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];
39const struct object_id null_oid;
40
41/*
42 * This is meant to hold a *small* number of objects that you would
43 * want read_sha1_file() to be able to return, but yet you do not want
44 * to write them into the object store (e.g. a browse-only
45 * application).
46 */
47static struct cached_object {
48 unsigned char sha1[20];
49 enum object_type type;
50 void *buf;
51 unsigned long size;
52} *cached_objects;
53static int cached_object_nr, cached_object_alloc;
54
55static struct cached_object empty_tree = {
56 EMPTY_TREE_SHA1_BIN_LITERAL,
57 OBJ_TREE,
58 "",
59 0
60};
61
62/*
63 * A pointer to the last packed_git in which an object was found.
64 * When an object is sought, we look in this packfile first, because
65 * objects that are looked up at similar times are often in the same
66 * packfile as one another.
67 */
68static struct packed_git *last_found_pack;
69
70static struct cached_object *find_cached_object(const unsigned char *sha1)
71{
72 int i;
73 struct cached_object *co = cached_objects;
74
75 for (i = 0; i < cached_object_nr; i++, co++) {
76 if (!hashcmp(co->sha1, sha1))
77 return co;
78 }
79 if (!hashcmp(sha1, empty_tree.sha1))
80 return &empty_tree;
81 return NULL;
82}
83
84int mkdir_in_gitdir(const char *path)
85{
86 if (mkdir(path, 0777)) {
87 int saved_errno = errno;
88 struct stat st;
89 struct strbuf sb = STRBUF_INIT;
90
91 if (errno != EEXIST)
92 return -1;
93 /*
94 * Are we looking at a path in a symlinked worktree
95 * whose original repository does not yet have it?
96 * e.g. .git/rr-cache pointing at its original
97 * repository in which the user hasn't performed any
98 * conflict resolution yet?
99 */
100 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
101 strbuf_readlink(&sb, path, st.st_size) ||
102 !is_absolute_path(sb.buf) ||
103 mkdir(sb.buf, 0777)) {
104 strbuf_release(&sb);
105 errno = saved_errno;
106 return -1;
107 }
108 strbuf_release(&sb);
109 }
110 return adjust_shared_perm(path);
111}
112
113enum scld_error safe_create_leading_directories(char *path)
114{
115 char *next_component = path + offset_1st_component(path);
116 enum scld_error ret = SCLD_OK;
117
118 while (ret == SCLD_OK && next_component) {
119 struct stat st;
120 char *slash = next_component, slash_character;
121
122 while (*slash && !is_dir_sep(*slash))
123 slash++;
124
125 if (!*slash)
126 break;
127
128 next_component = slash + 1;
129 while (is_dir_sep(*next_component))
130 next_component++;
131 if (!*next_component)
132 break;
133
134 slash_character = *slash;
135 *slash = '\0';
136 if (!stat(path, &st)) {
137 /* path exists */
138 if (!S_ISDIR(st.st_mode))
139 ret = SCLD_EXISTS;
140 } else if (mkdir(path, 0777)) {
141 if (errno == EEXIST &&
142 !stat(path, &st) && S_ISDIR(st.st_mode))
143 ; /* somebody created it since we checked */
144 else if (errno == ENOENT)
145 /*
146 * Either mkdir() failed because
147 * somebody just pruned the containing
148 * directory, or stat() failed because
149 * the file that was in our way was
150 * just removed. Either way, inform
151 * the caller that it might be worth
152 * trying again:
153 */
154 ret = SCLD_VANISHED;
155 else
156 ret = SCLD_FAILED;
157 } else if (adjust_shared_perm(path)) {
158 ret = SCLD_PERMS;
159 }
160 *slash = slash_character;
161 }
162 return ret;
163}
164
165enum scld_error safe_create_leading_directories_const(const char *path)
166{
167 /* path points to cache entries, so xstrdup before messing with it */
168 char *buf = xstrdup(path);
169 enum scld_error result = safe_create_leading_directories(buf);
170 free(buf);
171 return result;
172}
173
174static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175{
176 int i;
177 for (i = 0; i < 20; i++) {
178 static char hex[] = "0123456789abcdef";
179 unsigned int val = sha1[i];
180 char *pos = pathbuf + i*2 + (i > 0);
181 *pos++ = hex[val >> 4];
182 *pos = hex[val & 0xf];
183 }
184}
185
186const char *sha1_file_name(const unsigned char *sha1)
187{
188 static char buf[PATH_MAX];
189 const char *objdir;
190 int len;
191
192 objdir = get_object_directory();
193 len = strlen(objdir);
194
195 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
196 if (len + 43 > PATH_MAX)
197 die("insanely long object directory %s", objdir);
198 memcpy(buf, objdir, len);
199 buf[len] = '/';
200 buf[len+3] = '/';
201 buf[len+42] = '\0';
202 fill_sha1_path(buf + len + 1, sha1);
203 return buf;
204}
205
206/*
207 * Return the name of the pack or index file with the specified sha1
208 * in its filename. *base and *name are scratch space that must be
209 * provided by the caller. which should be "pack" or "idx".
210 */
211static char *sha1_get_pack_name(const unsigned char *sha1,
212 struct strbuf *buf,
213 const char *which)
214{
215 strbuf_reset(buf);
216 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
217 sha1_to_hex(sha1), which);
218 return buf->buf;
219}
220
221char *sha1_pack_name(const unsigned char *sha1)
222{
223 static struct strbuf buf = STRBUF_INIT;
224 return sha1_get_pack_name(sha1, &buf, "pack");
225}
226
227char *sha1_pack_index_name(const unsigned char *sha1)
228{
229 static struct strbuf buf = STRBUF_INIT;
230 return sha1_get_pack_name(sha1, &buf, "idx");
231}
232
233struct alternate_object_database *alt_odb_list;
234static struct alternate_object_database **alt_odb_tail;
235
236/*
237 * Prepare alternate object database registry.
238 *
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
250 */
251static int link_alt_odb_entry(const char *entry, const char *relative_base,
252 int depth, const char *normalized_objdir)
253{
254 struct alternate_object_database *ent;
255 struct alternate_object_database *alt;
256 int pfxlen, entlen;
257 struct strbuf pathbuf = STRBUF_INIT;
258
259 if (!is_absolute_path(entry) && relative_base) {
260 strbuf_addstr(&pathbuf, real_path(relative_base));
261 strbuf_addch(&pathbuf, '/');
262 }
263 strbuf_addstr(&pathbuf, entry);
264
265 normalize_path_copy(pathbuf.buf, pathbuf.buf);
266
267 pfxlen = strlen(pathbuf.buf);
268
269 /*
270 * The trailing slash after the directory name is given by
271 * this function at the end. Remove duplicates.
272 */
273 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
274 pfxlen -= 1;
275
276 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent = xmalloc(sizeof(*ent) + entlen);
278 memcpy(ent->base, pathbuf.buf, pfxlen);
279 strbuf_release(&pathbuf);
280
281 ent->name = ent->base + pfxlen + 1;
282 ent->base[pfxlen + 3] = '/';
283 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284
285 /* Detect cases where alternate disappeared */
286 if (!is_directory(ent->base)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
289 ent->base);
290 free(ent);
291 return -1;
292 }
293
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
296 */
297 for (alt = alt_odb_list; alt; alt = alt->next) {
298 if (pfxlen == alt->name - alt->base - 1 &&
299 !memcmp(ent->base, alt->base, pfxlen)) {
300 free(ent);
301 return -1;
302 }
303 }
304 if (!strcmp_icase(ent->base, normalized_objdir)) {
305 free(ent);
306 return -1;
307 }
308
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
313
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
316
317 ent->base[pfxlen] = '/';
318
319 return 0;
320}
321
322static void link_alt_odb_entries(const char *alt, int len, int sep,
323 const char *relative_base, int depth)
324{
325 struct string_list entries = STRING_LIST_INIT_NODUP;
326 char *alt_copy;
327 int i;
328 struct strbuf objdirbuf = STRBUF_INIT;
329
330 if (depth > 5) {
331 error("%s: ignoring alternate object stores, nesting too deep.",
332 relative_base);
333 return;
334 }
335
336 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
337 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
338
339 alt_copy = xmemdupz(alt, len);
340 string_list_split_in_place(&entries, alt_copy, sep, -1);
341 for (i = 0; i < entries.nr; i++) {
342 const char *entry = entries.items[i].string;
343 if (entry[0] == '\0' || entry[0] == '#')
344 continue;
345 if (!is_absolute_path(entry) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, entry);
348 } else {
349 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
350 }
351 }
352 string_list_clear(&entries, 0);
353 free(alt_copy);
354 strbuf_release(&objdirbuf);
355}
356
357void read_info_alternates(const char * relative_base, int depth)
358{
359 char *map;
360 size_t mapsz;
361 struct stat st;
362 char *path;
363 int fd;
364
365 path = xstrfmt("%s/info/alternates", relative_base);
366 fd = git_open_noatime(path);
367 free(path);
368 if (fd < 0)
369 return;
370 if (fstat(fd, &st) || (st.st_size == 0)) {
371 close(fd);
372 return;
373 }
374 mapsz = xsize_t(st.st_size);
375 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
376 close(fd);
377
378 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
379
380 munmap(map, mapsz);
381}
382
383void add_to_alternates_file(const char *reference)
384{
385 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
386 char *alts = git_pathdup("objects/info/alternates");
387 FILE *in, *out;
388
389 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
390 out = fdopen_lock_file(lock, "w");
391 if (!out)
392 die_errno("unable to fdopen alternates lockfile");
393
394 in = fopen(alts, "r");
395 if (in) {
396 struct strbuf line = STRBUF_INIT;
397 int found = 0;
398
399 while (strbuf_getline(&line, in) != EOF) {
400 if (!strcmp(reference, line.buf)) {
401 found = 1;
402 break;
403 }
404 fprintf_or_die(out, "%s\n", line.buf);
405 }
406
407 strbuf_release(&line);
408 fclose(in);
409
410 if (found) {
411 rollback_lock_file(lock);
412 lock = NULL;
413 }
414 }
415 else if (errno != ENOENT)
416 die_errno("unable to read alternates file");
417
418 if (lock) {
419 fprintf_or_die(out, "%s\n", reference);
420 if (commit_lock_file(lock))
421 die_errno("unable to move new alternates file into place");
422 if (alt_odb_tail)
423 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
424 }
425 free(alts);
426}
427
428int foreach_alt_odb(alt_odb_fn fn, void *cb)
429{
430 struct alternate_object_database *ent;
431 int r = 0;
432
433 prepare_alt_odb();
434 for (ent = alt_odb_list; ent; ent = ent->next) {
435 r = fn(ent, cb);
436 if (r)
437 break;
438 }
439 return r;
440}
441
442void prepare_alt_odb(void)
443{
444 const char *alt;
445
446 if (alt_odb_tail)
447 return;
448
449 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
450 if (!alt) alt = "";
451
452 alt_odb_tail = &alt_odb_list;
453 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
454
455 read_info_alternates(get_object_directory(), 0);
456}
457
458/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
459static int freshen_file(const char *fn)
460{
461 struct utimbuf t;
462 t.actime = t.modtime = time(NULL);
463 return !utime(fn, &t);
464}
465
466/*
467 * All of the check_and_freshen functions return 1 if the file exists and was
468 * freshened (if freshening was requested), 0 otherwise. If they return
469 * 0, you should not assume that it is safe to skip a write of the object (it
470 * either does not exist on disk, or has a stale mtime and may be subject to
471 * pruning).
