18922b435ebd82f95b495a40b76c4ceb15416dc0
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9#include "cache.h"
10#include "string-list.h"
11#include "lockfile.h"
12#include "delta.h"
13#include "pack.h"
14#include "blob.h"
15#include "commit.h"
16#include "run-command.h"
17#include "tag.h"
18#include "tree.h"
19#include "tree-walk.h"
20#include "refs.h"
21#include "pack-revindex.h"
22#include "sha1-lookup.h"
23#include "bulk-checkin.h"
24#include "streaming.h"
25#include "dir.h"
26
27#ifndef O_NOATIME
28#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29#define O_NOATIME 01000000
30#else
31#define O_NOATIME 0
32#endif
33#endif
34
35#define SZ_FMT PRIuMAX
36static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38const unsigned char null_sha1[20];
39
40/*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51} *cached_objects;
52static int cached_object_nr, cached_object_alloc;
53
54static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59};
60
61/*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67static struct packed_git *last_found_pack;
68
69static struct cached_object *find_cached_object(const unsigned char *sha1)
70{
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81}
82
83int mkdir_in_gitdir(const char *path)
84{
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110}
111
112enum scld_error safe_create_leading_directories(char *path)
113{
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162}
163
164enum scld_error safe_create_leading_directories_const(const char *path)
165{
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171}
172
173static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174{
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183}
184
185const char *sha1_file_name(const unsigned char *sha1)
186{
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203}
204
205/*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212{
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235}
236
237char *sha1_pack_name(const unsigned char *sha1)
238{
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242}
243
244char *sha1_pack_index_name(const unsigned char *sha1)
245{
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249}
250
251struct alternate_object_database *alt_odb_list;
252static struct alternate_object_database **alt_odb_tail;
253
254/*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271{
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338}
339
340static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342{
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373}
374
375void read_info_alternates(const char * relative_base, int depth)
376{
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 char *path;
381 int fd;
382
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
391 }
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
395
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
397
398 munmap(map, mapsz);
399}
400
401void add_to_alternates_file(const char *reference)
402{
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
405 char *alt = mkpath("%s\n", reference);
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
411}
412
413int foreach_alt_odb(alt_odb_fn fn, void *cb)
414{
415 struct alternate_object_database *ent;
416 int r = 0;
417
418 prepare_alt_odb();
419 for (ent = alt_odb_list; ent; ent = ent->next) {
420 r = fn(ent, cb);
421 if (r)
422 break;
423 }
424 return r;
425}
426
427void prepare_alt_odb(void)
428{
429 const char *alt;
430
431 if (alt_odb_tail)
432 return;
433
434 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
435 if (!alt) alt = "";
436
437 alt_odb_tail = &alt_odb_list;
438 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
439
440 read_info_alternates(get_object_directory(), 0);
441}
442
443/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
444static int freshen_file(const char *fn)
445{
446 struct utimbuf t;
447 t.actime = t.modtime = time(NULL);
448 return !utime(fn, &t);
449}
450
451/*
452 * All of the check_and_freshen functions return 1 if the file exists and was
453 * freshened (if freshening was requested), 0 otherwise. If they return
454 * 0, you should not assume that it is safe to skip a write of the object (it
455 * either does not exist on disk, or has a stale mtime and may be subject to
456 * pruning).
457 */
458static int check_and_freshen_file(const char *fn, int freshen)
459{
460 if (access(fn, F_OK))
461 return 0;
462 if (freshen && !freshen_file(fn))
463 return 0;
464 return 1;
465}
466
467static int check_and_freshen_local(const unsigned char *sha1, int freshen)
468{
469 return check_and_freshen_file(sha1_file_name(sha1), freshen);
470}
471
472static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
473{
474 struct alternate_object_database *alt;
475 prepare_alt_odb();
476 for (alt = alt_odb_list; alt; alt = alt->next) {
477 fill_sha1_path(alt->name, sha1);
478 if (check_and_freshen_file(alt->base, freshen))
479 return 1;
480 }
481 return 0;
482}
483
484static int check_and_freshen(const unsigned char *sha1, int freshen)
485{
486 return check_and_freshen_local(sha1, freshen) ||
487 check_and_freshen_nonlocal(sha1, freshen);
488}
489
490int has_loose_object_nonlocal(const unsigned char *sha1)
491{
492 return check_and_freshen_nonlocal(sha1, 0);
493}
494
495static int has_loose_object(const unsigned char *sha1)
496{
497 return check_and_freshen(sha1, 0);
498}
499
500static unsigned int pack_used_ctr;
501static unsigned int pack_mmap_calls;
502static unsigned int peak_pack_open_windows;
503static unsigned int pack_open_windows;
504static unsigned int pack_open_fds;
505static unsigned int pack_max_fds;
506static size_t peak_pack_mapped;
507static size_t pack_mapped;
508struct packed_git *packed_git;
509
510void pack_report(void)
511{
512 fprintf(stderr,
513 "pack_report: getpagesize() = %10" SZ_FMT "\n"
514 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
515 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
516 sz_fmt(getpagesize()),
517 sz_fmt(packed_git_window_size),
518 sz_fmt(packed_git_limit));
519 fprintf(stderr,
520 "pack_report: pack_used_ctr = %10u\n"
521 "pack_report: pack_mmap_calls = %10u\n"
522 "pack_report: pack_open_windows = %10u / %10u\n"
523 "pack_report: pack_mapped = "
524 "%10" SZ_FMT " / %10" SZ_FMT "\n",
525 pack_used_ctr,
526 pack_mmap_calls,
527 pack_open_windows, peak_pack_open_windows,
528 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
529}
530
531/*
532 * Open and mmap the index file at path, perform a couple of
533 * consistency checks, then record its information to p. Return 0 on
534 * success.
535 */
536static int check_packed_git_idx(const char *path, struct packed_git *p)
537{
538 void *idx_map;
539 struct pack_idx_header *hdr;
540 size_t idx_size;
541 uint32_t version, nr, i, *index;
542 int fd = git_open_noatime(path);
543 struct stat st;
544
545 if (fd < 0)
546 return -1;
547 if (fstat(fd, &st)) {
548 close(fd);
549 return -1;
550 }
551 idx_size = xsize_t(st.st_size);
552 if (idx_size < 4 * 256 + 20 + 20) {
553 close(fd);
554 return error("index file %s is too small", path);
555 }
556 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
557 close(fd);
558
559 hdr = idx_map;
560 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
561 version = ntohl(hdr->idx_version);
562 if (version < 2 || version > 2) {
563 munmap(idx_map, idx_size);
564 return error("index file %s is version %"PRIu32
565 " and is not supported by this binary"
566 " (try upgrading GIT to a newer version)",
567 path, version);
568 }
569 } else
570 version = 1;
571
572 nr = 0;
573 index = idx_map;
574 if (version > 1)
575 index += 2; /* skip index header */
576 for (i = 0; i < 256; i++) {
577 uint32_t n = ntohl(index[i]);
578 if (n < nr) {
579 munmap(idx_map, idx_size);
580 return error("non-monotonic index %s", path);
581 }
582 nr = n;
583 }
584
585 if (version == 1) {
586 /*
587 * Total size:
588 * - 256 index entries 4 bytes each
589 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
590 * - 20-byte SHA1 of the packfile
591 * - 20-byte SHA1 file checksum
592 */
593 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
594 munmap(idx_map, idx_size);
595 return error("wrong index v1 file size in %s", path);
596 }
597 } else if (version == 2) {
598 /*
599 * Minimum size:
600 * - 8 bytes of header
601 * - 256 index entries 4 bytes each
602 * - 20-byte sha1 entry * nr
603 * - 4-byte crc entry * nr
604 * - 4-byte offset entry * nr
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 * And after the 4-byte offset table might be a
608 * variable sized table containing 8-byte entries
609 * for offsets larger than 2^31.
610 */
611 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
612 unsigned long max_size = min_size;
613 if (nr)
614 max_size += (nr - 1)*8;
615 if (idx_size < min_size || idx_size > max_size) {
616 munmap(idx_map, idx_size);
617 return error("wrong index v2 file size in %s", path);
618 }
619 if (idx_size != min_size &&
620 /*
621 * make sure we can deal with large pack offsets.
622 * 31-bit signed offset won't be enough, neither
623 * 32-bit unsigned one will be.
624 */
625 (sizeof(off_t) <= 4)) {
626 munmap(idx_map, idx_size);
627 return error("pack too large for current definition of off_t in %s", path);
628 }
629 }
630
631 p->index_version = version;
632 p->index_data = idx_map;
633 p->index_size = idx_size;
634 p->num_objects = nr;
635 return 0;
636}
637
638int open_pack_index(struct packed_git *p)
639{
640 char *idx_name;
641 int ret;
642
643 if (p->index_data)
644 return 0;
645
646 idx_name = xstrdup(p->pack_name);
647 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
648 ret = check_packed_git_idx(idx_name, p);
649 free(idx_name);
650 return ret;
651}
652
653static void scan_windows(struct packed_git *p,
654 struct packed_git **lru_p,
655 struct pack_window **lru_w,
656 struct pack_window **lru_l)
657{
658 struct pack_window *w, *w_l;
659
660 for (w_l = NULL, w = p->windows; w; w = w->next) {
661 if (!w->inuse_cnt) {
662 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
663 *lru_p = p;
664 *lru_w = w;
665 *lru_l = w_l;
666 }
667 }
668 w_l = w;
669 }
670}
671
672static int unuse_one_window(struct packed_git *current)
673{
674 struct packed_git *p, *lru_p = NULL;
675 struct pack_window *lru_w = NULL, *lru_l = NULL;
676
677 if (current)
678 scan_windows(current, &lru_p, &lru_w, &lru_l);
679 for (p = packed_git; p; p = p->next)
680 scan_windows(p, &lru_p, &lru_w, &lru_l);
681 if (lru_p) {
682 munmap(lru_w->base, lru_w->len);
683 pack_mapped -= lru_w->len;
684 if (lru_l)
685 lru_l->next = lru_w->next;
686 else
687 lru_p->windows = lru_w->next;
688 free(lru_w);
689 pack_open_windows--;
690 return 1;
691 }
692 return 0;
693}
694
695void release_pack_memory(size_t need)
696{
697 size_t cur = pack_mapped;
698 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
699 ; /* nothing */
700}
701
702static void mmap_limit_check(size_t length)
703{
704 static size_t limit = 0;
705 if (!limit) {
706 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
707 if (!limit)
708 limit = SIZE_MAX;
709 }
710 if (length > limit)
711 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
712 (uintmax_t)length, (uintmax_t)limit);
713}
714
715void *xmmap_gently(void *start, size_t length,
716 int prot, int flags, int fd, off_t offset)
717{
718 void *ret;
719
720 mmap_limit_check(length);
721 ret = mmap(start, length, prot, flags, fd, offset);
722 if (ret == MAP_FAILED) {
723 if (!length)
724 return NULL;
725 release_pack_memory(length);
726 ret = mmap(start, length, prot, flags, fd, offset);
727 }
728 return ret;
729}
730
731void *xmmap(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
733{
734 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
735 if (ret == MAP_FAILED)
736 die_errno("mmap failed");
737 return ret;
738}
739
740void close_pack_windows(struct packed_git *p)
741{
742 while (p->windows) {
743 struct pack_window *w = p->windows;
744
745 if (w->inuse_cnt)
746 die("pack '%s' still has open windows to it",
747 p->pack_name);
748 munmap(w->base, w->len);
749 pack_mapped -= w->len;
750 pack_open_windows--;
751 p->windows = w->next;
752 free(w);
753 }
754}
755
756static int close_pack_fd(struct packed_git *p)
757{
758 if (p->pack_fd < 0)
759 return 0;
760
761 close(p->pack_fd);
762 pack_open_fds--;
763 p->pack_fd = -1;
764
765 return 1;
766}
767
768/*
769 * The LRU pack is the one with the oldest MRU window, preferring packs
770 * with no used windows, or the oldest mtime if it has no windows allocated.
