requested.
3362 */
3363static int index_stream(unsigned char *sha1, int fd, size_t size,
3364 enum object_type type, const char *path,
3365 unsigned flags)
3366{
3367 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3368}
3369
3370int index_fd(unsigned char *sha1, int fd, struct stat *st,
3371 enum object_type type, const char *path, unsigned flags)
3372{
3373 int ret;
3374
3375 /*
3376 * Call xsize_t() only when needed to avoid potentially unnecessary
3377 * die() for large files.
3378 */
3379 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3380 ret = index_stream_convert_blob(sha1, fd, path, flags);
3381 else if (!S_ISREG(st->st_mode))
3382 ret = index_pipe(sha1, fd, type, path, flags);
3383 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3384 (path && would_convert_to_git(path)))
3385 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3386 flags);
3387 else
3388 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3389 flags);
3390 close(fd);
3391 return ret;
3392}
3393
3394int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3395{
3396 int fd;
3397 struct strbuf sb = STRBUF_INIT;
3398
3399 switch (st->st_mode & S_IFMT) {
3400 case S_IFREG:
3401 fd = open(path, O_RDONLY);
3402 if (fd < 0)
3403 return error_errno("open(\"%s\")", path);
3404 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3405 return error("%s: failed to insert into database",
3406 path);
3407 break;
3408 case S_IFLNK:
3409 if (strbuf_readlink(&sb, path, st->st_size))
3410 return error_errno("readlink(\"%s\")", path);
3411 if (!(flags & HASH_WRITE_OBJECT))
3412 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3413 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3414 return error("%s: failed to insert into database",
3415 path);
3416 strbuf_release(&sb);
3417 break;
3418 case S_IFDIR:
3419 return resolve_gitlink_ref(path, "HEAD", sha1);
3420 default:
3421 return error("%s: unsupported file type", path);
3422 }
3423 return 0;
3424}
3425
3426int read_pack_header(int fd, struct pack_header *header)
3427{
3428 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3429 /* "eof before pack header was fully read" */
3430 return PH_ERROR_EOF;
3431
3432 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3433 /* "protocol error (pack signature mismatch detected)" */
3434 return PH_ERROR_PACK_SIGNATURE;
3435 if (!pack_version_ok(header->hdr_version))
3436 /* "protocol error (pack version unsupported)" */
3437 return PH_ERROR_PROTOCOL;
3438 return 0;
3439}
3440
3441void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3442{
3443 enum object_type type = sha1_object_info(sha1, NULL);
3444 if (type < 0)
3445 die("%s is not a valid object", sha1_to_hex(sha1));
3446 if (type != expect)
3447 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3448 typename(expect));
3449}
3450
3451static int for_each_file_in_obj_subdir(int subdir_nr,
3452 struct strbuf *path,
3453 each_loose_object_fn obj_cb,
3454 each_loose_cruft_fn cruft_cb,
3455 each_loose_subdir_fn subdir_cb,
3456 void *data)
3457{
3458 size_t baselen = path->len;
3459 DIR *dir = opendir(path->buf);
3460 struct dirent *de;
3461 int r = 0;
3462
3463 if (!dir) {
3464 if (errno == ENOENT)
3465 return 0;
3466 return error_errno("unable to open %s", path->buf);
3467 }
3468
3469 while ((de = readdir(dir))) {
3470 if (is_dot_or_dotdot(de->d_name))
3471 continue;
3472
3473 strbuf_setlen(path, baselen);
3474 strbuf_addf(path, "/%s", de->d_name);
3475
3476 if (strlen(de->d_name) == 38) {
3477 char hex[41];
3478 unsigned char sha1[20];
3479
3480 snprintf(hex, sizeof(hex), "%02x%s",
3481 subdir_nr, de->d_name);
3482 if (!get_sha1_hex(hex, sha1)) {
3483 if (obj_cb) {
3484 r = obj_cb(sha1, path->buf, data);
3485 if (r)
3486 break;
3487 }
3488 continue;
3489 }
3490 }
3491
3492 if (cruft_cb) {
3493 r = cruft_cb(de->d_name, path->buf, data);
3494 if (r)
3495 break;
3496 }
3497 }
3498 closedir(dir);
3499
3500 strbuf_setlen(path, baselen);
3501 if (!r && subdir_cb)
3502 r = subdir_cb(subdir_nr, path->buf, data);
3503
3504 return r;
3505}
3506
3507int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3508 each_loose_object_fn obj_cb,
3509 each_loose_cruft_fn cruft_cb,
3510 each_loose_subdir_fn subdir_cb,
3511 void *data)
3512{
3513 size_t baselen = path->len;
3514 int r = 0;
3515 int i;
3516
3517 for (i = 0; i < 256; i++) {
3518 strbuf_addf(path, "/%02x", i);
3519 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3520 subdir_cb, data);
3521 strbuf_setlen(path, baselen);
3522 if (r)
3523 break;
3524 }
3525
3526 return r;
3527}
3528
3529int for_each_loose_file_in_objdir(const char *path,
3530 each_loose_object_fn obj_cb,
3531 each_loose_cruft_fn cruft_cb,
3532 each_loose_subdir_fn subdir_cb,
3533 void *data)
3534{
3535 struct strbuf buf = STRBUF_INIT;
3536 int r;
3537
3538 strbuf_addstr(&buf, path);
3539 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3540 subdir_cb, data);
3541 strbuf_release(&buf);
3542
3543 return r;
3544}
3545
3546struct loose_alt_odb_data {
3547 each_loose_object_fn *cb;
3548 void *data;
3549};
3550
3551static int loose_from_alt_odb(struct alternate_object_database *alt,
3552 void *vdata)
3553{
3554 struct loose_alt_odb_data *data = vdata;
3555 struct strbuf buf = STRBUF_INIT;
3556 int r;
3557
3558 /* copy base not including trailing '/' */
3559 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3560 r = for_each_loose_file_in_objdir_buf(&buf,
3561 data->cb, NULL, NULL,
3562 data->data);
3563 strbuf_release(&buf);
3564 return r;
3565}
3566
3567int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3568{
3569 struct loose_alt_odb_data alt;
3570 int r;
3571
3572 r = for_each_loose_file_in_objdir(get_object_directory(),
3573 cb, NULL, NULL, data);
3574 if (r)
3575 return r;
3576
3577 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3578 return 0;
3579
3580 alt.cb = cb;
3581 alt.data = data;
3582 return foreach_alt_odb(loose_from_alt_odb, &alt);
3583}
3584
3585static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3586{
3587 uint32_t i;
3588 int r = 0;
3589
3590 for (i = 0; i < p->num_objects; i++) {
3591 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3592
3593 if (!sha1)
3594 return error("unable to get sha1 of object %u in %s",
3595 i, p->pack_name);
3596
3597 r = cb(sha1, p, i, data);
3598 if (r)
3599 break;
3600 }
3601 return r;
3602}
3603
3604int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3605{
3606 struct packed_git *p;
3607 int r = 0;
3608 int pack_errors = 0;
3609
3610 prepare_packed_git();
3611 for (p = packed_git; p; p = p->next) {
3612 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3613 continue;
3614 if (open_pack_index(p)) {
3615 pack_errors = 1;
3616 continue;
3617 }
3618 r = for_each_object_in_pack(p, cb, data);
3619 if (r)
3620 break;
3621 }
3622 return r ? r : pack_errors;
3623}