472 */
473static int check_and_freshen_file(const char *fn, int freshen)
474{
475 if (access(fn, F_OK))
476 return 0;
477 if (freshen && !freshen_file(fn))
478 return 0;
479 return 1;
480}
481
482static int check_and_freshen_local(const unsigned char *sha1, int freshen)
483{
484 return check_and_freshen_file(sha1_file_name(sha1), freshen);
485}
486
487static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
488{
489 struct alternate_object_database *alt;
490 prepare_alt_odb();
491 for (alt = alt_odb_list; alt; alt = alt->next) {
492 fill_sha1_path(alt->name, sha1);
493 if (check_and_freshen_file(alt->base, freshen))
494 return 1;
495 }
496 return 0;
497}
498
499static int check_and_freshen(const unsigned char *sha1, int freshen)
500{
501 return check_and_freshen_local(sha1, freshen) ||
502 check_and_freshen_nonlocal(sha1, freshen);
503}
504
505int has_loose_object_nonlocal(const unsigned char *sha1)
506{
507 return check_and_freshen_nonlocal(sha1, 0);
508}
509
510static int has_loose_object(const unsigned char *sha1)
511{
512 return check_and_freshen(sha1, 0);
513}
514
515static unsigned int pack_used_ctr;
516static unsigned int pack_mmap_calls;
517static unsigned int peak_pack_open_windows;
518static unsigned int pack_open_windows;
519static unsigned int pack_open_fds;
520static unsigned int pack_max_fds;
521static size_t peak_pack_mapped;
522static size_t pack_mapped;
523struct packed_git *packed_git;
524
525void pack_report(void)
526{
527 fprintf(stderr,
528 "pack_report: getpagesize() = %10" SZ_FMT "\n"
529 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
530 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
531 sz_fmt(getpagesize()),
532 sz_fmt(packed_git_window_size),
533 sz_fmt(packed_git_limit));
534 fprintf(stderr,
535 "pack_report: pack_used_ctr = %10u\n"
536 "pack_report: pack_mmap_calls = %10u\n"
537 "pack_report: pack_open_windows = %10u / %10u\n"
538 "pack_report: pack_mapped = "
539 "%10" SZ_FMT " / %10" SZ_FMT "\n",
540 pack_used_ctr,
541 pack_mmap_calls,
542 pack_open_windows, peak_pack_open_windows,
543 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
544}
545
546/*
547 * Open and mmap the index file at path, perform a couple of
548 * consistency checks, then record its information to p. Return 0 on
549 * success.
550 */
551static int check_packed_git_idx(const char *path, struct packed_git *p)
552{
553 void *idx_map;
554 struct pack_idx_header *hdr;
555 size_t idx_size;
556 uint32_t version, nr, i, *index;
557 int fd = git_open_noatime(path);
558 struct stat st;
559
560 if (fd < 0)
561 return -1;
562 if (fstat(fd, &st)) {
563 close(fd);
564 return -1;
565 }
566 idx_size = xsize_t(st.st_size);
567 if (idx_size < 4 * 256 + 20 + 20) {
568 close(fd);
569 return error("index file %s is too small", path);
570 }
571 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
572 close(fd);
573
574 hdr = idx_map;
575 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
576 version = ntohl(hdr->idx_version);
577 if (version < 2 || version > 2) {
578 munmap(idx_map, idx_size);
579 return error("index file %s is version %"PRIu32
580 " and is not supported by this binary"
581 " (try upgrading GIT to a newer version)",
582 path, version);
583 }
584 } else
585 version = 1;
586
587 nr = 0;
588 index = idx_map;
589 if (version > 1)
590 index += 2; /* skip index header */
591 for (i = 0; i < 256; i++) {
592 uint32_t n = ntohl(index[i]);
593 if (n < nr) {
594 munmap(idx_map, idx_size);
595 return error("non-monotonic index %s", path);
596 }
597 nr = n;
598 }
599
600 if (version == 1) {
601 /*
602 * Total size:
603 * - 256 index entries 4 bytes each
604 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 */
608 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
609 munmap(idx_map, idx_size);
610 return error("wrong index v1 file size in %s", path);
611 }
612 } else if (version == 2) {
613 /*
614 * Minimum size:
615 * - 8 bytes of header
616 * - 256 index entries 4 bytes each
617 * - 20-byte sha1 entry * nr
618 * - 4-byte crc entry * nr
619 * - 4-byte offset entry * nr
620 * - 20-byte SHA1 of the packfile
621 * - 20-byte SHA1 file checksum
622 * And after the 4-byte offset table might be a
623 * variable sized table containing 8-byte entries
624 * for offsets larger than 2^31.
625 */
626 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
627 unsigned long max_size = min_size;
628 if (nr)
629 max_size += (nr - 1)*8;
630 if (idx_size < min_size || idx_size > max_size) {
631 munmap(idx_map, idx_size);
632 return error("wrong index v2 file size in %s", path);
633 }
634 if (idx_size != min_size &&
635 /*
636 * make sure we can deal with large pack offsets.
637 * 31-bit signed offset won't be enough, neither
638 * 32-bit unsigned one will be.
639 */
640 (sizeof(off_t) <= 4)) {
641 munmap(idx_map, idx_size);
642 return error("pack too large for current definition of off_t in %s", path);
643 }
644 }
645
646 p->index_version = version;
647 p->index_data = idx_map;
648 p->index_size = idx_size;
649 p->num_objects = nr;
650 return 0;
651}
652
653int open_pack_index(struct packed_git *p)
654{
655 char *idx_name;
656 size_t len;
657 int ret;
658
659 if (p->index_data)
660 return 0;
661
662 if (!strip_suffix(p->pack_name, ".pack", &len))
663 die("BUG: pack_name does not end in .pack");
664 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
665 ret = check_packed_git_idx(idx_name, p);
666 free(idx_name);
667 return ret;
668}
669
670static void scan_windows(struct packed_git *p,
671 struct packed_git **lru_p,
672 struct pack_window **lru_w,
673 struct pack_window **lru_l)
674{
675 struct pack_window *w, *w_l;
676
677 for (w_l = NULL, w = p->windows; w; w = w->next) {
678 if (!w->inuse_cnt) {
679 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
680 *lru_p = p;
681 *lru_w = w;
682 *lru_l = w_l;
683 }
684 }
685 w_l = w;
686 }
687}
688
689static int unuse_one_window(struct packed_git *current)
690{
691 struct packed_git *p, *lru_p = NULL;
692 struct pack_window *lru_w = NULL, *lru_l = NULL;
693
694 if (current)
695 scan_windows(current, &lru_p, &lru_w, &lru_l);
696 for (p = packed_git; p; p = p->next)
697 scan_windows(p, &lru_p, &lru_w, &lru_l);
698 if (lru_p) {
699 munmap(lru_w->base, lru_w->len);
700 pack_mapped -= lru_w->len;
701 if (lru_l)
702 lru_l->next = lru_w->next;
703 else
704 lru_p->windows = lru_w->next;
705 free(lru_w);
706 pack_open_windows--;
707 return 1;
708 }
709 return 0;
710}
711
712void release_pack_memory(size_t need)
713{
714 size_t cur = pack_mapped;
715 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
716 ; /* nothing */
717}
718
719static void mmap_limit_check(size_t length)
720{
721 static size_t limit = 0;
722 if (!limit) {
723 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
724 if (!limit)
725 limit = SIZE_MAX;
726 }
727 if (length > limit)
728 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
729 (uintmax_t)length, (uintmax_t)limit);
730}
731
732void *xmmap_gently(void *start, size_t length,
733 int prot, int flags, int fd, off_t offset)
734{
735 void *ret;
736
737 mmap_limit_check(length);
738 ret = mmap(start, length, prot, flags, fd, offset);
739 if (ret == MAP_FAILED) {
740 if (!length)
741 return NULL;
742 release_pack_memory(length);
743 ret = mmap(start, length, prot, flags, fd, offset);
744 }
745 return ret;
746}
747
748void *xmmap(void *start, size_t length,
749 int prot, int flags, int fd, off_t offset)
750{
751 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
752 if (ret == MAP_FAILED)
753 die_errno("mmap failed");
754 return ret;
755}
756
757void close_pack_windows(struct packed_git *p)
758{
759 while (p->windows) {
760 struct pack_window *w = p->windows;
761
762 if (w->inuse_cnt)
763 die("pack '%s' still has open windows to it",
764 p->pack_name);
765 munmap(w->base, w->len);
766 pack_mapped -= w->len;
767 pack_open_windows--;
768 p->windows = w->next;
769 free(w);
770 }
771}
772
773static int close_pack_fd(struct packed_git *p)
774{
775 if (p->pack_fd < 0)
776 return 0;
777
778 close(p->pack_fd);
779 pack_open_fds--;
780 p->pack_fd = -1;
781
782 return 1;
783}
784
785static void close_pack(struct packed_git *p)
786{
787 close_pack_windows(p);
788 close_pack_fd(p);
789 close_pack_index(p);
790}
791
792void close_all_packs(void)
793{
794 struct packed_git *p;
795
796 for (p = packed_git; p; p = p->next)
797 if (p->do_not_close)
798 die("BUG! Want to close pack marked 'do-not-close'");
799 else
800 close_pack(p);
801}
802
803
804/*
805 * The LRU pack is the one with the oldest MRU window, preferring packs
806 * with no used windows, or the oldest mtime if it has no windows allocated.