771 */
772static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
773{
774 struct pack_window *w, *this_mru_w;
775 int has_windows_inuse = 0;
776
777 /*
778 * Reject this pack if it has windows and the previously selected
779 * one does not. If this pack does not have windows, reject
780 * it if the pack file is newer than the previously selected one.
781 */
782 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
783 return;
784
785 for (w = this_mru_w = p->windows; w; w = w->next) {
786 /*
787 * Reject this pack if any of its windows are in use,
788 * but the previously selected pack did not have any
789 * inuse windows. Otherwise, record that this pack
790 * has windows in use.
791 */
792 if (w->inuse_cnt) {
793 if (*accept_windows_inuse)
794 has_windows_inuse = 1;
795 else
796 return;
797 }
798
799 if (w->last_used > this_mru_w->last_used)
800 this_mru_w = w;
801
802 /*
803 * Reject this pack if it has windows that have been
804 * used more recently than the previously selected pack.
805 * If the previously selected pack had windows inuse and
806 * we have not encountered a window in this pack that is
807 * inuse, skip this check since we prefer a pack with no
808 * inuse windows to one that has inuse windows.
809 */
810 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
811 this_mru_w->last_used > (*mru_w)->last_used)
812 return;
813 }
814
815 /*
816 * Select this pack.
817 */
818 *mru_w = this_mru_w;
819 *lru_p = p;
820 *accept_windows_inuse = has_windows_inuse;
821}
822
823static int close_one_pack(void)
824{
825 struct packed_git *p, *lru_p = NULL;
826 struct pack_window *mru_w = NULL;
827 int accept_windows_inuse = 1;
828
829 for (p = packed_git; p; p = p->next) {
830 if (p->pack_fd == -1)
831 continue;
832 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
833 }
834
835 if (lru_p)
836 return close_pack_fd(lru_p);
837
838 return 0;
839}
840
841void unuse_pack(struct pack_window **w_cursor)
842{
843 struct pack_window *w = *w_cursor;
844 if (w) {
845 w->inuse_cnt--;
846 *w_cursor = NULL;
847 }
848}
849
850void close_pack_index(struct packed_git *p)
851{
852 if (p->index_data) {
853 munmap((void *)p->index_data, p->index_size);
854 p->index_data = NULL;
855 }
856}
857
858/*
859 * This is used by git-repack in case a newly created pack happens to
860 * contain the same set of objects as an existing one. In that case
861 * the resulting file might be different even if its name would be the
862 * same. It is best to close any reference to the old pack before it is
863 * replaced on disk. Of course no index pointers or windows for given pack
864 * must subsist at this point. If ever objects from this pack are requested
865 * again, the new version of the pack will be reinitialized through
866 * reprepare_packed_git().
867 */
868void free_pack_by_name(const char *pack_name)
869{
870 struct packed_git *p, **pp = &packed_git;
871
872 while (*pp) {
873 p = *pp;
874 if (strcmp(pack_name, p->pack_name) == 0) {
875 clear_delta_base_cache();
876 close_pack_windows(p);
877 close_pack_fd(p);
878 close_pack_index(p);
879 free(p->bad_object_sha1);
880 *pp = p->next;
881 if (last_found_pack == p)
882 last_found_pack = NULL;
883 free(p);
884 return;
885 }
886 pp = &p->next;
887 }
888}
889
890static unsigned int get_max_fd_limit(void)
891{
892#ifdef RLIMIT_NOFILE
893 {
894 struct rlimit lim;
895
896 if (!getrlimit(RLIMIT_NOFILE, &lim))
897 return lim.rlim_cur;
898 }
899#endif
900
901#ifdef _SC_OPEN_MAX
902 {
903 long open_max = sysconf(_SC_OPEN_MAX);
904 if (0 < open_max)
905 return open_max;
906 /*
907 * Otherwise, we got -1 for one of the two
908 * reasons:
909 *
910 * (1) sysconf() did not understand _SC_OPEN_MAX
911 * and signaled an error with -1; or
912 * (2) sysconf() said there is no limit.
913 *
914 * We _could_ clear errno before calling sysconf() to
915 * tell these two cases apart and return a huge number
916 * in the latter case to let the caller cap it to a
917 * value that is not so selfish, but letting the
918 * fallback OPEN_MAX codepath take care of these cases
919 * is a lot simpler.
920 */
921 }
922#endif
923
924#ifdef OPEN_MAX
925 return OPEN_MAX;
926#else
927 return 1; /* see the caller ;-) */
928#endif
929}
930
931/*
932 * Do not call this directly as this leaks p->pack_fd on error return;
933 * call open_packed_git() instead.
934 */
935static int open_packed_git_1(struct packed_git *p)
936{
937 struct stat st;
938 struct pack_header hdr;
939 unsigned char sha1[20];
940 unsigned char *idx_sha1;
941 long fd_flag;
942
943 if (!p->index_data && open_pack_index(p))
944 return error("packfile %s index unavailable", p->pack_name);
945
946 if (!pack_max_fds) {
947 unsigned int max_fds = get_max_fd_limit();
948
949 /* Save 3 for stdin/stdout/stderr, 22 for work */
950 if (25 < max_fds)
951 pack_max_fds = max_fds - 25;
952 else
953 pack_max_fds = 1;
954 }
955
956 while (pack_max_fds <= pack_open_fds && close_one_pack())
957 ; /* nothing */
958
959 p->pack_fd = git_open_noatime(p->pack_name);
960 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
961 return -1;
962 pack_open_fds++;
963
964 /* If we created the struct before we had the pack we lack size. */
965 if (!p->pack_size) {
966 if (!S_ISREG(st.st_mode))
967 return error("packfile %s not a regular file", p->pack_name);
968 p->pack_size = st.st_size;
969 } else if (p->pack_size != st.st_size)
970 return error("packfile %s size changed", p->pack_name);
971
972 /* We leave these file descriptors open with sliding mmap;
973 * there is no point keeping them open across exec(), though.
974 */
975 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
976 if (fd_flag < 0)
977 return error("cannot determine file descriptor flags");
978 fd_flag |= FD_CLOEXEC;
979 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
980 return error("cannot set FD_CLOEXEC");
981
982 /* Verify we recognize this pack file format. */
983 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
984 return error("file %s is far too short to be a packfile", p->pack_name);
985 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
986 return error("file %s is not a GIT packfile", p->pack_name);
987 if (!pack_version_ok(hdr.hdr_version))
988 return error("packfile %s is version %"PRIu32" and not"
989 " supported (try upgrading GIT to a newer version)",
990 p->pack_name, ntohl(hdr.hdr_version));
991
992 /* Verify the pack matches its index. */
993 if (p->num_objects != ntohl(hdr.hdr_entries))
994 return error("packfile %s claims to have %"PRIu32" objects"
995 " while index indicates %"PRIu32" objects",
996 p->pack_name, ntohl(hdr.hdr_entries),
997 p->num_objects);
998 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
999 return error("end of packfile %s is unavailable", p->pack_name);
1000 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1001 return error("packfile %s signature is unavailable", p->pack_name);
1002 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1003 if (hashcmp(sha1, idx_sha1))
1004 return error("packfile %s does not match index", p->pack_name);
1005 return 0;
1006}
1007
1008static int open_packed_git(struct packed_git *p)
1009{
1010 if (!open_packed_git_1(p))
1011 return 0;
1012 close_pack_fd(p);
1013 return -1;
1014}
1015
1016static int in_window(struct pack_window *win, off_t offset)
1017{
1018 /* We must promise at least 20 bytes (one hash) after the
1019 * offset is available from this window, otherwise the offset
1020 * is not actually in this window and a different window (which
1021 * has that one hash excess) must be used. This is to support
1022 * the object header and delta base parsing routines below.
1023 */
1024 off_t win_off = win->offset;
1025 return win_off <= offset
1026 && (offset + 20) <= (win_off + win->len);
1027}
1028
1029unsigned char *use_pack(struct packed_git *p,
1030 struct pack_window **w_cursor,
1031 off_t offset,
1032 unsigned long *left)
1033{
1034 struct pack_window *win = *w_cursor;
1035
1036 /* Since packfiles end in a hash of their content and it's
1037 * pointless to ask for an offset into the middle of that
1038 * hash, and the in_window function above wouldn't match
1039 * don't allow an offset too close to the end of the file.
1040 */
1041 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1042 die("packfile %s cannot be accessed", p->pack_name);
1043 if (offset > (p->pack_size - 20))
1044 die("offset beyond end of packfile (truncated pack?)");
1045
1046 if (!win || !in_window(win, offset)) {
1047 if (win)
1048 win->inuse_cnt--;
1049 for (win = p->windows; win; win = win->next) {
1050 if (in_window(win, offset))
1051 break;
1052 }
1053 if (!win) {
1054 size_t window_align = packed_git_window_size / 2;
1055 off_t len;
1056
1057 if (p->pack_fd == -1 && open_packed_git(p))
1058 die("packfile %s cannot be accessed", p->pack_name);
1059
1060 win = xcalloc(1, sizeof(*win));
1061 win->offset = (offset / window_align) * window_align;
1062 len = p->pack_size - win->offset;
1063 if (len > packed_git_window_size)
1064 len = packed_git_window_size;
1065 win->len = (size_t)len;
1066 pack_mapped += win->len;
1067 while (packed_git_limit < pack_mapped
1068 && unuse_one_window(p))
1069 ; /* nothing */
1070 win->base = xmmap(NULL, win->len,
1071 PROT_READ, MAP_PRIVATE,
1072 p->pack_fd, win->offset);
1073 if (win->base == MAP_FAILED)
1074 die("packfile %s cannot be mapped: %s",
1075 p->pack_name,
1076 strerror(errno));
1077 if (!win->offset && win->len == p->pack_size
1078 && !p->do_not_close)
1079 close_pack_fd(p);
1080 pack_mmap_calls++;
1081 pack_open_windows++;
1082 if (pack_mapped > peak_pack_mapped)
1083 peak_pack_mapped = pack_mapped;
1084 if (pack_open_windows > peak_pack_open_windows)
1085 peak_pack_open_windows = pack_open_windows;
1086 win->next = p->windows;
1087 p->windows = win;
1088 }
1089 }
1090 if (win != *w_cursor) {
1091 win->last_used = pack_used_ctr++;
1092 win->inuse_cnt++;
1093 *w_cursor = win;
1094 }
1095 offset -= win->offset;
1096 if (left)
1097 *left = win->len - xsize_t(offset);
1098 return win->base + offset;
1099}
1100
1101static struct packed_git *alloc_packed_git(int extra)
1102{
1103 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1104 memset(p, 0, sizeof(*p));
1105 p->pack_fd = -1;
1106 return p;
1107}
1108
1109static void try_to_free_pack_memory(size_t size)
1110{
1111 release_pack_memory(size);
1112}
1113
1114struct packed_git *add_packed_git(const char *path, int path_len, int local)
1115{
1116 static int have_set_try_to_free_routine;
1117 struct stat st;
1118 struct packed_git *p = alloc_packed_git(path_len + 2);
1119
1120 if (!have_set_try_to_free_routine) {
1121 have_set_try_to_free_routine = 1;
1122 set_try_to_free_routine(try_to_free_pack_memory);
1123 }
1124
1125 /*
1126 * Make sure a corresponding .pack file exists and that
1127 * the index looks sane.