807 */
808static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
809{
810 struct pack_window *w, *this_mru_w;
811 int has_windows_inuse = 0;
812
813 /*
814 * Reject this pack if it has windows and the previously selected
815 * one does not. If this pack does not have windows, reject
816 * it if the pack file is newer than the previously selected one.
817 */
818 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
819 return;
820
821 for (w = this_mru_w = p->windows; w; w = w->next) {
822 /*
823 * Reject this pack if any of its windows are in use,
824 * but the previously selected pack did not have any
825 * inuse windows. Otherwise, record that this pack
826 * has windows in use.
827 */
828 if (w->inuse_cnt) {
829 if (*accept_windows_inuse)
830 has_windows_inuse = 1;
831 else
832 return;
833 }
834
835 if (w->last_used > this_mru_w->last_used)
836 this_mru_w = w;
837
838 /*
839 * Reject this pack if it has windows that have been
840 * used more recently than the previously selected pack.
841 * If the previously selected pack had windows inuse and
842 * we have not encountered a window in this pack that is
843 * inuse, skip this check since we prefer a pack with no
844 * inuse windows to one that has inuse windows.
845 */
846 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
847 this_mru_w->last_used > (*mru_w)->last_used)
848 return;
849 }
850
851 /*
852 * Select this pack.
853 */
854 *mru_w = this_mru_w;
855 *lru_p = p;
856 *accept_windows_inuse = has_windows_inuse;
857}
858
859static int close_one_pack(void)
860{
861 struct packed_git *p, *lru_p = NULL;
862 struct pack_window *mru_w = NULL;
863 int accept_windows_inuse = 1;
864
865 for (p = packed_git; p; p = p->next) {
866 if (p->pack_fd == -1)
867 continue;
868 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
869 }
870
871 if (lru_p)
872 return close_pack_fd(lru_p);
873
874 return 0;
875}
876
877void unuse_pack(struct pack_window **w_cursor)
878{
879 struct pack_window *w = *w_cursor;
880 if (w) {
881 w->inuse_cnt--;
882 *w_cursor = NULL;
883 }
884}
885
886void close_pack_index(struct packed_git *p)
887{
888 if (p->index_data) {
889 munmap((void *)p->index_data, p->index_size);
890 p->index_data = NULL;
891 }
892}
893
894/*
895 * This is used by git-repack in case a newly created pack happens to
896 * contain the same set of objects as an existing one. In that case
897 * the resulting file might be different even if its name would be the
898 * same. It is best to close any reference to the old pack before it is
899 * replaced on disk. Of course no index pointers or windows for given pack
900 * must subsist at this point. If ever objects from this pack are requested
901 * again, the new version of the pack will be reinitialized through
902 * reprepare_packed_git().
903 */
904void free_pack_by_name(const char *pack_name)
905{
906 struct packed_git *p, **pp = &packed_git;
907
908 while (*pp) {
909 p = *pp;
910 if (strcmp(pack_name, p->pack_name) == 0) {
911 clear_delta_base_cache();
912 close_pack(p);
913 free(p->bad_object_sha1);
914 *pp = p->next;
915 if (last_found_pack == p)
916 last_found_pack = NULL;
917 free(p);
918 return;
919 }
920 pp = &p->next;
921 }
922}
923
924static unsigned int get_max_fd_limit(void)
925{
926#ifdef RLIMIT_NOFILE
927 {
928 struct rlimit lim;
929
930 if (!getrlimit(RLIMIT_NOFILE, &lim))
931 return lim.rlim_cur;
932 }
933#endif
934
935#ifdef _SC_OPEN_MAX
936 {
937 long open_max = sysconf(_SC_OPEN_MAX);
938 if (0 < open_max)
939 return open_max;
940 /*
941 * Otherwise, we got -1 for one of the two
942 * reasons:
943 *
944 * (1) sysconf() did not understand _SC_OPEN_MAX
945 * and signaled an error with -1; or
946 * (2) sysconf() said there is no limit.
947 *
948 * We _could_ clear errno before calling sysconf() to
949 * tell these two cases apart and return a huge number
950 * in the latter case to let the caller cap it to a
951 * value that is not so selfish, but letting the
952 * fallback OPEN_MAX codepath take care of these cases
953 * is a lot simpler.
954 */
955 }
956#endif
957
958#ifdef OPEN_MAX
959 return OPEN_MAX;
960#else
961 return 1; /* see the caller ;-) */
962#endif
963}
964
965/*
966 * Do not call this directly as this leaks p->pack_fd on error return;
967 * call open_packed_git() instead.
968 */
969static int open_packed_git_1(struct packed_git *p)
970{
971 struct stat st;
972 struct pack_header hdr;
973 unsigned char sha1[20];
974 unsigned char *idx_sha1;
975 long fd_flag;
976
977 if (!p->index_data && open_pack_index(p))
978 return error("packfile %s index unavailable", p->pack_name);
979
980 if (!pack_max_fds) {
981 unsigned int max_fds = get_max_fd_limit();
982
983 /* Save 3 for stdin/stdout/stderr, 22 for work */
984 if (25 < max_fds)
985 pack_max_fds = max_fds - 25;
986 else
987 pack_max_fds = 1;
988 }
989
990 while (pack_max_fds <= pack_open_fds && close_one_pack())
991 ; /* nothing */
992
993 p->pack_fd = git_open_noatime(p->pack_name);
994 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
995 return -1;
996 pack_open_fds++;
997
998 /* If we created the struct before we had the pack we lack size. */
999 if (!p->pack_size) {
1000 if (!S_ISREG(st.st_mode))
1001 return error("packfile %s not a regular file", p->pack_name);
1002 p->pack_size = st.st_size;
1003 } else if (p->pack_size != st.st_size)
1004 return error("packfile %s size changed", p->pack_name);
1005
1006 /* We leave these file descriptors open with sliding mmap;
1007 * there is no point keeping them open across exec(), though.
1008 */
1009 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1010 if (fd_flag < 0)
1011 return error("cannot determine file descriptor flags");
1012 fd_flag |= FD_CLOEXEC;
1013 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1014 return error("cannot set FD_CLOEXEC");
1015
1016 /* Verify we recognize this pack file format. */
1017 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1018 return error("file %s is far too short to be a packfile", p->pack_name);
1019 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1020 return error("file %s is not a GIT packfile", p->pack_name);
1021 if (!pack_version_ok(hdr.hdr_version))
1022 return error("packfile %s is version %"PRIu32" and not"
1023 " supported (try upgrading GIT to a newer version)",
1024 p->pack_name, ntohl(hdr.hdr_version));
1025
1026 /* Verify the pack matches its index. */
1027 if (p->num_objects != ntohl(hdr.hdr_entries))
1028 return error("packfile %s claims to have %"PRIu32" objects"
1029 " while index indicates %"PRIu32" objects",
1030 p->pack_name, ntohl(hdr.hdr_entries),
1031 p->num_objects);
1032 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1033 return error("end of packfile %s is unavailable", p->pack_name);
1034 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1035 return error("packfile %s signature is unavailable", p->pack_name);
1036 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1037 if (hashcmp(sha1, idx_sha1))
1038 return error("packfile %s does not match index", p->pack_name);
1039 return 0;
1040}
1041
1042static int open_packed_git(struct packed_git *p)
1043{
1044 if (!open_packed_git_1(p))
1045 return 0;
1046 close_pack_fd(p);
1047 return -1;
1048}
1049
1050static int in_window(struct pack_window *win, off_t offset)
1051{
1052 /* We must promise at least 20 bytes (one hash) after the
1053 * offset is available from this window, otherwise the offset
1054 * is not actually in this window and a different window (which
1055 * has that one hash excess) must be used. This is to support
1056 * the object header and delta base parsing routines below.