1128 */
1129 path_len -= strlen(".idx");
1130 if (path_len < 1) {
1131 free(p);
1132 return NULL;
1133 }
1134 memcpy(p->pack_name, path, path_len);
1135
1136 strcpy(p->pack_name + path_len, ".keep");
1137 if (!access(p->pack_name, F_OK))
1138 p->pack_keep = 1;
1139
1140 strcpy(p->pack_name + path_len, ".pack");
1141 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1142 free(p);
1143 return NULL;
1144 }
1145
1146 /* ok, it looks sane as far as we can check without
1147 * actually mapping the pack file.
1148 */
1149 p->pack_size = st.st_size;
1150 p->pack_local = local;
1151 p->mtime = st.st_mtime;
1152 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1153 hashclr(p->sha1);
1154 return p;
1155}
1156
1157struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1158{
1159 const char *path = sha1_pack_name(sha1);
1160 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1161
1162 strcpy(p->pack_name, path);
1163 hashcpy(p->sha1, sha1);
1164 if (check_packed_git_idx(idx_path, p)) {
1165 free(p);
1166 return NULL;
1167 }
1168
1169 return p;
1170}
1171
1172void install_packed_git(struct packed_git *pack)
1173{
1174 if (pack->pack_fd != -1)
1175 pack_open_fds++;
1176
1177 pack->next = packed_git;
1178 packed_git = pack;
1179}
1180
1181void (*report_garbage)(const char *desc, const char *path);
1182
1183static void report_helper(const struct string_list *list,
1184 int seen_bits, int first, int last)
1185{
1186 const char *msg;
1187 switch (seen_bits) {
1188 case 0:
1189 msg = "no corresponding .idx or .pack";
1190 break;
1191 case 1:
1192 msg = "no corresponding .idx";
1193 break;
1194 case 2:
1195 msg = "no corresponding .pack";
1196 break;
1197 default:
1198 return;
1199 }
1200 for (; first < last; first++)
1201 report_garbage(msg, list->items[first].string);
1202}
1203
1204static void report_pack_garbage(struct string_list *list)
1205{
1206 int i, baselen = -1, first = 0, seen_bits = 0;
1207
1208 if (!report_garbage)
1209 return;
1210
1211 string_list_sort(list);
1212
1213 for (i = 0; i < list->nr; i++) {
1214 const char *path = list->items[i].string;
1215 if (baselen != -1 &&
1216 strncmp(path, list->items[first].string, baselen)) {
1217 report_helper(list, seen_bits, first, i);
1218 baselen = -1;
1219 seen_bits = 0;
1220 }
1221 if (baselen == -1) {
1222 const char *dot = strrchr(path, '.');
1223 if (!dot) {
1224 report_garbage("garbage found", path);
1225 continue;
1226 }
1227 baselen = dot - path + 1;
1228 first = i;
1229 }
1230 if (!strcmp(path + baselen, "pack"))
1231 seen_bits |= 1;
1232 else if (!strcmp(path + baselen, "idx"))
1233 seen_bits |= 2;
1234 }
1235 report_helper(list, seen_bits, first, list->nr);
1236}
1237
1238static void prepare_packed_git_one(char *objdir, int local)
1239{
1240 struct strbuf path = STRBUF_INIT;
1241 size_t dirnamelen;
1242 DIR *dir;
1243 struct dirent *de;
1244 struct string_list garbage = STRING_LIST_INIT_DUP;
1245
1246 strbuf_addstr(&path, objdir);
1247 strbuf_addstr(&path, "/pack");
1248 dir = opendir(path.buf);
1249 if (!dir) {
1250 if (errno != ENOENT)
1251 error("unable to open object pack directory: %s: %s",
1252 path.buf, strerror(errno));
1253 strbuf_release(&path);
1254 return;
1255 }
1256 strbuf_addch(&path, '/');
1257 dirnamelen = path.len;
1258 while ((de = readdir(dir)) != NULL) {
1259 struct packed_git *p;
1260 size_t base_len;
1261
1262 if (is_dot_or_dotdot(de->d_name))
1263 continue;
1264
1265 strbuf_setlen(&path, dirnamelen);
1266 strbuf_addstr(&path, de->d_name);
1267
1268 base_len = path.len;
1269 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1270 /* Don't reopen a pack we already have. */
1271 for (p = packed_git; p; p = p->next) {
1272 size_t len;
1273 if (strip_suffix(p->pack_name, ".pack", &len) &&
1274 len == base_len &&
1275 !memcmp(p->pack_name, path.buf, len))
1276 break;
1277 }
1278 if (p == NULL &&
1279 /*
1280 * See if it really is a valid .idx file with
1281 * corresponding .pack file that we can map.
1282 */
1283 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1284 install_packed_git(p);
1285 }
1286
1287 if (!report_garbage)
1288 continue;
1289
1290 if (ends_with(de->d_name, ".idx") ||
1291 ends_with(de->d_name, ".pack") ||
1292 ends_with(de->d_name, ".bitmap") ||
1293 ends_with(de->d_name, ".keep"))
1294 string_list_append(&garbage, path.buf);
1295 else
1296 report_garbage("garbage found", path.buf);
1297 }
1298 closedir(dir);
1299 report_pack_garbage(&garbage);
1300 string_list_clear(&garbage, 0);
1301 strbuf_release(&path);
1302}
1303
1304static int sort_pack(const void *a_, const void *b_)
1305{
1306 struct packed_git *a = *((struct packed_git **)a_);
1307 struct packed_git *b = *((struct packed_git **)b_);
1308 int st;
1309
1310 /*
1311 * Local packs tend to contain objects specific to our
1312 * variant of the project than remote ones. In addition,
1313 * remote ones could be on a network mounted filesystem.
1314 * Favor local ones for these reasons.
1315 */
1316 st = a->pack_local - b->pack_local;
1317 if (st)
1318 return -st;
1319
1320 /*
1321 * Younger packs tend to contain more recent objects,
1322 * and more recent objects tend to get accessed more
1323 * often.
1324 */
1325 if (a->mtime < b->mtime)
1326 return 1;
1327 else if (a->mtime == b->mtime)
1328 return 0;
1329 return -1;
1330}
1331
1332static void rearrange_packed_git(void)
1333{
1334 struct packed_git **ary, *p;
1335 int i, n;
1336
1337 for (n = 0, p = packed_git; p; p = p->next)
1338 n++;
1339 if (n < 2)
1340 return;
1341
1342 /* prepare an array of packed_git for easier sorting */
1343 ary = xcalloc(n, sizeof(struct packed_git *));
1344 for (n = 0, p = packed_git; p; p = p->next)
1345 ary[n++] = p;
1346
1347 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1348
1349 /* link them back again */
1350 for (i = 0; i < n - 1; i++)
1351 ary[i]->next = ary[i + 1];
1352 ary[n - 1]->next = NULL;
1353 packed_git = ary[0];
1354
1355 free(ary);
1356}
1357
1358static int prepare_packed_git_run_once = 0;
1359void prepare_packed_git(void)
1360{
1361 struct alternate_object_database *alt;
1362
1363 if (prepare_packed_git_run_once)
1364 return;
1365 prepare_packed_git_one(get_object_directory(), 1);
1366 prepare_alt_odb();
1367 for (alt = alt_odb_list; alt; alt = alt->next) {
1368 alt->name[-1] = 0;
1369 prepare_packed_git_one(alt->base, 0);
1370 alt->name[-1] = '/';
1371 }
1372 rearrange_packed_git();
1373 prepare_packed_git_run_once = 1;
1374}
1375
1376void reprepare_packed_git(void)
1377{
1378 prepare_packed_git_run_once = 0;
1379 prepare_packed_git();
1380}
1381
1382static void mark_bad_packed_object(struct packed_git *p,
1383 const unsigned char *sha1)
1384{
1385 unsigned i;
1386 for (i = 0; i < p->num_bad_objects; i++)
1387 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1388 return;
1389 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1390 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1391 p->num_bad_objects++;
1392}
1393
1394static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1395{
1396 struct packed_git *p;
1397 unsigned i;
1398
1399 for (p = packed_git; p; p = p->next)
1400 for (i = 0; i < p->num_bad_objects; i++)
1401 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1402 return p;
1403 return NULL;
1404}
1405
1406/*
1407 * With an in-core object data in "map", rehash it to make sure the
1408 * object name actually matches "sha1" to detect object corruption.
1409 * With "map" == NULL, try reading the object named with "sha1" using
1410 * the streaming interface and rehash it to do the same.
1411 */
1412int check_sha1_signature(const unsigned char *sha1, void *map,
1413 unsigned long size, const char *type)
1414{
1415 unsigned char real_sha1[20];
1416 enum object_type obj_type;
1417 struct git_istream *st;
1418 git_SHA_CTX c;
1419 char hdr[32];
1420 int hdrlen;
1421
1422 if (map) {
1423 hash_sha1_file(map, size, type, real_sha1);
1424 return hashcmp(sha1, real_sha1) ? -1 : 0;
1425 }
1426
1427 st = open_istream(sha1, &obj_type, &size, NULL);
1428 if (!st)
1429 return -1;
1430
1431 /* Generate the header */
1432 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1433
1434 /* Sha1.. */
1435 git_SHA1_Init(&c);
1436 git_SHA1_Update(&c, hdr, hdrlen);
1437 for (;;) {
1438 char buf[1024 * 16];
1439 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1440
1441 if (readlen < 0) {
1442 close_istream(st);
1443 return -1;
1444 }
1445 if (!readlen)
1446 break;
1447 git_SHA1_Update(&c, buf, readlen);
1448 }
1449 git_SHA1_Final(real_sha1, &c);
1450 close_istream(st);
1451 return hashcmp(sha1, real_sha1) ? -1 : 0;
1452}
1453
1454int git_open_noatime(const char *name)
1455{
1456 static int sha1_file_open_flag = O_NOATIME;
1457
1458 for (;;) {
1459 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1460 if (fd >= 0)
1461 return fd;
1462
1463 /* Might the failure be due to O_NOATIME? */
1464 if (errno != ENOENT && sha1_file_open_flag) {
1465 sha1_file_open_flag = 0;
1466 continue;
1467 }
1468
1469 return -1;
1470 }
1471}
1472
1473static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1474{
1475 struct alternate_object_database *alt;
1476
1477 if (!lstat(sha1_file_name(sha1), st))
1478 return 0;
1479
1480 prepare_alt_odb();
1481 errno = ENOENT;
1482 for (alt = alt_odb_list; alt; alt = alt->next) {
1483 fill_sha1_path(alt->name, sha1);
1484 if (!lstat(alt->base, st))
1485 return 0;
1486 }
1487
1488 return -1;
1489}
1490
1491static int open_sha1_file(const unsigned char *sha1)
1492{
1493 int fd;
1494 struct alternate_object_database *alt;
1495 int most_interesting_errno;
1496
1497 fd = git_open_noatime(sha1_file_name(sha1));
1498 if (fd >= 0)
1499 return fd;
1500 most_interesting_errno = errno;
1501
1502 prepare_alt_odb();
1503 for (alt = alt_odb_list; alt; alt = alt->next) {
1504 fill_sha1_path(alt->name, sha1);
1505 fd = git_open_noatime(alt->base);
1506 if (fd >= 0)
1507 return fd;
1508 if (most_interesting_errno == ENOENT)
1509 most_interesting_errno = errno;
1510 }
1511 errno = most_interesting_errno;
1512 return -1;
1513}
1514
1515void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1516{
1517 void *map;
1518 int fd;
1519
1520 fd = open_sha1_file(sha1);
1521 map = NULL;
1522 if (fd >= 0) {
1523 struct stat st;
1524
1525 if (!fstat(fd, &st)) {
1526 *size = xsize_t(st.st_size);
1527 if (!*size) {
1528 /* mmap() is forbidden on empty files */
1529 error("object file %s is empty", sha1_file_name(sha1));
1530 return NULL;
1531 }
1532 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1533 }
1534 close(fd);
1535 }
1536 return map;
1537}
1538
1539unsigned long unpack_object_header_buffer(const unsigned char *buf,
1540 unsigned long len, enum object_type *type, unsigned long *sizep)
1541{
1542 unsigned shift;
1543 unsigned long size, c;
1544 unsigned long used = 0;
1545
1546 c = buf[used++];
1547 *type = (c >> 4) & 7;
1548 size = c & 15;
1549 shift = 4;
1550 while (c & 0x80) {
1551 if (len <= used || bitsizeof(long) <= shift) {
1552 error("bad object header");
1553 size = used = 0;
1554 break;
1555 }
1556 c = buf[used++];
1557 size += (c & 0x7f) << shift;
1558 shift += 7;
1559 }
1560 *sizep = size;
1561 return used;
1562}
1563
1564int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1565{
1566 /* Get the data stream */
1567 memset(stream, 0, sizeof(*stream));
1568 stream->next_in = map;
1569 stream->avail_in = mapsize;
1570 stream->next_out = buffer;
1571 stream->avail_out = bufsiz;
1572
1573 git_inflate_init(stream);
1574 return git_inflate(stream, 0);
1575}
1576
1577static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1578{
1579 int bytes = strlen(buffer) + 1;
1580 unsigned char *buf = xmallocz(size);
1581 unsigned long n;
1582 int status = Z_OK;
1583
1584 n = stream->total_out - bytes;
1585 if (n > size)
1586 n = size;
1587 memcpy(buf, (char *) buffer + bytes, n);
1588 bytes = n;
1589 if (bytes <= size) {
1590 /*
1591 * The above condition must be (bytes <= size), not
1592 * (bytes < size). In other words, even though we
1593 * expect no more output and set avail_out to zero,
1594 * the input zlib stream may have bytes that express
1595 * "this concludes the stream", and we *do* want to
1596 * eat that input.