1057 */
1058 off_t win_off = win->offset;
1059 return win_off <= offset
1060 && (offset + 20) <= (win_off + win->len);
1061}
1062
1063unsigned char *use_pack(struct packed_git *p,
1064 struct pack_window **w_cursor,
1065 off_t offset,
1066 unsigned long *left)
1067{
1068 struct pack_window *win = *w_cursor;
1069
1070 /* Since packfiles end in a hash of their content and it's
1071 * pointless to ask for an offset into the middle of that
1072 * hash, and the in_window function above wouldn't match
1073 * don't allow an offset too close to the end of the file.
1074 */
1075 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1076 die("packfile %s cannot be accessed", p->pack_name);
1077 if (offset > (p->pack_size - 20))
1078 die("offset beyond end of packfile (truncated pack?)");
1079
1080 if (!win || !in_window(win, offset)) {
1081 if (win)
1082 win->inuse_cnt--;
1083 for (win = p->windows; win; win = win->next) {
1084 if (in_window(win, offset))
1085 break;
1086 }
1087 if (!win) {
1088 size_t window_align = packed_git_window_size / 2;
1089 off_t len;
1090
1091 if (p->pack_fd == -1 && open_packed_git(p))
1092 die("packfile %s cannot be accessed", p->pack_name);
1093
1094 win = xcalloc(1, sizeof(*win));
1095 win->offset = (offset / window_align) * window_align;
1096 len = p->pack_size - win->offset;
1097 if (len > packed_git_window_size)
1098 len = packed_git_window_size;
1099 win->len = (size_t)len;
1100 pack_mapped += win->len;
1101 while (packed_git_limit < pack_mapped
1102 && unuse_one_window(p))
1103 ; /* nothing */
1104 win->base = xmmap(NULL, win->len,
1105 PROT_READ, MAP_PRIVATE,
1106 p->pack_fd, win->offset);
1107 if (win->base == MAP_FAILED)
1108 die("packfile %s cannot be mapped: %s",
1109 p->pack_name,
1110 strerror(errno));
1111 if (!win->offset && win->len == p->pack_size
1112 && !p->do_not_close)
1113 close_pack_fd(p);
1114 pack_mmap_calls++;
1115 pack_open_windows++;
1116 if (pack_mapped > peak_pack_mapped)
1117 peak_pack_mapped = pack_mapped;
1118 if (pack_open_windows > peak_pack_open_windows)
1119 peak_pack_open_windows = pack_open_windows;
1120 win->next = p->windows;
1121 p->windows = win;
1122 }
1123 }
1124 if (win != *w_cursor) {
1125 win->last_used = pack_used_ctr++;
1126 win->inuse_cnt++;
1127 *w_cursor = win;
1128 }
1129 offset -= win->offset;
1130 if (left)
1131 *left = win->len - xsize_t(offset);
1132 return win->base + offset;
1133}
1134
1135static struct packed_git *alloc_packed_git(int extra)
1136{
1137 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1138 memset(p, 0, sizeof(*p));
1139 p->pack_fd = -1;
1140 return p;
1141}
1142
1143static void try_to_free_pack_memory(size_t size)
1144{
1145 release_pack_memory(size);
1146}
1147
1148struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1149{
1150 static int have_set_try_to_free_routine;
1151 struct stat st;
1152 size_t alloc;
1153 struct packed_git *p;
1154
1155 if (!have_set_try_to_free_routine) {
1156 have_set_try_to_free_routine = 1;
1157 set_try_to_free_routine(try_to_free_pack_memory);
1158 }
1159
1160 /*
1161 * Make sure a corresponding .pack file exists and that
1162 * the index looks sane.
1163 */
1164 if (!strip_suffix_mem(path, &path_len, ".idx"))
1165 return NULL;
1166
1167 /*
1168 * ".pack" is long enough to hold any suffix we're adding (and
1169 * the use xsnprintf double-checks that)
1170 */
1171 alloc = path_len + strlen(".pack") + 1;
1172 p = alloc_packed_git(alloc);
1173 memcpy(p->pack_name, path, path_len);
1174
1175 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1176 if (!access(p->pack_name, F_OK))
1177 p->pack_keep = 1;
1178
1179 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1180 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1181 free(p);
1182 return NULL;
1183 }
1184
1185 /* ok, it looks sane as far as we can check without
1186 * actually mapping the pack file.
1187 */
1188 p->pack_size = st.st_size;
1189 p->pack_local = local;
1190 p->mtime = st.st_mtime;
1191 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1192 hashclr(p->sha1);
1193 return p;
1194}
1195
1196struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1197{
1198 const char *path = sha1_pack_name(sha1);
1199 int alloc = strlen(path) + 1;
1200 struct packed_git *p = alloc_packed_git(alloc);
1201
1202 memcpy(p->pack_name, path, alloc); /* includes NUL */
1203 hashcpy(p->sha1, sha1);
1204 if (check_packed_git_idx(idx_path, p)) {
1205 free(p);
1206 return NULL;
1207 }
1208
1209 return p;
1210}
1211
1212void install_packed_git(struct packed_git *pack)
1213{
1214 if (pack->pack_fd != -1)
1215 pack_open_fds++;
1216
1217 pack->next = packed_git;
1218 packed_git = pack;
1219}
1220
1221void (*report_garbage)(unsigned seen_bits, const char *path);
1222
1223static void report_helper(const struct string_list *list,
1224 int seen_bits, int first, int last)
1225{
1226 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1227 return;
1228
1229 for (; first < last; first++)
1230 report_garbage(seen_bits, list->items[first].string);
1231}
1232
1233static void report_pack_garbage(struct string_list *list)
1234{
1235 int i, baselen = -1, first = 0, seen_bits = 0;
1236
1237 if (!report_garbage)
1238 return;
1239
1240 string_list_sort(list);
1241
1242 for (i = 0; i < list->nr; i++) {
1243 const char *path = list->items[i].string;
1244 if (baselen != -1 &&
1245 strncmp(path, list->items[first].string, baselen)) {
1246 report_helper(list, seen_bits, first, i);
1247 baselen = -1;
1248 seen_bits = 0;
1249 }
1250 if (baselen == -1) {
1251 const char *dot = strrchr(path, '.');
1252 if (!dot) {
1253 report_garbage(PACKDIR_FILE_GARBAGE, path);
1254 continue;
1255 }
1256 baselen = dot - path + 1;
1257 first = i;
1258 }
1259 if (!strcmp(path + baselen, "pack"))
1260 seen_bits |= 1;
1261 else if (!strcmp(path + baselen, "idx"))
1262 seen_bits |= 2;
1263 }
1264 report_helper(list, seen_bits, first, list->nr);
1265}
1266
1267static void prepare_packed_git_one(char *objdir, int local)
1268{
1269 struct strbuf path = STRBUF_INIT;
1270 size_t dirnamelen;
1271 DIR *dir;
1272 struct dirent *de;
1273 struct string_list garbage = STRING_LIST_INIT_DUP;
1274
1275 strbuf_addstr(&path, objdir);
1276 strbuf_addstr(&path, "/pack");
1277 dir = opendir(path.buf);
1278 if (!dir) {
1279 if (errno != ENOENT)
1280 error("unable to open object pack directory: %s: %s",
1281 path.buf, strerror(errno));
1282 strbuf_release(&path);
1283 return;
1284 }
1285 strbuf_addch(&path, '/');
1286 dirnamelen = path.len;
1287 while ((de = readdir(dir)) != NULL) {
1288 struct packed_git *p;
1289 size_t base_len;
1290
1291 if (is_dot_or_dotdot(de->d_name))
1292 continue;
1293
1294 strbuf_setlen(&path, dirnamelen);
1295 strbuf_addstr(&path, de->d_name);
1296
1297 base_len = path.len;
1298 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1299 /* Don't reopen a pack we already have. */
1300 for (p = packed_git; p; p = p->next) {
1301 size_t len;
1302 if (strip_suffix(p->pack_name, ".pack", &len) &&
1303 len == base_len &&
1304 !memcmp(p->pack_name, path.buf, len))
1305 break;
1306 }
1307 if (p == NULL &&
1308 /*
1309 * See if it really is a valid .idx file with
1310 * corresponding .pack file that we can map.
1311 */
1312 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1313 install_packed_git(p);
1314 }
1315
1316 if (!report_garbage)
1317 continue;
1318
1319 if (ends_with(de->d_name, ".idx") ||
1320 ends_with(de->d_name, ".pack") ||
1321 ends_with(de->d_name, ".bitmap") ||
1322 ends_with(de->d_name, ".keep"))
1323 string_list_append(&garbage, path.buf);
1324 else
1325 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1326 }
1327 closedir(dir);
1328 report_pack_garbage(&garbage);
1329 string_list_clear(&garbage, 0);
1330 strbuf_release(&path);
1331}
1332
1333static int sort_pack(const void *a_, const void *b_)
1334{
1335 struct packed_git *a = *((struct packed_git **)a_);
1336 struct packed_git *b = *((struct packed_git **)b_);
1337 int st;
1338
1339 /*
1340 * Local packs tend to contain objects specific to our
1341 * variant of the project than remote ones. In addition,
1342 * remote ones could be on a network mounted filesystem.
1343 * Favor local ones for these reasons.
1344 */
1345 st = a->pack_local - b->pack_local;
1346 if (st)
1347 return -st;
1348
1349 /*
1350 * Younger packs tend to contain more recent objects,
1351 * and more recent objects tend to get accessed more
1352 * often.
1353 */
1354 if (a->mtime < b->mtime)
1355 return 1;
1356 else if (a->mtime == b->mtime)
1357 return 0;
1358 return -1;
1359}
1360
1361static void rearrange_packed_git(void)
1362{
1363 struct packed_git **ary, *p;
1364 int i, n;
1365
1366 for (n = 0, p = packed_git; p; p = p->next)
1367 n++;
1368 if (n < 2)
1369 return;
1370
1371 /* prepare an array of packed_git for easier sorting */
1372 ary = xcalloc(n, sizeof(struct packed_git *));
1373 for (n = 0, p = packed_git; p; p = p->next)
1374 ary[n++] = p;
1375
1376 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1377
1378 /* link them back again */
1379 for (i = 0; i < n - 1; i++)
1380 ary[i]->next = ary[i + 1];
1381 ary[n - 1]->next = NULL;
1382 packed_git = ary[0];
1383
1384 free(ary);
1385}
1386
1387static int prepare_packed_git_run_once = 0;
1388void prepare_packed_git(void)
1389{
1390 struct alternate_object_database *alt;
1391
1392 if (prepare_packed_git_run_once)
1393 return;
1394 prepare_packed_git_one(get_object_directory(), 1);
1395 prepare_alt_odb();
1396 for (alt = alt_odb_list; alt; alt = alt->next) {
1397 alt->name[-1] = 0;
1398 prepare_packed_git_one(alt->base, 0);
1399 alt->name[-1] = '/';
1400 }
1401 rearrange_packed_git();
1402 prepare_packed_git_run_once = 1;
1403}
1404
1405void reprepare_packed_git(void)
1406{
1407 prepare_packed_git_run_once = 0;
1408 prepare_packed_git();
1409}
1410
1411static void mark_bad_packed_object(struct packed_git *p,
1412 const unsigned char *sha1)
1413{
1414 unsigned i;
1415 for (i = 0; i < p->num_bad_objects; i++)
1416 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1417 return;
1418 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1419 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1420 p->num_bad_objects++;
1421}
1422
1423static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1424{
1425 struct packed_git *p;
1426 unsigned i;
1427
1428 for (p = packed_git; p; p = p->next)
1429 for (i = 0; i < p->num_bad_objects; i++)
1430 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1431 return p;
1432 return NULL;
1433}
1434
1435/*
1436 * With an in-core object data in "map", rehash it to make sure the
1437 * object name actually matches "sha1" to detect object corruption.
1438 * With "map" == NULL, try reading the object named with "sha1" using
1439 * the streaming interface and rehash it to do the same.
1440 */
1441int check_sha1_signature(const unsigned char *sha1, void *map,
1442 unsigned long size, const char *type)
1443{
1444 unsigned char real_sha1[20];
1445 enum object_type obj_type;
1446 struct git_istream *st;
1447 git_SHA_CTX c;
1448 char hdr[32];
1449 int hdrlen;
1450
1451 if (map) {
1452 hash_sha1_file(map, size, type, real_sha1);
1453 return hashcmp(sha1, real_sha1) ? -1 : 0;
1454 }
1455
1456 st = open_istream(sha1, &obj_type, &size, NULL);
1457 if (!st)
1458 return -1;
1459
1460 /* Generate the header */
1461 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1462
1463 /* Sha1.. */
1464 git_SHA1_Init(&c);
1465 git_SHA1_Update(&c, hdr, hdrlen);
1466 for (;;) {
1467 char buf[1024 * 16];
1468 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1469
1470 if (readlen < 0) {
1471 close_istream(st);
1472 return -1;
1473 }
1474 if (!readlen)
1475 break;
1476 git_SHA1_Update(&c, buf, readlen);
1477 }
1478 git_SHA1_Final(real_sha1, &c);
1479 close_istream(st);
1480 return hashcmp(sha1, real_sha1) ? -1 : 0;
1481}
1482
1483int git_open_noatime(const char *name)
1484{
1485 static int sha1_file_open_flag = O_NOATIME;
1486
1487 for (;;) {
1488 int fd;
1489
1490 errno = 0;
1491 fd = open(name, O_RDONLY | sha1_file_open_flag);
1492 if (fd >= 0)
1493 return fd;
1494
1495 /* Might the failure be due to O_NOATIME? */
1496 if (errno != ENOENT && sha1_file_open_flag) {
1497 sha1_file_open_flag = 0;
1498 continue;
1499 }
1500
1501 return -1;
1502 }
1503}
1504
1505static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1506{
1507 struct alternate_object_database *alt;
1508
1509 if (!lstat(sha1_file_name(sha1), st))
1510 return 0;
1511
1512 prepare_alt_odb();
1513 errno = ENOENT;
1514 for (alt = alt_odb_list; alt; alt = alt->next) {
1515 fill_sha1_path(alt->name, sha1);
1516 if (!lstat(alt->base, st))
1517 return 0;
1518 }
1519
1520 return -1;
1521}
1522
1523static int open_sha1_file(const unsigned char *sha1)
1524{
1525 int fd;
1526 struct alternate_object_database *alt;
1527 int most_interesting_errno;
1528
1529 fd = git_open_noatime(sha1_file_name(sha1));
1530 if (fd >= 0)
1531 return fd;
1532 most_interesting_errno = errno;
1533
1534 prepare_alt_odb();
1535 for (alt = alt_odb_list; alt; alt = alt->next) {
1536 fill_sha1_path(alt->name, sha1);
1537 fd = git_open_noatime(alt->base);
1538 if (fd >= 0)
1539 return fd;
1540 if (most_interesting_errno == ENOENT)
1541 most_interesting_errno = errno;
1542 }
1543 errno = most_interesting_errno;
1544 return -1;
1545}
1546
1547void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1548{
1549 void *map;
1550 int fd;
1551
1552 fd = open_sha1_file(sha1);
1553 map = NULL;
1554 if (fd >= 0) {
1555 struct stat st;
1556
1557 if (!fstat(fd, &st)) {
1558 *size = xsize_t(st.st_size);
1559 if (!*size) {
1560 /* mmap() is forbidden on empty files */
1561 error("object file %s is empty", sha1_file_name(sha1));
1562 return NULL;
1563 }
1564 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1565 }
1566 close(fd);
1567 }
1568 return map;
1569}
1570
1571unsigned long unpack_object_header_buffer(const unsigned char *buf,
1572 unsigned long len, enum object_type *type, unsigned long *sizep)
1573{
1574 unsigned shift;
1575 unsigned long size, c;
1576 unsigned long used = 0;
1577
1578 c = buf[used++];
1579 *type = (c >> 4) & 7;
1580 size = c & 15;
1581 shift = 4;
1582 while (c & 0x80) {
1583 if (len <= used || bitsizeof(long) <= shift) {
1584 error("bad object header");
1585 size = used = 0;
1586 break;
1587 }
1588 c = buf[used++];
1589 size += (c & 0x7f) << shift;
1590 shift += 7;
1591 }
1592 *sizep = size;
1593 return used;
1594}
1595
1596int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1597{
1598 /* Get the data stream */
1599 memset(stream, 0, sizeof(*stream));
1600 stream->next_in = map;
1601 stream->avail_in = mapsize;
1602 stream->next_out = buffer;
1603 stream->avail_out = bufsiz;
1604
1605 git_inflate_init(stream);
1606 return git_inflate(stream, 0);
1607}
1608
1609static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1610 unsigned long mapsize, void *buffer,
1611 unsigned long bufsiz, struct strbuf *header)
1612{
1613 int status;
1614
1615 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1616
1617 /*
1618 * Check if entire header is unpacked in the first iteration.
1619 */
1620 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1621 return 0;
1622
1623 /*
1624 * buffer[0..bufsiz] was not large enough. Copy the partial
1625 * result out to header, and then append the result of further
1626 * reading the stream.
1627 */
1628 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1629 stream->next_out = buffer;
1630 stream->avail_out = bufsiz;
1631
1632 do {
1633 status = git_inflate(stream, 0);
1634 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1635 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1636 return 0;
1637 stream->next_out = buffer;
1638 stream->avail_out = bufsiz;
1639 } while (status != Z_STREAM_END);
1640 return -1;
1641}
1642
1643static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1644{
1645 int bytes = strlen(buffer) + 1;
1646 unsigned char *buf = xmallocz(size);
1647 unsigned long n;
1648 int status = Z_OK;
1649
1650 n = stream->total_out - bytes;
1651 if (n > size)
1652 n = size;
1653 memcpy(buf, (char *) buffer + bytes, n);
1654 bytes = n;
1655 if (bytes <= size) {
1656 /*
1657 * The above condition must be (bytes <= size), not
1658 * (bytes < size). In other words, even though we
1659 * expect no more output and set avail_out to zero,
1660 * the input zlib stream may have bytes that express
1661 * "this concludes the stream", and we *do* want to
1662 * eat that input.
1663 *
1664 * Otherwise we would not be able to test that we
1665 * consumed all the input to reach the expected size;
1666 * we also want to check that zlib tells us that all
1667 * went well with status == Z_STREAM_END at the end.