1597 *
1598 * Otherwise we would not be able to test that we
1599 * consumed all the input to reach the expected size;
1600 * we also want to check that zlib tells us that all
1601 * went well with status == Z_STREAM_END at the end.
1602 */
1603 stream->next_out = buf + bytes;
1604 stream->avail_out = size - bytes;
1605 while (status == Z_OK)
1606 status = git_inflate(stream, Z_FINISH);
1607 }
1608 if (status == Z_STREAM_END && !stream->avail_in) {
1609 git_inflate_end(stream);
1610 return buf;
1611 }
1612
1613 if (status < 0)
1614 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1615 else if (stream->avail_in)
1616 error("garbage at end of loose object '%s'",
1617 sha1_to_hex(sha1));
1618 free(buf);
1619 return NULL;
1620}
1621
1622/*
1623 * We used to just use "sscanf()", but that's actually way
1624 * too permissive for what we want to check. So do an anal
1625 * object header parse by hand.
1626 */
1627int parse_sha1_header(const char *hdr, unsigned long *sizep)
1628{
1629 char type[10];
1630 int i;
1631 unsigned long size;
1632
1633 /*
1634 * The type can be at most ten bytes (including the
1635 * terminating '\0' that we add), and is followed by
1636 * a space.
1637 */
1638 i = 0;
1639 for (;;) {
1640 char c = *hdr++;
1641 if (c == ' ')
1642 break;
1643 type[i++] = c;
1644 if (i >= sizeof(type))
1645 return -1;
1646 }
1647 type[i] = 0;
1648
1649 /*
1650 * The length must follow immediately, and be in canonical
1651 * decimal format (ie "010" is not valid).
1652 */
1653 size = *hdr++ - '0';
1654 if (size > 9)
1655 return -1;
1656 if (size) {
1657 for (;;) {
1658 unsigned long c = *hdr - '0';
1659 if (c > 9)
1660 break;
1661 hdr++;
1662 size = size * 10 + c;
1663 }
1664 }
1665 *sizep = size;
1666
1667 /*
1668 * The length must be followed by a zero byte
1669 */
1670 return *hdr ? -1 : type_from_string(type);
1671}
1672
1673static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1674{
1675 int ret;
1676 git_zstream stream;
1677 char hdr[8192];
1678
1679 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1680 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1681 return NULL;
1682
1683 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1684}
1685
1686unsigned long get_size_from_delta(struct packed_git *p,
1687 struct pack_window **w_curs,
1688 off_t curpos)
1689{
1690 const unsigned char *data;
1691 unsigned char delta_head[20], *in;
1692 git_zstream stream;
1693 int st;
1694
1695 memset(&stream, 0, sizeof(stream));
1696 stream.next_out = delta_head;
1697 stream.avail_out = sizeof(delta_head);
1698
1699 git_inflate_init(&stream);
1700 do {
1701 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1702 stream.next_in = in;
1703 st = git_inflate(&stream, Z_FINISH);
1704 curpos += stream.next_in - in;
1705 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1706 stream.total_out < sizeof(delta_head));
1707 git_inflate_end(&stream);
1708 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1709 error("delta data unpack-initial failed");
1710 return 0;
1711 }
1712
1713 /* Examine the initial part of the delta to figure out
1714 * the result size.
1715 */
1716 data = delta_head;
1717
1718 /* ignore base size */
1719 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1720
1721 /* Read the result size */
1722 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1723}
1724
1725static off_t get_delta_base(struct packed_git *p,
1726 struct pack_window **w_curs,
1727 off_t *curpos,
1728 enum object_type type,
1729 off_t delta_obj_offset)
1730{
1731 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1732 off_t base_offset;
1733
1734 /* use_pack() assured us we have [base_info, base_info + 20)
1735 * as a range that we can look at without walking off the
1736 * end of the mapped window. Its actually the hash size
1737 * that is assured. An OFS_DELTA longer than the hash size
1738 * is stupid, as then a REF_DELTA would be smaller to store.
1739 */
1740 if (type == OBJ_OFS_DELTA) {
1741 unsigned used = 0;
1742 unsigned char c = base_info[used++];
1743 base_offset = c & 127;
1744 while (c & 128) {
1745 base_offset += 1;
1746 if (!base_offset || MSB(base_offset, 7))
1747 return 0; /* overflow */
1748 c = base_info[used++];
1749 base_offset = (base_offset << 7) + (c & 127);
1750 }
1751 base_offset = delta_obj_offset - base_offset;
1752 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1753 return 0; /* out of bound */
1754 *curpos += used;
1755 } else if (type == OBJ_REF_DELTA) {
1756 /* The base entry _must_ be in the same pack */
1757 base_offset = find_pack_entry_one(base_info, p);
1758 *curpos += 20;
1759 } else
1760 die("I am totally screwed");
1761 return base_offset;
1762}
1763
1764/*
1765 * Like get_delta_base above, but we return the sha1 instead of the pack
1766 * offset. This means it is cheaper for REF deltas (we do not have to do
1767 * the final object lookup), but more expensive for OFS deltas (we
1768 * have to load the revidx to convert the offset back into a sha1).
1769 */
1770static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1771 struct pack_window **w_curs,
1772 off_t curpos,
1773 enum object_type type,
1774 off_t delta_obj_offset)
1775{
1776 if (type == OBJ_REF_DELTA) {
1777 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1778 return base;
1779 } else if (type == OBJ_OFS_DELTA) {
1780 struct revindex_entry *revidx;
1781 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1782 type, delta_obj_offset);
1783
1784 if (!base_offset)
1785 return NULL;
1786
1787 revidx = find_pack_revindex(p, base_offset);
1788 if (!revidx)
1789 return NULL;
1790
1791 return nth_packed_object_sha1(p, revidx->nr);
1792 } else
1793 return NULL;
1794}
1795
1796int unpack_object_header(struct packed_git *p,
1797 struct pack_window **w_curs,
1798 off_t *curpos,
1799 unsigned long *sizep)
1800{
1801 unsigned char *base;
1802 unsigned long left;
1803 unsigned long used;
1804 enum object_type type;
1805
1806 /* use_pack() assures us we have [base, base + 20) available
1807 * as a range that we can look at. (Its actually the hash
1808 * size that is assured.) With our object header encoding
1809 * the maximum deflated object size is 2^137, which is just
1810 * insane, so we know won't exceed what we have been given.
1811 */
1812 base = use_pack(p, w_curs, *curpos, &left);
1813 used = unpack_object_header_buffer(base, left, &type, sizep);
1814 if (!used) {
1815 type = OBJ_BAD;
1816 } else
1817 *curpos += used;
1818
1819 return type;
1820}
1821
1822static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1823{
1824 int type;
1825 struct revindex_entry *revidx;
1826 const unsigned char *sha1;
1827 revidx = find_pack_revindex(p, obj_offset);
1828 if (!revidx)
1829 return OBJ_BAD;
1830 sha1 = nth_packed_object_sha1(p, revidx->nr);
1831 mark_bad_packed_object(p, sha1);
1832 type = sha1_object_info(sha1, NULL);
1833 if (type <= OBJ_NONE)
1834 return OBJ_BAD;
1835 return type;
1836}
1837
1838#define POI_STACK_PREALLOC 64
1839
1840static enum object_type packed_to_object_type(struct packed_git *p,
1841 off_t obj_offset,
1842 enum object_type type,
1843 struct pack_window **w_curs,
1844 off_t curpos)
1845{
1846 off_t small_poi_stack[POI_STACK_PREALLOC];
1847 off_t *poi_stack = small_poi_stack;
1848 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1849
1850 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1851 off_t base_offset;
1852 unsigned long size;
1853 /* Push the object we're going to leave behind */
1854 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1855 poi_stack_alloc = alloc_nr(poi_stack_nr);
1856 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1857 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1858 } else {
1859 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1860 }
1861 poi_stack[poi_stack_nr++] = obj_offset;
1862 /* If parsing the base offset fails, just unwind */
1863 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1864 if (!base_offset)
1865 goto unwind;
1866 curpos = obj_offset = base_offset;
1867 type = unpack_object_header(p, w_curs, &curpos, &size);
1868 if (type <= OBJ_NONE) {
1869 /* If getting the base itself fails, we first
1870 * retry the base, otherwise unwind */
1871 type = retry_bad_packed_offset(p, base_offset);
1872 if (type > OBJ_NONE)
1873 goto out;
1874 goto unwind;
1875 }
1876 }
1877
1878 switch (type) {
1879 case OBJ_BAD:
1880 case OBJ_COMMIT:
1881 case OBJ_TREE:
1882 case OBJ_BLOB:
1883 case OBJ_TAG:
1884 break;
1885 default:
1886 error("unknown object type %i at offset %"PRIuMAX" in %s",
1887 type, (uintmax_t)obj_offset, p->pack_name);
1888 type = OBJ_BAD;
1889 }
1890
1891out:
1892 if (poi_stack != small_poi_stack)
1893 free(poi_stack);
1894 return type;
1895
1896unwind:
1897 while (poi_stack_nr) {
1898 obj_offset = poi_stack[--poi_stack_nr];
1899 type = retry_bad_packed_offset(p, obj_offset);
1900 if (type > OBJ_NONE)
1901 goto out;
1902 }
1903 type = OBJ_BAD;
1904 goto out;
1905}
1906
1907static int packed_object_info(struct packed_git *p, off_t obj_offset,
1908 struct object_info *oi)
1909{
1910 struct pack_window *w_curs = NULL;
1911 unsigned long size;
1912 off_t curpos = obj_offset;
1913 enum object_type type;
1914
1915 /*
1916 * We always get the representation type, but only convert it to
1917 * a "real" type later if the caller is interested.