1668 */
1669 stream->next_out = buf + bytes;
1670 stream->avail_out = size - bytes;
1671 while (status == Z_OK)
1672 status = git_inflate(stream, Z_FINISH);
1673 }
1674 if (status == Z_STREAM_END && !stream->avail_in) {
1675 git_inflate_end(stream);
1676 return buf;
1677 }
1678
1679 if (status < 0)
1680 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1681 else if (stream->avail_in)
1682 error("garbage at end of loose object '%s'",
1683 sha1_to_hex(sha1));
1684 free(buf);
1685 return NULL;
1686}
1687
1688/*
1689 * We used to just use "sscanf()", but that's actually way
1690 * too permissive for what we want to check. So do an anal
1691 * object header parse by hand.
1692 */
1693static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1694 unsigned int flags)
1695{
1696 const char *type_buf = hdr;
1697 unsigned long size;
1698 int type, type_len = 0;
1699
1700 /*
1701 * The type can be of any size but is followed by
1702 * a space.
1703 */
1704 for (;;) {
1705 char c = *hdr++;
1706 if (c == ' ')
1707 break;
1708 type_len++;
1709 }
1710
1711 type = type_from_string_gently(type_buf, type_len, 1);
1712 if (oi->typename)
1713 strbuf_add(oi->typename, type_buf, type_len);
1714 /*
1715 * Set type to 0 if its an unknown object and
1716 * we're obtaining the type using '--allow-unkown-type'
1717 * option.
1718 */
1719 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1720 type = 0;
1721 else if (type < 0)
1722 die("invalid object type");
1723 if (oi->typep)
1724 *oi->typep = type;
1725
1726 /*
1727 * The length must follow immediately, and be in canonical
1728 * decimal format (ie "010" is not valid).
1729 */
1730 size = *hdr++ - '0';
1731 if (size > 9)
1732 return -1;
1733 if (size) {
1734 for (;;) {
1735 unsigned long c = *hdr - '0';
1736 if (c > 9)
1737 break;
1738 hdr++;
1739 size = size * 10 + c;
1740 }
1741 }
1742
1743 if (oi->sizep)
1744 *oi->sizep = size;
1745
1746 /*
1747 * The length must be followed by a zero byte
1748 */
1749 return *hdr ? -1 : type;
1750}
1751
1752int parse_sha1_header(const char *hdr, unsigned long *sizep)
1753{
1754 struct object_info oi;
1755
1756 oi.sizep = sizep;
1757 oi.typename = NULL;
1758 oi.typep = NULL;
1759 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1760}
1761
1762static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1763{
1764 int ret;
1765 git_zstream stream;
1766 char hdr[8192];
1767
1768 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1769 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1770 return NULL;
1771
1772 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1773}
1774
1775unsigned long get_size_from_delta(struct packed_git *p,
1776 struct pack_window **w_curs,
1777 off_t curpos)
1778{
1779 const unsigned char *data;
1780 unsigned char delta_head[20], *in;
1781 git_zstream stream;
1782 int st;
1783
1784 memset(&stream, 0, sizeof(stream));
1785 stream.next_out = delta_head;
1786 stream.avail_out = sizeof(delta_head);
1787
1788 git_inflate_init(&stream);
1789 do {
1790 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1791 stream.next_in = in;
1792 st = git_inflate(&stream, Z_FINISH);
1793 curpos += stream.next_in - in;
1794 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1795 stream.total_out < sizeof(delta_head));
1796 git_inflate_end(&stream);
1797 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1798 error("delta data unpack-initial failed");
1799 return 0;
1800 }
1801
1802 /* Examine the initial part of the delta to figure out
1803 * the result size.
1804 */
1805 data = delta_head;
1806
1807 /* ignore base size */
1808 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1809
1810 /* Read the result size */
1811 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1812}
1813
1814static off_t get_delta_base(struct packed_git *p,
1815 struct pack_window **w_curs,
1816 off_t *curpos,
1817 enum object_type type,
1818 off_t delta_obj_offset)
1819{
1820 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1821 off_t base_offset;
1822
1823 /* use_pack() assured us we have [base_info, base_info + 20)
1824 * as a range that we can look at without walking off the
1825 * end of the mapped window. Its actually the hash size
1826 * that is assured. An OFS_DELTA longer than the hash size
1827 * is stupid, as then a REF_DELTA would be smaller to store.
1828 */
1829 if (type == OBJ_OFS_DELTA) {
1830 unsigned used = 0;
1831 unsigned char c = base_info[used++];
1832 base_offset = c & 127;
1833 while (c & 128) {
1834 base_offset += 1;
1835 if (!base_offset || MSB(base_offset, 7))
1836 return 0; /* overflow */
1837 c = base_info[used++];
1838 base_offset = (base_offset << 7) + (c & 127);
1839 }
1840 base_offset = delta_obj_offset - base_offset;
1841 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1842 return 0; /* out of bound */
1843 *curpos += used;
1844 } else if (type == OBJ_REF_DELTA) {
1845 /* The base entry _must_ be in the same pack */
1846 base_offset = find_pack_entry_one(base_info, p);
1847 *curpos += 20;
1848 } else
1849 die("I am totally screwed");
1850 return base_offset;
1851}
1852
1853/*
1854 * Like get_delta_base above, but we return the sha1 instead of the pack
1855 * offset. This means it is cheaper for REF deltas (we do not have to do
1856 * the final object lookup), but more expensive for OFS deltas (we
1857 * have to load the revidx to convert the offset back into a sha1).
1858 */
1859static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1860 struct pack_window **w_curs,
1861 off_t curpos,
1862 enum object_type type,
1863 off_t delta_obj_offset)
1864{
1865 if (type == OBJ_REF_DELTA) {
1866 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1867 return base;
1868 } else if (type == OBJ_OFS_DELTA) {
1869 struct revindex_entry *revidx;
1870 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1871 type, delta_obj_offset);
1872
1873 if (!base_offset)
1874 return NULL;
1875
1876 revidx = find_pack_revindex(p, base_offset);
1877 if (!revidx)
1878 return NULL;
1879
1880 return nth_packed_object_sha1(p, revidx->nr);
1881 } else
1882 return NULL;
1883}
1884
1885int unpack_object_header(struct packed_git *p,
1886 struct pack_window **w_curs,
1887 off_t *curpos,
1888 unsigned long *sizep)
1889{
1890 unsigned char *base;
1891 unsigned long left;
1892 unsigned long used;
1893 enum object_type type;
1894
1895 /* use_pack() assures us we have [base, base + 20) available
1896 * as a range that we can look at. (Its actually the hash
1897 * size that is assured.) With our object header encoding
1898 * the maximum deflated object size is 2^137, which is just
1899 * insane, so we know won't exceed what we have been given.
1900 */
1901 base = use_pack(p, w_curs, *curpos, &left);
1902 used = unpack_object_header_buffer(base, left, &type, sizep);
1903 if (!used) {
1904 type = OBJ_BAD;
1905 } else
1906 *curpos += used;
1907
1908 return type;
1909}
1910
1911static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1912{
1913 int type;
1914 struct revindex_entry *revidx;
1915 const unsigned char *sha1;
1916 revidx = find_pack_revindex(p, obj_offset);
1917 if (!revidx)
1918 return OBJ_BAD;
1919 sha1 = nth_packed_object_sha1(p, revidx->nr);
1920 mark_bad_packed_object(p, sha1);
1921 type = sha1_object_info(sha1, NULL);
1922 if (type <= OBJ_NONE)
1923 return OBJ_BAD;
1924 return type;
1925}
1926
1927#define POI_STACK_PREALLOC 64
1928
1929static enum object_type packed_to_object_type(struct packed_git *p,
1930 off_t obj_offset,
1931 enum object_type type,
1932 struct pack_window **w_curs,
1933 off_t curpos)
1934{
1935 off_t small_poi_stack[POI_STACK_PREALLOC];
1936 off_t *poi_stack = small_poi_stack;
1937 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1938
1939 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1940 off_t base_offset;
1941 unsigned long size;
1942 /* Push the object we're going to leave behind */
1943 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1944 poi_stack_alloc = alloc_nr(poi_stack_nr);
1945 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1946 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1947 } else {
1948 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1949 }
1950 poi_stack[poi_stack_nr++] = obj_offset;
1951 /* If parsing the base offset fails, just unwind */
1952 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1953 if (!base_offset)
1954 goto unwind;
1955 curpos = obj_offset = base_offset;
1956 type = unpack_object_header(p, w_curs, &curpos, &size);
1957 if (type <= OBJ_NONE) {
1958 /* If getting the base itself fails, we first
1959 * retry the base, otherwise unwind */
1960 type = retry_bad_packed_offset(p, base_offset);
1961 if (type > OBJ_NONE)
1962 goto out;
1963 goto unwind;
1964 }
1965 }
1966
1967 switch (type) {
1968 case OBJ_BAD:
1969 case OBJ_COMMIT:
1970 case OBJ_TREE:
1971 case OBJ_BLOB:
1972 case OBJ_TAG:
1973 break;
1974 default:
1975 error("unknown object type %i at offset %"PRIuMAX" in %s",
1976 type, (uintmax_t)obj_offset, p->pack_name);
1977 type = OBJ_BAD;
1978 }
1979
1980out:
1981 if (poi_stack != small_poi_stack)
1982 free(poi_stack);
1983 return type;
1984
1985unwind:
1986 while (poi_stack_nr) {
1987 obj_offset = poi_stack[--poi_stack_nr];
1988 type = retry_bad_packed_offset(p, obj_offset);
1989 if (type > OBJ_NONE)
1990 goto out;
1991 }
1992 type = OBJ_BAD;
1993 goto out;
1994}
1995
1996static int packed_object_info(struct packed_git *p, off_t obj_offset,
1997 struct object_info *oi)
1998{
1999 struct pack_window *w_curs = NULL;
2000 unsigned long size;
2001 off_t curpos = obj_offset;
2002 enum object_type type;
2003
2004 /*
2005 * We always get the representation type, but only convert it to
2006 * a "real" type later if the caller is interested.