1918 */
1919 type = unpack_object_header(p, &w_curs, &curpos, &size);
1920
1921 if (oi->sizep) {
1922 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1923 off_t tmp_pos = curpos;
1924 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1925 type, obj_offset);
1926 if (!base_offset) {
1927 type = OBJ_BAD;
1928 goto out;
1929 }
1930 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1931 if (*oi->sizep == 0) {
1932 type = OBJ_BAD;
1933 goto out;
1934 }
1935 } else {
1936 *oi->sizep = size;
1937 }
1938 }
1939
1940 if (oi->disk_sizep) {
1941 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1942 *oi->disk_sizep = revidx[1].offset - obj_offset;
1943 }
1944
1945 if (oi->typep) {
1946 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1947 if (*oi->typep < 0) {
1948 type = OBJ_BAD;
1949 goto out;
1950 }
1951 }
1952
1953 if (oi->delta_base_sha1) {
1954 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1955 const unsigned char *base;
1956
1957 base = get_delta_base_sha1(p, &w_curs, curpos,
1958 type, obj_offset);
1959 if (!base) {
1960 type = OBJ_BAD;
1961 goto out;
1962 }
1963
1964 hashcpy(oi->delta_base_sha1, base);
1965 } else
1966 hashclr(oi->delta_base_sha1);
1967 }
1968
1969out:
1970 unuse_pack(&w_curs);
1971 return type;
1972}
1973
1974static void *unpack_compressed_entry(struct packed_git *p,
1975 struct pack_window **w_curs,
1976 off_t curpos,
1977 unsigned long size)
1978{
1979 int st;
1980 git_zstream stream;
1981 unsigned char *buffer, *in;
1982
1983 buffer = xmallocz_gently(size);
1984 if (!buffer)
1985 return NULL;
1986 memset(&stream, 0, sizeof(stream));
1987 stream.next_out = buffer;
1988 stream.avail_out = size + 1;
1989
1990 git_inflate_init(&stream);
1991 do {
1992 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1993 stream.next_in = in;
1994 st = git_inflate(&stream, Z_FINISH);
1995 if (!stream.avail_out)
1996 break; /* the payload is larger than it should be */
1997 curpos += stream.next_in - in;
1998 } while (st == Z_OK || st == Z_BUF_ERROR);
1999 git_inflate_end(&stream);
2000 if ((st != Z_STREAM_END) || stream.total_out != size) {
2001 free(buffer);
2002 return NULL;
2003 }
2004
2005 return buffer;
2006}
2007
2008#define MAX_DELTA_CACHE (256)
2009
2010static size_t delta_base_cached;
2011
2012static struct delta_base_cache_lru_list {
2013 struct delta_base_cache_lru_list *prev;
2014 struct delta_base_cache_lru_list *next;
2015} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2016
2017static struct delta_base_cache_entry {
2018 struct delta_base_cache_lru_list lru;
2019 void *data;
2020 struct packed_git *p;
2021 off_t base_offset;
2022 unsigned long size;
2023 enum object_type type;
2024} delta_base_cache[MAX_DELTA_CACHE];
2025
2026static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2027{
2028 unsigned long hash;
2029
2030 hash = (unsigned long)p + (unsigned long)base_offset;
2031 hash += (hash >> 8) + (hash >> 16);
2032 return hash % MAX_DELTA_CACHE;
2033}
2034
2035static struct delta_base_cache_entry *
2036get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2037{
2038 unsigned long hash = pack_entry_hash(p, base_offset);
2039 return delta_base_cache + hash;
2040}
2041
2042static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2043 struct packed_git *p, off_t base_offset)
2044{
2045 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2046}
2047
2048static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2049{
2050 struct delta_base_cache_entry *ent;
2051 ent = get_delta_base_cache_entry(p, base_offset);
2052 return eq_delta_base_cache_entry(ent, p, base_offset);
2053}
2054
2055static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2056{
2057 ent->data = NULL;
2058 ent->lru.next->prev = ent->lru.prev;
2059 ent->lru.prev->next = ent->lru.next;
2060 delta_base_cached -= ent->size;
2061}
2062
2063static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2064 unsigned long *base_size, enum object_type *type, int keep_cache)
2065{
2066 struct delta_base_cache_entry *ent;
2067 void *ret;
2068
2069 ent = get_delta_base_cache_entry(p, base_offset);
2070
2071 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2072 return unpack_entry(p, base_offset, type, base_size);
2073
2074 ret = ent->data;
2075
2076 if (!keep_cache)
2077 clear_delta_base_cache_entry(ent);
2078 else
2079 ret = xmemdupz(ent->data, ent->size);
2080 *type = ent->type;
2081 *base_size = ent->size;
2082 return ret;
2083}
2084
2085static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2086{
2087 if (ent->data) {
2088 free(ent->data);
2089 ent->data = NULL;
2090 ent->lru.next->prev = ent->lru.prev;
2091 ent->lru.prev->next = ent->lru.next;
2092 delta_base_cached -= ent->size;
2093 }
2094}
2095
2096void clear_delta_base_cache(void)
2097{
2098 unsigned long p;
2099 for (p = 0; p < MAX_DELTA_CACHE; p++)
2100 release_delta_base_cache(&delta_base_cache[p]);
2101}
2102
2103static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2104 void *base, unsigned long base_size, enum object_type type)
2105{
2106 unsigned long hash = pack_entry_hash(p, base_offset);
2107 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2108 struct delta_base_cache_lru_list *lru;
2109
2110 release_delta_base_cache(ent);
2111 delta_base_cached += base_size;
2112
2113 for (lru = delta_base_cache_lru.next;
2114 delta_base_cached > delta_base_cache_limit
2115 && lru != &delta_base_cache_lru;
2116 lru = lru->next) {
2117 struct delta_base_cache_entry *f = (void *)lru;
2118 if (f->type == OBJ_BLOB)
2119 release_delta_base_cache(f);
2120 }
2121 for (lru = delta_base_cache_lru.next;
2122 delta_base_cached > delta_base_cache_limit
2123 && lru != &delta_base_cache_lru;
2124 lru = lru->next) {
2125 struct delta_base_cache_entry *f = (void *)lru;
2126 release_delta_base_cache(f);
2127 }
2128
2129 ent->p = p;
2130 ent->base_offset = base_offset;
2131 ent->type = type;
2132 ent->data = base;
2133 ent->size = base_size;
2134 ent->lru.next = &delta_base_cache_lru;
2135 ent->lru.prev = delta_base_cache_lru.prev;
2136 delta_base_cache_lru.prev->next = &ent->lru;
2137 delta_base_cache_lru.prev = &ent->lru;
2138}
2139
2140static void *read_object(const unsigned char *sha1, enum object_type *type,
2141 unsigned long *size);
2142
2143static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2144{
2145 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2146 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2147 p->pack_name, (uintmax_t)obj_offset);
2148}
2149
2150int do_check_packed_object_crc;
2151
2152#define UNPACK_ENTRY_STACK_PREALLOC 64
2153struct unpack_entry_stack_ent {
2154 off_t obj_offset;
2155 off_t curpos;
2156 unsigned long size;
2157};
2158
2159void *unpack_entry(struct packed_git *p, off_t obj_offset,
2160 enum object_type *final_type, unsigned long *final_size)
2161{
2162 struct pack_window *w_curs = NULL;
2163 off_t curpos = obj_offset;
2164 void *data = NULL;
2165 unsigned long size;
2166 enum object_type type;
2167 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2168 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2169 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2170 int base_from_cache = 0;
2171
2172 write_pack_access_log(p, obj_offset);
2173
2174 /* PHASE 1: drill down to the innermost base object */
2175 for (;;) {
2176 off_t base_offset;
2177 int i;
2178 struct delta_base_cache_entry *ent;
2179
2180 ent = get_delta_base_cache_entry(p, curpos);
2181 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2182 type = ent->type;
2183 data = ent->data;
2184 size = ent->size;
2185 clear_delta_base_cache_entry(ent);
2186 base_from_cache = 1;
2187 break;
2188 }
2189
2190 if (do_check_packed_object_crc && p->index_version > 1) {
2191 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2192 unsigned long len = revidx[1].offset - obj_offset;
2193 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2194 const unsigned char *sha1 =
2195 nth_packed_object_sha1(p, revidx->nr);
2196 error("bad packed object CRC for %s",
2197 sha1_to_hex(sha1));
2198 mark_bad_packed_object(p, sha1);
2199 unuse_pack(&w_curs);
2200 return NULL;
2201 }
2202 }
2203
2204 type = unpack_object_header(p, &w_curs, &curpos, &size);
2205 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2206 break;
2207
2208 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2209 if (!base_offset) {
2210 error("failed to validate delta base reference "
2211 "at offset %"PRIuMAX" from %s",
2212 (uintmax_t)curpos, p->pack_name);
2213 /* bail to phase 2, in hopes of recovery */
2214 data = NULL;
2215 break;
2216 }
2217
2218 /* push object, proceed to base */
2219 if (delta_stack_nr >= delta_stack_alloc
2220 && delta_stack == small_delta_stack) {
2221 delta_stack_alloc = alloc_nr(delta_stack_nr);
2222 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2223 memcpy(delta_stack, small_delta_stack,
2224 sizeof(*delta_stack)*delta_stack_nr);
2225 } else {
2226 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2227 }
2228 i = delta_stack_nr++;
2229 delta_stack[i].obj_offset = obj_offset;
2230 delta_stack[i].curpos = curpos;
2231 delta_stack[i].size = size;
2232
2233 curpos = obj_offset = base_offset;
2234 }
2235
2236 /* PHASE 2: handle the base */
2237 switch (type) {
2238 case OBJ_OFS_DELTA:
2239 case OBJ_REF_DELTA:
2240 if (data)
2241 die("BUG in unpack_entry: left loop at a valid delta");
2242 break;
2243 case OBJ_COMMIT:
2244 case OBJ_TREE:
2245 case OBJ_BLOB:
2246 case OBJ_TAG:
2247 if (!base_from_cache)
2248 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2249 break;
2250 default:
2251 data = NULL;
2252 error("unknown object type %i at offset %"PRIuMAX" in %s",
2253 type, (uintmax_t)obj_offset, p->pack_name);
2254 }
2255
2256 /* PHASE 3: apply deltas in order */
2257
2258 /* invariants:
2259 * 'data' holds the base data, or NULL if there was corruption
2260 */
2261 while (delta_stack_nr) {
2262 void *delta_data;
2263 void *base = data;
2264 unsigned long delta_size, base_size = size;
2265 int i;
2266
2267 data = NULL;
2268
2269 if (base)
2270 add_delta_base_cache(p, obj_offset, base, base_size, type);
2271
2272 if (!base) {
2273 /*
2274 * We're probably in deep shit, but let's try to fetch
2275 * the required base anyway from another pack or loose.
2276 * This is costly but should happen only in the presence
2277 * of a corrupted pack, and is better than failing outright.
2278 */
2279 struct revindex_entry *revidx;
2280 const unsigned char *base_sha1;
2281 revidx = find_pack_revindex(p, obj_offset);
2282 if (revidx) {
2283 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2284 error("failed to read delta base object %s"
2285 " at offset %"PRIuMAX" from %s",
2286 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2287 p->pack_name);
2288 mark_bad_packed_object(p, base_sha1);
2289 base = read_object(base_sha1, &type, &base_size);
2290 }
2291 }
2292
2293 i = --delta_stack_nr;
2294 obj_offset = delta_stack[i].obj_offset;
2295 curpos = delta_stack[i].curpos;
2296 delta_size = delta_stack[i].size;
2297
2298 if (!base)
2299 continue;
2300
2301 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2302
2303 if (!delta_data) {
2304 error("failed to unpack compressed delta "
2305 "at offset %"PRIuMAX" from %s",
2306 (uintmax_t)curpos, p->pack_name);
2307 data = NULL;
2308 continue;
2309 }
2310
2311 data = patch_delta(base, base_size,
2312 delta_data, delta_size,
2313 &size);
2314
2315 /*
2316 * We could not apply the delta; warn the user, but keep going.