2007 */
2008 type = unpack_object_header(p, &w_curs, &curpos, &size);
2009
2010 if (oi->sizep) {
2011 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2012 off_t tmp_pos = curpos;
2013 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2014 type, obj_offset);
2015 if (!base_offset) {
2016 type = OBJ_BAD;
2017 goto out;
2018 }
2019 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2020 if (*oi->sizep == 0) {
2021 type = OBJ_BAD;
2022 goto out;
2023 }
2024 } else {
2025 *oi->sizep = size;
2026 }
2027 }
2028
2029 if (oi->disk_sizep) {
2030 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2031 *oi->disk_sizep = revidx[1].offset - obj_offset;
2032 }
2033
2034 if (oi->typep) {
2035 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2036 if (*oi->typep < 0) {
2037 type = OBJ_BAD;
2038 goto out;
2039 }
2040 }
2041
2042 if (oi->delta_base_sha1) {
2043 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2044 const unsigned char *base;
2045
2046 base = get_delta_base_sha1(p, &w_curs, curpos,
2047 type, obj_offset);
2048 if (!base) {
2049 type = OBJ_BAD;
2050 goto out;
2051 }
2052
2053 hashcpy(oi->delta_base_sha1, base);
2054 } else
2055 hashclr(oi->delta_base_sha1);
2056 }
2057
2058out:
2059 unuse_pack(&w_curs);
2060 return type;
2061}
2062
2063static void *unpack_compressed_entry(struct packed_git *p,
2064 struct pack_window **w_curs,
2065 off_t curpos,
2066 unsigned long size)
2067{
2068 int st;
2069 git_zstream stream;
2070 unsigned char *buffer, *in;
2071
2072 buffer = xmallocz_gently(size);
2073 if (!buffer)
2074 return NULL;
2075 memset(&stream, 0, sizeof(stream));
2076 stream.next_out = buffer;
2077 stream.avail_out = size + 1;
2078
2079 git_inflate_init(&stream);
2080 do {
2081 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2082 stream.next_in = in;
2083 st = git_inflate(&stream, Z_FINISH);
2084 if (!stream.avail_out)
2085 break; /* the payload is larger than it should be */
2086 curpos += stream.next_in - in;
2087 } while (st == Z_OK || st == Z_BUF_ERROR);
2088 git_inflate_end(&stream);
2089 if ((st != Z_STREAM_END) || stream.total_out != size) {
2090 free(buffer);
2091 return NULL;
2092 }
2093
2094 return buffer;
2095}
2096
2097#define MAX_DELTA_CACHE (256)
2098
2099static size_t delta_base_cached;
2100
2101static struct delta_base_cache_lru_list {
2102 struct delta_base_cache_lru_list *prev;
2103 struct delta_base_cache_lru_list *next;
2104} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2105
2106static struct delta_base_cache_entry {
2107 struct delta_base_cache_lru_list lru;
2108 void *data;
2109 struct packed_git *p;
2110 off_t base_offset;
2111 unsigned long size;
2112 enum object_type type;
2113} delta_base_cache[MAX_DELTA_CACHE];
2114
2115static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2116{
2117 unsigned long hash;
2118
2119 hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
2120 hash += (hash >> 8) + (hash >> 16);
2121 return hash % MAX_DELTA_CACHE;
2122}
2123
2124static struct delta_base_cache_entry *
2125get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2126{
2127 unsigned long hash = pack_entry_hash(p, base_offset);
2128 return delta_base_cache + hash;
2129}
2130
2131static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2132 struct packed_git *p, off_t base_offset)
2133{
2134 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2135}
2136
2137static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2138{
2139 struct delta_base_cache_entry *ent;
2140 ent = get_delta_base_cache_entry(p, base_offset);
2141 return eq_delta_base_cache_entry(ent, p, base_offset);
2142}
2143
2144static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2145{
2146 ent->data = NULL;
2147 ent->lru.next->prev = ent->lru.prev;
2148 ent->lru.prev->next = ent->lru.next;
2149 delta_base_cached -= ent->size;
2150}
2151
2152static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2153 unsigned long *base_size, enum object_type *type, int keep_cache)
2154{
2155 struct delta_base_cache_entry *ent;
2156 void *ret;
2157
2158 ent = get_delta_base_cache_entry(p, base_offset);
2159
2160 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2161 return unpack_entry(p, base_offset, type, base_size);
2162
2163 ret = ent->data;
2164
2165 if (!keep_cache)
2166 clear_delta_base_cache_entry(ent);
2167 else
2168 ret = xmemdupz(ent->data, ent->size);
2169 *type = ent->type;
2170 *base_size = ent->size;
2171 return ret;
2172}
2173
2174static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2175{
2176 if (ent->data) {
2177 free(ent->data);
2178 ent->data = NULL;
2179 ent->lru.next->prev = ent->lru.prev;
2180 ent->lru.prev->next = ent->lru.next;
2181 delta_base_cached -= ent->size;
2182 }
2183}
2184
2185void clear_delta_base_cache(void)
2186{
2187 unsigned long p;
2188 for (p = 0; p < MAX_DELTA_CACHE; p++)
2189 release_delta_base_cache(&delta_base_cache[p]);
2190}
2191
2192static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2193 void *base, unsigned long base_size, enum object_type type)
2194{
2195 unsigned long hash = pack_entry_hash(p, base_offset);
2196 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2197 struct delta_base_cache_lru_list *lru;
2198
2199 release_delta_base_cache(ent);
2200 delta_base_cached += base_size;
2201
2202 for (lru = delta_base_cache_lru.next;
2203 delta_base_cached > delta_base_cache_limit
2204 && lru != &delta_base_cache_lru;
2205 lru = lru->next) {
2206 struct delta_base_cache_entry *f = (void *)lru;
2207 if (f->type == OBJ_BLOB)
2208 release_delta_base_cache(f);
2209 }
2210 for (lru = delta_base_cache_lru.next;
2211 delta_base_cached > delta_base_cache_limit
2212 && lru != &delta_base_cache_lru;
2213 lru = lru->next) {
2214 struct delta_base_cache_entry *f = (void *)lru;
2215 release_delta_base_cache(f);
2216 }
2217
2218 ent->p = p;
2219 ent->base_offset = base_offset;
2220 ent->type = type;
2221 ent->data = base;
2222 ent->size = base_size;
2223 ent->lru.next = &delta_base_cache_lru;
2224 ent->lru.prev = delta_base_cache_lru.prev;
2225 delta_base_cache_lru.prev->next = &ent->lru;
2226 delta_base_cache_lru.prev = &ent->lru;
2227}
2228
2229static void *read_object(const unsigned char *sha1, enum object_type *type,
2230 unsigned long *size);
2231
2232static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2233{
2234 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2235 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2236 p->pack_name, (uintmax_t)obj_offset);
2237}
2238
2239int do_check_packed_object_crc;
2240
2241#define UNPACK_ENTRY_STACK_PREALLOC 64
2242struct unpack_entry_stack_ent {
2243 off_t obj_offset;
2244 off_t curpos;
2245 unsigned long size;
2246};
2247
2248void *unpack_entry(struct packed_git *p, off_t obj_offset,
2249 enum object_type *final_type, unsigned long *final_size)
2250{
2251 struct pack_window *w_curs = NULL;
2252 off_t curpos = obj_offset;
2253 void *data = NULL;
2254 unsigned long size;
2255 enum object_type type;
2256 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2257 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2258 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2259 int base_from_cache = 0;
2260
2261 write_pack_access_log(p, obj_offset);
2262
2263 /* PHASE 1: drill down to the innermost base object */
2264 for (;;) {
2265 off_t base_offset;
2266 int i;
2267 struct delta_base_cache_entry *ent;
2268
2269 ent = get_delta_base_cache_entry(p, curpos);
2270 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2271 type = ent->type;
2272 data = ent->data;
2273 size = ent->size;
2274 clear_delta_base_cache_entry(ent);
2275 base_from_cache = 1;
2276 break;
2277 }
2278
2279 if (do_check_packed_object_crc && p->index_version > 1) {
2280 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2281 unsigned long len = revidx[1].offset - obj_offset;
2282 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2283 const unsigned char *sha1 =
2284 nth_packed_object_sha1(p, revidx->nr);
2285 error("bad packed object CRC for %s",
2286 sha1_to_hex(sha1));
2287 mark_bad_packed_object(p, sha1);
2288 unuse_pack(&w_curs);
2289 return NULL;
2290 }
2291 }
2292
2293 type = unpack_object_header(p, &w_curs, &curpos, &size);
2294 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2295 break;
2296
2297 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2298 if (!base_offset) {
2299 error("failed to validate delta base reference "
2300 "at offset %"PRIuMAX" from %s",
2301 (uintmax_t)curpos, p->pack_name);
2302 /* bail to phase 2, in hopes of recovery */
2303 data = NULL;
2304 break;
2305 }
2306
2307 /* push object, proceed to base */
2308 if (delta_stack_nr >= delta_stack_alloc
2309 && delta_stack == small_delta_stack) {
2310 delta_stack_alloc = alloc_nr(delta_stack_nr);
2311 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2312 memcpy(delta_stack, small_delta_stack,
2313 sizeof(*delta_stack)*delta_stack_nr);
2314 } else {
2315 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2316 }
2317 i = delta_stack_nr++;
2318 delta_stack[i].obj_offset = obj_offset;
2319 delta_stack[i].curpos = curpos;
2320 delta_stack[i].size = size;
2321
2322 curpos = obj_offset = base_offset;
2323 }
2324
2325 /* PHASE 2: handle the base */
2326 switch (type) {
2327 case OBJ_OFS_DELTA:
2328 case OBJ_REF_DELTA:
2329 if (data)
2330 die("BUG in unpack_entry: left loop at a valid delta");
2331 break;
2332 case OBJ_COMMIT:
2333 case OBJ_TREE:
2334 case OBJ_BLOB:
2335 case OBJ_TAG:
2336 if (!base_from_cache)
2337 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2338 break;
2339 default:
2340 data = NULL;
2341 error("unknown object type %i at offset %"PRIuMAX" in %s",
2342 type, (uintmax_t)obj_offset, p->pack_name);
2343 }
2344
2345 /* PHASE 3: apply deltas in order */
2346
2347 /* invariants:
2348 * 'data' holds the base data, or NULL if there was corruption
2349 */
2350 while (delta_stack_nr) {
2351 void *delta_data;
2352 void *base = data;
2353 unsigned long delta_size, base_size = size;
2354 int i;
2355
2356 data = NULL;
2357
2358 if (base)
2359 add_delta_base_cache(p, obj_offset, base, base_size, type);
2360
2361 if (!base) {
2362 /*
2363 * We're probably in deep shit, but let's try to fetch
2364 * the required base anyway from another pack or loose.