2317 * Our failure will be noticed either in the next iteration of
2318 * the loop, or if this is the final delta, in the caller when
2319 * we return NULL. Those code paths will take care of making
2320 * a more explicit warning and retrying with another copy of
2321 * the object.
2322 */
2323 if (!data)
2324 error("failed to apply delta");
2325
2326 free(delta_data);
2327 }
2328
2329 *final_type = type;
2330 *final_size = size;
2331
2332 unuse_pack(&w_curs);
2333
2334 if (delta_stack != small_delta_stack)
2335 free(delta_stack);
2336
2337 return data;
2338}
2339
2340const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2341 uint32_t n)
2342{
2343 const unsigned char *index = p->index_data;
2344 if (!index) {
2345 if (open_pack_index(p))
2346 return NULL;
2347 index = p->index_data;
2348 }
2349 if (n >= p->num_objects)
2350 return NULL;
2351 index += 4 * 256;
2352 if (p->index_version == 1) {
2353 return index + 24 * n + 4;
2354 } else {
2355 index += 8;
2356 return index + 20 * n;
2357 }
2358}
2359
2360off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2361{
2362 const unsigned char *index = p->index_data;
2363 index += 4 * 256;
2364 if (p->index_version == 1) {
2365 return ntohl(*((uint32_t *)(index + 24 * n)));
2366 } else {
2367 uint32_t off;
2368 index += 8 + p->num_objects * (20 + 4);
2369 off = ntohl(*((uint32_t *)(index + 4 * n)));
2370 if (!(off & 0x80000000))
2371 return off;
2372 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2373 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2374 ntohl(*((uint32_t *)(index + 4)));
2375 }
2376}
2377
2378off_t find_pack_entry_one(const unsigned char *sha1,
2379 struct packed_git *p)
2380{
2381 const uint32_t *level1_ofs = p->index_data;
2382 const unsigned char *index = p->index_data;
2383 unsigned hi, lo, stride;
2384 static int use_lookup = -1;
2385 static int debug_lookup = -1;
2386
2387 if (debug_lookup < 0)
2388 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2389
2390 if (!index) {
2391 if (open_pack_index(p))
2392 return 0;
2393 level1_ofs = p->index_data;
2394 index = p->index_data;
2395 }
2396 if (p->index_version > 1) {
2397 level1_ofs += 2;
2398 index += 8;
2399 }
2400 index += 4 * 256;
2401 hi = ntohl(level1_ofs[*sha1]);
2402 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2403 if (p->index_version > 1) {
2404 stride = 20;
2405 } else {
2406 stride = 24;
2407 index += 4;
2408 }
2409
2410 if (debug_lookup)
2411 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2412 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2413
2414 if (use_lookup < 0)
2415 use_lookup = !!getenv("GIT_USE_LOOKUP");
2416 if (use_lookup) {
2417 int pos = sha1_entry_pos(index, stride, 0,
2418 lo, hi, p->num_objects, sha1);
2419 if (pos < 0)
2420 return 0;
2421 return nth_packed_object_offset(p, pos);
2422 }
2423
2424 do {
2425 unsigned mi = (lo + hi) / 2;
2426 int cmp = hashcmp(index + mi * stride, sha1);
2427
2428 if (debug_lookup)
2429 printf("lo %u hi %u rg %u mi %u\n",
2430 lo, hi, hi - lo, mi);
2431 if (!cmp)
2432 return nth_packed_object_offset(p, mi);
2433 if (cmp > 0)
2434 hi = mi;
2435 else
2436 lo = mi+1;
2437 } while (lo < hi);
2438 return 0;
2439}
2440
2441int is_pack_valid(struct packed_git *p)
2442{
2443 /* An already open pack is known to be valid. */
2444 if (p->pack_fd != -1)
2445 return 1;
2446
2447 /* If the pack has one window completely covering the
2448 * file size, the pack is known to be valid even if
2449 * the descriptor is not currently open.
2450 */
2451 if (p->windows) {
2452 struct pack_window *w = p->windows;
2453
2454 if (!w->offset && w->len == p->pack_size)
2455 return 1;
2456 }
2457
2458 /* Force the pack to open to prove its valid. */
2459 return !open_packed_git(p);
2460}
2461
2462static int fill_pack_entry(const unsigned char *sha1,
2463 struct pack_entry *e,
2464 struct packed_git *p)
2465{
2466 off_t offset;
2467
2468 if (p->num_bad_objects) {
2469 unsigned i;
2470 for (i = 0; i < p->num_bad_objects; i++)
2471 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2472 return 0;
2473 }
2474
2475 offset = find_pack_entry_one(sha1, p);
2476 if (!offset)
2477 return 0;
2478
2479 /*
2480 * We are about to tell the caller where they can locate the
2481 * requested object. We better make sure the packfile is
2482 * still here and can be accessed before supplying that
2483 * answer, as it may have been deleted since the index was
2484 * loaded!
2485 */
2486 if (!is_pack_valid(p))
2487 return 0;
2488 e->offset = offset;
2489 e->p = p;
2490 hashcpy(e->sha1, sha1);
2491 return 1;
2492}
2493
2494/*
2495 * Iff a pack file contains the object named by sha1, return true and
2496 * store its location to e.
2497 */
2498static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2499{
2500 struct packed_git *p;
2501
2502 prepare_packed_git();
2503 if (!packed_git)
2504 return 0;
2505
2506 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2507 return 1;
2508
2509 for (p = packed_git; p; p = p->next) {
2510 if (p == last_found_pack)
2511 continue; /* we already checked this one */
2512
2513 if (fill_pack_entry(sha1, e, p)) {
2514 last_found_pack = p;
2515 return 1;
2516 }
2517 }
2518 return 0;
2519}
2520
2521struct packed_git *find_sha1_pack(const unsigned char *sha1,
2522 struct packed_git *packs)
2523{
2524 struct packed_git *p;
2525
2526 for (p = packs; p; p = p->next) {
2527 if (find_pack_entry_one(sha1, p))
2528 return p;
2529 }
2530 return NULL;
2531
2532}
2533
2534static int sha1_loose_object_info(const unsigned char *sha1,
2535 struct object_info *oi)
2536{
2537 int status;
2538 unsigned long mapsize, size;
2539 void *map;
2540 git_zstream stream;
2541 char hdr[32];
2542
2543 if (oi->delta_base_sha1)
2544 hashclr(oi->delta_base_sha1);
2545
2546 /*
2547 * If we don't care about type or size, then we don't
2548 * need to look inside the object at all. Note that we
2549 * do not optimize out the stat call, even if the
2550 * caller doesn't care about the disk-size, since our
2551 * return value implicitly indicates whether the
2552 * object even exists.
2553 */
2554 if (!oi->typep && !oi->sizep) {
2555 struct stat st;
2556 if (stat_sha1_file(sha1, &st) < 0)
2557 return -1;
2558 if (oi->disk_sizep)
2559 *oi->disk_sizep = st.st_size;
2560 return 0;
2561 }
2562
2563 map = map_sha1_file(sha1, &mapsize);
2564 if (!map)
2565 return -1;
2566 if (oi->disk_sizep)
2567 *oi->disk_sizep = mapsize;
2568 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2569 status = error("unable to unpack %s header",
2570 sha1_to_hex(sha1));
2571 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2572 status = error("unable to parse %s header", sha1_to_hex(sha1));
2573 else if (oi->sizep)
2574 *oi->sizep = size;
2575 git_inflate_end(&stream);
2576 munmap(map, mapsize);
2577 if (oi->typep)
2578 *oi->typep = status;
2579 return 0;
2580}
2581
2582int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2583{
2584 struct cached_object *co;
2585 struct pack_entry e;
2586 int rtype;
2587 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2588
2589 co = find_cached_object(real);
2590 if (co) {
2591 if (oi->typep)
2592 *(oi->typep) = co->type;
2593 if (oi->sizep)
2594 *(oi->sizep) = co->size;
2595 if (oi->disk_sizep)
2596 *(oi->disk_sizep) = 0;
2597 if (oi->delta_base_sha1)
2598 hashclr(oi->delta_base_sha1);
2599 oi->whence = OI_CACHED;
2600 return 0;
2601 }
2602
2603 if (!find_pack_entry(real, &e)) {
2604 /* Most likely it's a loose object. */
2605 if (!sha1_loose_object_info(real, oi)) {
2606 oi->whence = OI_LOOSE;
2607 return 0;
2608 }
2609
2610 /* Not a loose object; someone else may have just packed it. */
2611 reprepare_packed_git();
2612 if (!find_pack_entry(real, &e))
2613 return -1;
2614 }
2615
2616 rtype = packed_object_info(e.p, e.offset, oi);
2617 if (rtype < 0) {
2618 mark_bad_packed_object(e.p, real);
2619 return sha1_object_info_extended(real, oi, 0);
2620 } else if (in_delta_base_cache(e.p, e.offset)) {
2621 oi->whence = OI_DBCACHED;
2622 } else {
2623 oi->whence = OI_PACKED;
2624 oi->u.packed.offset = e.offset;
2625 oi->u.packed.pack = e.p;
2626 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2627 rtype == OBJ_OFS_DELTA);
2628 }
2629
2630 return 0;
2631}
2632
2633/* returns enum object_type or negative */
2634int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2635{
2636 enum object_type type;
2637 struct object_info oi = {NULL};
2638
2639 oi.typep = &type;
2640 oi.sizep = sizep;
2641 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2642 return -1;
2643 return type;
2644}
2645
2646static void *read_packed_sha1(const unsigned char *sha1,
2647 enum object_type *type, unsigned long *size)
2648{
2649 struct pack_entry e;
2650 void *data;
2651
2652 if (!find_pack_entry(sha1, &e))
2653 return NULL;
2654 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2655 if (!data) {
2656 /*
2657 * We're probably in deep shit, but let's try to fetch
2658 * the required object anyway from another pack or loose.
2659 * This should happen only in the presence of a corrupted
2660 * pack, and is better than failing outright.
2661 */
2662 error("failed to read object %s at offset %"PRIuMAX" from %s",
2663 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2664 mark_bad_packed_object(e.p, sha1);
2665 data = read_object(sha1, type, size);
2666 }
2667 return data;
2668}
2669
2670int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2671 unsigned char *sha1)
2672{
2673 struct cached_object *co;
2674
2675 hash_sha1_file(buf, len, typename(type), sha1);
2676 if (has_sha1_file(sha1) || find_cached_object(sha1))
2677 return 0;
2678 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2679 co = &cached_objects[cached_object_nr++];
2680 co->size = len;
2681 co->type = type;
2682 co->buf = xmalloc(len);
2683 memcpy(co->buf, buf, len);
2684 hashcpy(co->sha1, sha1);
2685 return 0;
2686}
2687
2688static void *read_object(const unsigned char *sha1, enum object_type *type,
2689 unsigned long *size)
2690{
2691 unsigned long mapsize;
2692 void *map, *buf;
2693 struct cached_object *co;
2694
2695 co = find_cached_object(sha1);
2696 if (co) {
2697 *type = co->type;
2698 *size = co->size;
2699 return xmemdupz(co->buf, co->size);
2700 }
2701
2702 buf = read_packed_sha1(sha1, type, size);
2703 if (buf)
2704 return buf;
2705 map = map_sha1_file(sha1, &mapsize);
2706 if (map) {
2707 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2708 munmap(map, mapsize);
2709 return buf;
2710 }
2711 reprepare_packed_git();
2712 return read_packed_sha1(sha1, type, size);
2713}
2714
2715/*
2716 * This function dies on corrupt objects; the callers who want to
2717 * deal with them should arrange to call read_object() and give error
2718 * messages themselves.
2719 */
2720void *read_sha1_file_extended(const unsigned char *sha1,
2721 enum object_type *type,
2722 unsigned long *size,
2723 unsigned flag)
2724{
2725 void *data;
2726 const struct packed_git *p;
2727 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2728
2729 errno = 0;
2730 data = read_object(repl, type, size);
2731 if (data)
2732 return data;
2733
2734 if (errno && errno != ENOENT)
2735 die_errno("failed to read object %s", sha1_to_hex(sha1));
2736
2737 /* die if we replaced an object with one that does not exist */
2738 if (repl != sha1)
2739 die("replacement %s not found for %s",
2740 sha1_to_hex(repl), sha1_to_hex(sha1));
2741
2742 if (has_loose_object(repl)) {
2743 const char *path = sha1_file_name(sha1);
2744
2745 die("loose object %s (stored in %s) is corrupt",
2746 sha1_to_hex(repl), path);
2747 }
2748
2749 if ((p = has_packed_and_bad(repl)) != NULL)
2750 die("packed object %s (stored in %s) is corrupt",
2751 sha1_to_hex(repl), p->pack_name);
2752
2753 return NULL;
2754}
2755
2756void *read_object_with_reference(const unsigned char *sha1,
2757 const char *required_type_name,
2758 unsigned long *size,
2759 unsigned char *actual_sha1_return)
2760{
2761 enum object_type type, required_type;
2762 void *buffer;
2763 unsigned long isize;
2764 unsigned char actual_sha1[20];
2765
2766 required_type = type_from_string(required_type_name);
2767 hashcpy(actual_sha1, sha1);
2768 while (1) {
2769 int ref_length = -1;
2770 const char *ref_type = NULL;
2771
2772 buffer = read_sha1_file(actual_sha1, &type, &isize);
2773 if (!buffer)
2774 return NULL;
2775 if (type == required_type) {
2776 *size = isize;
2777 if (actual_sha1_return)
2778 hashcpy(actual_sha1_return, actual_sha1);
2779 return buffer;
2780 }
2781 /* Handle references */
2782 else if (type == OBJ_COMMIT)
2783 ref_type = "tree ";
2784 else if (type == OBJ_TAG)
2785 ref_type = "object ";
2786 else {
2787 free(buffer);
2788 return NULL;
2789 }
2790 ref_length = strlen(ref_type);
2791
2792 if (ref_length + 40 > isize ||
2793 memcmp(buffer, ref_type, ref_length) ||
2794 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2795 free(buffer);
2796 return NULL;
2797 }
2798 free(buffer);
2799 /* Now we have the ID of the referred-to object in
2800 * actual_sha1. Check again. */
2801 }
2802}
2803
2804static void write_sha1_file_prepare(const void *buf, unsigned long len,
2805 const char *type, unsigned char *sha1,
2806 char *hdr, int *hdrlen)
2807{
2808 git_SHA_CTX c;
2809
2810 /* Generate the header */
2811 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2812
2813 /* Sha1.. */
2814 git_SHA1_Init(&c);
2815 git_SHA1_Update(&c, hdr, *hdrlen);
2816 git_SHA1_Update(&c, buf, len);
2817 git_SHA1_Final(sha1, &c);
2818}
2819
2820/*
2821 * Move the just written object into its final resting place.
2822 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2823 * "moving" is only a part of what it does, when no patch between
2824 * master to pu changes the call sites of this function.
2825 */
2826int move_temp_to_file(const char *tmpfile, const char *filename)
2827{
2828 int ret = 0;
2829
2830 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2831 goto try_rename;
2832 else if (link(tmpfile, filename))
2833 ret = errno;
2834
2835 /*
2836 * Coda hack - coda doesn't like cross-directory links,
2837 * so we fall back to a rename, which will mean that it
2838 * won't be able to check collisions, but that's not a
2839 * big deal.
2840 *
2841 * The same holds for FAT formatted media.
2842 *
2843 * When this succeeds, we just return. We have nothing
2844 * left to unlink.
2845 */
2846 if (ret && ret != EEXIST) {
2847 try_rename:
2848 if (!rename(tmpfile, filename))
2849 goto out;
2850 ret = errno;
2851 }
2852 unlink_or_warn(tmpfile);
2853 if (ret) {
2854 if (ret != EEXIST) {
2855 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2856 }
2857 /* FIXME!!! Collision check here ? */
2858 }
2859
2860out:
2861 if (adjust_shared_perm(filename))
2862 return error("unable to set permission to '%s'", filename);
2863 return 0;
2864}
2865
2866static int write_buffer(int fd, const void *buf, size_t len)
2867{
2868 if (write_in_full(fd, buf, len) < 0)
2869 return error("file write error (%s)", strerror(errno));
2870 return 0;
2871}
2872
2873int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2874 unsigned char *sha1)
2875{
2876 char hdr[32];
2877 int hdrlen;
2878 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2879 return 0;
2880}
2881
2882/* Finalize a file on disk, and close it. */
2883static void close_sha1_file(int fd)
2884{
2885 if (fsync_object_files)
2886 fsync_or_die(fd, "sha1 file");
2887 if (close(fd) != 0)
2888 die_errno("error when closing sha1 file");
2889}
2890
2891/* Size of directory component, including the ending '/' */
2892static inline int directory_size(const char *filename)
2893{
2894 const char *s = strrchr(filename, '/');
2895 if (!s)
2896 return 0;
2897 return s - filename + 1;
2898}
2899
2900/*
2901 * This creates a temporary file in the same directory as the final
2902 * 'filename'
2903 *
2904 * We want to avoid cross-directory filename renames, because those
2905 * can have problems on various filesystems (FAT, NFS, Coda).
2906 */
2907static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2908{
2909 int fd, dirlen = directory_size(filename);
2910
2911 if (dirlen + 20 > bufsiz) {
2912 errno = ENAMETOOLONG;
2913 return -1;
2914 }
2915 memcpy(buffer, filename, dirlen);
2916 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2917 fd = git_mkstemp_mode(buffer, 0444);
2918 if (fd < 0 && dirlen && errno == ENOENT) {
2919 /* Make sure the directory exists */
2920 memcpy(buffer, filename, dirlen);
2921 buffer[dirlen-1] = 0;
2922 if (mkdir(buffer, 0777) && errno != EEXIST)
2923 return -1;
2924 if (adjust_shared_perm(buffer))
2925 return -1;
2926
2927 /* Try again */
2928 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2929 fd = git_mkstemp_mode(buffer, 0444);
2930 }
2931 return fd;
2932}
2933
2934static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2935 const void *buf, unsigned long len, time_t mtime)
2936{
2937 int fd, ret;
2938 unsigned char compressed[4096];
2939 git_zstream stream;
2940 git_SHA_CTX c;
2941 unsigned char parano_sha1[20];
2942 static char tmp_file[PATH_MAX];
2943 const char *filename = sha1_file_name(sha1);
2944
2945 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2946 if (fd < 0) {
2947 if (errno == EACCES)
2948 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2949 else
2950 return error("unable to create temporary file: %s", strerror(errno));
2951 }
2952
2953 /* Set it up */
2954 git_deflate_init(&stream, zlib_compression_level);
2955 stream.next_out = compressed;
2956 stream.avail_out = sizeof(compressed);
2957 git_SHA1_Init(&c);
2958
2959 /* First header.. */
2960 stream.next_in = (unsigned char *)hdr;
2961 stream.avail_in = hdrlen;
2962 while (git_deflate(&stream, 0) == Z_OK)
2963 ; /* nothing */
2964 git_SHA1_Update(&c, hdr, hdrlen);
2965
2966 /* Then the data itself.. */
2967 stream.next_in = (void *)buf;
2968 stream.avail_in = len;
2969 do {
2970 unsigned char *in0 = stream.next_in;
2971 ret = git_deflate(&stream, Z_FINISH);
2972 git_SHA1_Update(&c, in0, stream.next_in - in0);
2973 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2974 die("unable to write sha1 file");
2975 stream.next_out = compressed;
2976 stream.avail_out = sizeof(compressed);
2977 } while (ret == Z_OK);
2978
2979 if (ret != Z_STREAM_END)
2980 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2981 ret = git_deflate_end_gently(&stream);
2982 if (ret != Z_OK)
2983 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2984 git_SHA1_Final(parano_sha1, &c);
2985 if (hashcmp(sha1, parano_sha1) != 0)
2986 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2987
2988 close_sha1_file(fd);
2989
2990 if (mtime) {
2991 struct utimbuf utb;
2992 utb.actime = mtime;
2993 utb.modtime = mtime;
2994 if (utime(tmp_file, &utb) < 0)
2995 warning("failed utime() on %s: %s",
2996 tmp_file, strerror(errno));
2997 }
2998
2999 return move_temp_to_file(tmp_file, filename);
3000}
3001
3002static int freshen_loose_object(const unsigned char *sha1)
3003{
3004 return check_and_freshen(sha1, 1);
3005}
3006
3007static int freshen_packed_object(const unsigned char *sha1)
3008{
3009 struct pack_entry e;
3010 if (!find_pack_entry(sha1, &e))
3011 return 0;
3012 if (e.p->freshened)
3013 return 1;
3014 if (!freshen_file(e.p->pack_name))
3015 return 0;
3016 e.p->freshened = 1;
3017 return 1;
3018}
3019
3020int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3021{
3022 char hdr[32];
3023 int hdrlen;
3024
3025 /* Normally if we have it in the pack then we do not bother writing
3026 * it out into .git/objects/??/?{38} file.
3027 */
3028 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3029 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3030 return 0;
3031 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3032}
3033
3034int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3035 unsigned char *sha1, unsigned flags)
3036{
3037 char *header;
3038 int hdrlen, status = 0;
3039
3040 /* type string, SP, %lu of the length plus NUL must fit this */
3041 header = xmalloc(strlen(type) + 32);
3042 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3043
3044 if (!(flags & HASH_WRITE_OBJECT))
3045 goto cleanup;
3046 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3047 goto cleanup;
3048 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3049
3050cleanup:
3051 free(header);
3052 return status;
3053}
3054
3055int force_object_loose(const unsigned char *sha1, time_t mtime)
3056{
3057 void *buf;
3058 unsigned long len;
3059 enum object_type type;
3060 char hdr[32];
3061 int hdrlen;
3062 int ret;
3063
3064 if (has_loose_object(sha1))
3065 return 0;
3066 buf = read_packed_sha1(sha1, &type, &len);
3067 if (!buf)
3068 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3069 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3070 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3071 free(buf);
3072
3073 return ret;
3074}
3075
3076int has_pack_index(const unsigned char *sha1)
3077{
3078 struct stat st;
3079 if (stat(sha1_pack_index_name(sha1), &st))
3080 return 0;
3081 return 1;
3082}
3083
3084int has_sha1_pack(const unsigned char *sha1)
3085{
3086 struct pack_entry e;
3087 return find_pack_entry(sha1, &e);
3088}
3089
3090int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3091{
3092 struct pack_entry e;
3093
3094 if (find_pack_entry(sha1, &e))
3095 return 1;
3096 if (has_loose_object(sha1))
3097 return 1;
3098 if (flags & HAS_SHA1_QUICK)
3099 return 0;
3100 reprepare_packed_git();
3101 return find_pack_entry(sha1, &e);
3102}
3103
3104static void check_tree(const void *buf, size_t size)
3105{
3106 struct tree_desc desc;
3107 struct name_entry entry;
3108
3109 init_tree_desc(&desc, buf, size);
3110 while (tree_entry(&desc, &entry))
3111 /* do nothing
3112 * tree_entry() will die() on malformed entries */
3113 ;
3114}
3115
3116static void check_commit(const void *buf, size_t size)
3117{
3118 struct commit c;
3119 memset(&c, 0, sizeof(c));
3120 if (parse_commit_buffer(&c, buf, size))
3121 die("corrupt commit");
3122}
3123
3124static void check_tag(const void *buf, size_t size)
3125{
3126 struct tag t;
3127 memset(&t, 0, sizeof(t));
3128 if (parse_tag_buffer(&t, buf, size))
3129 die("corrupt tag");
3130}
3131
3132static int index_mem(unsigned char *sha1, void *buf, size_t size,
3133 enum object_type type,
3134 const char *path, unsigned flags)
3135{
3136 int ret, re_allocated = 0;
3137 int write_object = flags & HASH_WRITE_OBJECT;
3138
3139 if (!type)
3140 type = OBJ_BLOB;
3141
3142 /*
3143 * Convert blobs to git internal format
3144 */
3145 if ((type == OBJ_BLOB) && path) {
3146 struct strbuf nbuf = STRBUF_INIT;
3147 if (convert_to_git(path, buf, size, &nbuf,
3148 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3149 buf = strbuf_detach(&nbuf, &size);
3150 re_allocated = 1;
3151 }
3152 }
3153 if (flags & HASH_FORMAT_CHECK) {
3154 if (type == OBJ_TREE)
3155 check_tree(buf, size);
3156 if (type == OBJ_COMMIT)
3157 check_commit(buf, size);
3158 if (type == OBJ_TAG)
3159 check_tag(buf, size);
3160 }
3161
3162 if (write_object)
3163 ret = write_sha1_file(buf, size, typename(type), sha1);
3164 else
3165 ret = hash_sha1_file(buf, size, typename(type), sha1);
3166 if (re_allocated)
3167 free(buf);
3168 return ret;
3169}
3170
3171static int index_stream_convert_blob(unsigned char *sha1, int fd,
3172 const char *path, unsigned flags)
3173{
3174 int ret;
3175 const int write_object = flags & HASH_WRITE_OBJECT;
3176 struct strbuf sbuf = STRBUF_INIT;
3177
3178 assert(path);
3179 assert(would_convert_to_git_filter_fd(path));
3180
3181 convert_to_git_filter_fd(path, fd, &sbuf,
3182 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3183
3184 if (write_object)
3185 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3186 sha1);
3187 else
3188 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3189 sha1);
3190 strbuf_release(&sbuf);
3191 return ret;
3192}
3193
3194static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3195 const char *path, unsigned flags)
3196{
3197 struct strbuf sbuf = STRBUF_INIT;
3198 int ret;
3199
3200 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3201 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3202 else
3203 ret = -1;
3204 strbuf_release(&sbuf);
3205 return ret;
3206}
3207
3208#define SMALL_FILE_SIZE (32*1024)
3209
3210static int index_core(unsigned char *sha1, int fd, size_t size,
3211 enum object_type type, const char *path,
3212 unsigned flags)
3213{
3214 int ret;
3215
3216 if (!size) {
3217 ret = index_mem(sha1, "", size, type, path, flags);
3218 } else if (size <= SMALL_FILE_SIZE) {
3219 char *buf = xmalloc(size);
3220 if (size == read_in_full(fd, buf, size))
3221 ret = index_mem(sha1, buf, size, type, path, flags);
3222 else
3223 ret = error("short read %s", strerror(errno));
3224 free(buf);
3225 } else {
3226 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3227 ret = index_mem(sha1, buf, size, type, path, flags);
3228 munmap(buf, size);
3229 }
3230 return ret;
3231}
3232
3233/*
3234 * This creates one packfile per large blob unless bulk-checkin
3235 * machinery is "plugged".
3236 *
3237 * This also bypasses the usual "convert-to-git" dance, and that is on
3238 * purpose. We could write a streaming version of the converting
3239 * functions and insert that before feeding the data to fast-import
3240 * (or equivalent in-core API described above). However, that is
3241 * somewhat complicated, as we do not know the size of the filter
3242 * result, which we need to know beforehand when writing a git object.
3243 * Since the primary motivation for trying to stream from the working
3244 * tree file and to avoid mmaping it in core is to deal with large
3245 * binary blobs, they generally do not want to get any conversion, and
3246 * callers should avoid this code path when filters are requested.
3247 */
3248static int index_stream(unsigned char *sha1, int fd, size_t size,
3249 enum object_type type, const char *path,
3250 unsigned flags)
3251{
3252 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3253}
3254
3255int index_fd(unsigned char *sha1, int fd, struct stat *st,
3256 enum object_type type, const char *path, unsigned flags)
3257{
3258 int ret;
3259
3260 /*
3261 * Call xsize_t() only when needed to avoid potentially unnecessary
3262 * die() for large files.
3263 */
3264 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3265 ret = index_stream_convert_blob(sha1, fd, path, flags);
3266 else if (!S_ISREG(st->st_mode))
3267 ret = index_pipe(sha1, fd, type, path, flags);
3268 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3269 (path && would_convert_to_git(path)))
3270 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3271 flags);
3272 else
3273 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3274 flags);
3275 close(fd);
3276 return ret;
3277}
3278
3279int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3280{
3281 int fd;
3282 struct strbuf sb = STRBUF_INIT;
3283
3284 switch (st->st_mode & S_IFMT) {
3285 case S_IFREG:
3286 fd = open(path, O_RDONLY);
3287 if (fd < 0)
3288 return error("open(\"%s\"): %s", path,
3289 strerror(errno));
3290 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3291 return error("%s: failed to insert into database",
3292 path);
3293 break;
3294 case S_IFLNK:
3295 if (strbuf_readlink(&sb, path, st->st_size)) {
3296 char *errstr = strerror(errno);
3297 return error("readlink(\"%s\"): %s", path,
3298 errstr);
3299 }
3300 if (!(flags & HASH_WRITE_OBJECT))
3301 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3302 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3303 return error("%s: failed to insert into database",
3304 path);
3305 strbuf_release(&sb);
3306 break;
3307 case S_IFDIR:
3308 return resolve_gitlink_ref(path, "HEAD", sha1);
3309 default:
3310 return error("%s: unsupported file type", path);
3311 }
3312 return 0;
3313}
3314
3315int read_pack_header(int fd, struct pack_header *header)
3316{
3317 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3318 /* "eof before pack header was fully read" */
3319 return PH_ERROR_EOF;
3320
3321 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3322 /* "protocol error (pack signature mismatch detected)" */
3323 return PH_ERROR_PACK_SIGNATURE;
3324 if (!pack_version_ok(header->hdr_version))
3325 /* "protocol error (pack version unsupported)" */
3326 return PH_ERROR_PROTOCOL;
3327 return 0;
3328}
3329
3330void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3331{
3332 enum object_type type = sha1_object_info(sha1, NULL);
3333 if (type < 0)
3334 die("%s is not a valid object", sha1_to_hex(sha1));
3335 if (type != expect)
3336 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3337 typename(expect));
3338}
3339
3340static int for_each_file_in_obj_subdir(int subdir_nr,
3341 struct strbuf *path,
3342 each_loose_object_fn obj_cb,
3343 each_loose_cruft_fn cruft_cb,
3344 each_loose_subdir_fn subdir_cb,
3345 void *data)
3346{
3347 size_t baselen = path->len;
3348 DIR *dir = opendir(path->buf);
3349 struct dirent *de;
3350 int r = 0;
3351
3352 if (!dir) {
3353 if (errno == ENOENT)
3354 return 0;
3355 return error("unable to open %s: %s", path->buf, strerror(errno));
3356 }
3357
3358 while ((de = readdir(dir))) {
3359 if (is_dot_or_dotdot(de->d_name))
3360 continue;
3361
3362 strbuf_setlen(path, baselen);
3363 strbuf_addf(path, "/%s", de->d_name);
3364
3365 if (strlen(de->d_name) == 38) {
3366 char hex[41];
3367 unsigned char sha1[20];
3368
3369 snprintf(hex, sizeof(hex), "%02x%s",
3370 subdir_nr, de->d_name);
3371 if (!get_sha1_hex(hex, sha1)) {
3372 if (obj_cb) {
3373 r = obj_cb(sha1, path->buf, data);
3374 if (r)
3375 break;
3376 }
3377 continue;
3378 }
3379 }
3380
3381 if (cruft_cb) {
3382 r = cruft_cb(de->d_name, path->buf, data);
3383 if (r)
3384 break;
3385 }
3386 }
3387 strbuf_setlen(path, baselen);
3388
3389 if (!r && subdir_cb)
3390 r = subdir_cb(subdir_nr, path->buf, data);
3391
3392 closedir(dir);
3393 return r;
3394}
3395
3396int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3397 each_loose_object_fn obj_cb,
3398 each_loose_cruft_fn cruft_cb,
3399 each_loose_subdir_fn subdir_cb,
3400 void *data)
3401{
3402 size_t baselen = path->len;
3403 int r = 0;
3404 int i;
3405
3406 for (i = 0; i < 256; i++) {
3407 strbuf_addf(path, "/%02x", i);
3408 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3409 subdir_cb, data);
3410 strbuf_setlen(path, baselen);
3411 if (r)
3412 break;
3413 }
3414
3415 return r;
3416}
3417
3418int for_each_loose_file_in_objdir(const char *path,
3419 each_loose_object_fn obj_cb,
3420 each_loose_cruft_fn cruft_cb,
3421 each_loose_subdir_fn subdir_cb,
3422 void *data)
3423{
3424 struct strbuf buf = STRBUF_INIT;
3425 int r;
3426
3427 strbuf_addstr(&buf, path);
3428 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3429 subdir_cb, data);
3430 strbuf_release(&buf);
3431
3432 return r;
3433}
3434
3435struct loose_alt_odb_data {
3436 each_loose_object_fn *cb;
3437 void *data;
3438};
3439
3440static int loose_from_alt_odb(struct alternate_object_database *alt,
3441 void *vdata)
3442{
3443 struct loose_alt_odb_data *data = vdata;
3444 struct strbuf buf = STRBUF_INIT;
3445 int r;
3446
3447 /* copy base not including trailing '/' */
3448 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3449 r = for_each_loose_file_in_objdir_buf(&buf,
3450 data->cb, NULL, NULL,
3451 data->data);
3452 strbuf_release(&buf);
3453 return r;
3454}
3455
3456int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3457{
3458 struct loose_alt_odb_data alt;
3459 int r;
3460
3461 r = for_each_loose_file_in_objdir(get_object_directory(),
3462 cb, NULL, NULL, data);
3463 if (r)
3464 return r;
3465
3466 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3467 return 0;
3468
3469 alt.cb = cb;
3470 alt.data = data;
3471 return foreach_alt_odb(loose_from_alt_odb, &alt);
3472}
3473
3474static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3475{
3476 uint32_t i;
3477 int r = 0;
3478
3479 for (i = 0; i < p->num_objects; i++) {
3480 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3481
3482 if (!sha1)
3483 return error("unable to get sha1 of object %u in %s",
3484 i, p->pack_name);
3485
3486 r = cb(sha1, p, i, data);
3487 if (r)
3488 break;
3489 }
3490 return r;
3491}
3492
3493int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3494{
3495 struct packed_git *p;
3496 int r = 0;
3497
3498 prepare_packed_git();
3499 for (p = packed_git; p; p = p->next) {
3500 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3501 continue;
3502 r = for_each_object_in_pack(p, cb, data);
3503 if (r)
3504 break;
3505 }
3506 return r;
3507}