2365 * This is costly but should happen only in the presence
2366 * of a corrupted pack, and is better than failing outright.
2367 */
2368 struct revindex_entry *revidx;
2369 const unsigned char *base_sha1;
2370 revidx = find_pack_revindex(p, obj_offset);
2371 if (revidx) {
2372 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2373 error("failed to read delta base object %s"
2374 " at offset %"PRIuMAX" from %s",
2375 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2376 p->pack_name);
2377 mark_bad_packed_object(p, base_sha1);
2378 base = read_object(base_sha1, &type, &base_size);
2379 }
2380 }
2381
2382 i = --delta_stack_nr;
2383 obj_offset = delta_stack[i].obj_offset;
2384 curpos = delta_stack[i].curpos;
2385 delta_size = delta_stack[i].size;
2386
2387 if (!base)
2388 continue;
2389
2390 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2391
2392 if (!delta_data) {
2393 error("failed to unpack compressed delta "
2394 "at offset %"PRIuMAX" from %s",
2395 (uintmax_t)curpos, p->pack_name);
2396 data = NULL;
2397 continue;
2398 }
2399
2400 data = patch_delta(base, base_size,
2401 delta_data, delta_size,
2402 &size);
2403
2404 /*
2405 * We could not apply the delta; warn the user, but keep going.
2406 * Our failure will be noticed either in the next iteration of
2407 * the loop, or if this is the final delta, in the caller when
2408 * we return NULL. Those code paths will take care of making
2409 * a more explicit warning and retrying with another copy of
2410 * the object.
2411 */
2412 if (!data)
2413 error("failed to apply delta");
2414
2415 free(delta_data);
2416 }
2417
2418 *final_type = type;
2419 *final_size = size;
2420
2421 unuse_pack(&w_curs);
2422
2423 if (delta_stack != small_delta_stack)
2424 free(delta_stack);
2425
2426 return data;
2427}
2428
2429const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2430 uint32_t n)
2431{
2432 const unsigned char *index = p->index_data;
2433 if (!index) {
2434 if (open_pack_index(p))
2435 return NULL;
2436 index = p->index_data;
2437 }
2438 if (n >= p->num_objects)
2439 return NULL;
2440 index += 4 * 256;
2441 if (p->index_version == 1) {
2442 return index + 24 * n + 4;
2443 } else {
2444 index += 8;
2445 return index + 20 * n;
2446 }
2447}
2448
2449off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2450{
2451 const unsigned char *index = p->index_data;
2452 index += 4 * 256;
2453 if (p->index_version == 1) {
2454 return ntohl(*((uint32_t *)(index + 24 * n)));
2455 } else {
2456 uint32_t off;
2457 index += 8 + p->num_objects * (20 + 4);
2458 off = ntohl(*((uint32_t *)(index + 4 * n)));
2459 if (!(off & 0x80000000))
2460 return off;
2461 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2462 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2463 ntohl(*((uint32_t *)(index + 4)));
2464 }
2465}
2466
2467off_t find_pack_entry_one(const unsigned char *sha1,
2468 struct packed_git *p)
2469{
2470 const uint32_t *level1_ofs = p->index_data;
2471 const unsigned char *index = p->index_data;
2472 unsigned hi, lo, stride;
2473 static int use_lookup = -1;
2474 static int debug_lookup = -1;
2475
2476 if (debug_lookup < 0)
2477 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2478
2479 if (!index) {
2480 if (open_pack_index(p))
2481 return 0;
2482 level1_ofs = p->index_data;
2483 index = p->index_data;
2484 }
2485 if (p->index_version > 1) {
2486 level1_ofs += 2;
2487 index += 8;
2488 }
2489 index += 4 * 256;
2490 hi = ntohl(level1_ofs[*sha1]);
2491 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2492 if (p->index_version > 1) {
2493 stride = 20;
2494 } else {
2495 stride = 24;
2496 index += 4;
2497 }
2498
2499 if (debug_lookup)
2500 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2501 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2502
2503 if (use_lookup < 0)
2504 use_lookup = !!getenv("GIT_USE_LOOKUP");
2505 if (use_lookup) {
2506 int pos = sha1_entry_pos(index, stride, 0,
2507 lo, hi, p->num_objects, sha1);
2508 if (pos < 0)
2509 return 0;
2510 return nth_packed_object_offset(p, pos);
2511 }
2512
2513 do {
2514 unsigned mi = (lo + hi) / 2;
2515 int cmp = hashcmp(index + mi * stride, sha1);
2516
2517 if (debug_lookup)
2518 printf("lo %u hi %u rg %u mi %u\n",
2519 lo, hi, hi - lo, mi);
2520 if (!cmp)
2521 return nth_packed_object_offset(p, mi);
2522 if (cmp > 0)
2523 hi = mi;
2524 else
2525 lo = mi+1;
2526 } while (lo < hi);
2527 return 0;
2528}
2529
2530int is_pack_valid(struct packed_git *p)
2531{
2532 /* An already open pack is known to be valid. */
2533 if (p->pack_fd != -1)
2534 return 1;
2535
2536 /* If the pack has one window completely covering the
2537 * file size, the pack is known to be valid even if
2538 * the descriptor is not currently open.
2539 */
2540 if (p->windows) {
2541 struct pack_window *w = p->windows;
2542
2543 if (!w->offset && w->len == p->pack_size)
2544 return 1;
2545 }
2546
2547 /* Force the pack to open to prove its valid. */
2548 return !open_packed_git(p);
2549}
2550
2551static int fill_pack_entry(const unsigned char *sha1,
2552 struct pack_entry *e,
2553 struct packed_git *p)
2554{
2555 off_t offset;
2556
2557 if (p->num_bad_objects) {
2558 unsigned i;
2559 for (i = 0; i < p->num_bad_objects; i++)
2560 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2561 return 0;
2562 }
2563
2564 offset = find_pack_entry_one(sha1, p);
2565 if (!offset)
2566 return 0;
2567
2568 /*
2569 * We are about to tell the caller where they can locate the
2570 * requested object. We better make sure the packfile is
2571 * still here and can be accessed before supplying that
2572 * answer, as it may have been deleted since the index was
2573 * loaded!
2574 */
2575 if (!is_pack_valid(p))
2576 return 0;
2577 e->offset = offset;
2578 e->p = p;
2579 hashcpy(e->sha1, sha1);
2580 return 1;
2581}
2582
2583/*
2584 * Iff a pack file contains the object named by sha1, return true and
2585 * store its location to e.
2586 */
2587static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2588{
2589 struct packed_git *p;
2590
2591 prepare_packed_git();
2592 if (!packed_git)
2593 return 0;
2594
2595 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2596 return 1;
2597
2598 for (p = packed_git; p; p = p->next) {
2599 if (p == last_found_pack)
2600 continue; /* we already checked this one */
2601
2602 if (fill_pack_entry(sha1, e, p)) {
2603 last_found_pack = p;
2604 return 1;
2605 }
2606 }
2607 return 0;
2608}
2609
2610struct packed_git *find_sha1_pack(const unsigned char *sha1,
2611 struct packed_git *packs)
2612{
2613 struct packed_git *p;
2614
2615 for (p = packs; p; p = p->next) {
2616 if (find_pack_entry_one(sha1, p))
2617 return p;
2618 }
2619 return NULL;
2620
2621}
2622
2623static int sha1_loose_object_info(const unsigned char *sha1,
2624 struct object_info *oi,
2625 int flags)
2626{
2627 int status = 0;
2628 unsigned long mapsize;
2629 void *map;
2630 git_zstream stream;
2631 char hdr[32];
2632 struct strbuf hdrbuf = STRBUF_INIT;
2633
2634 if (oi->delta_base_sha1)
2635 hashclr(oi->delta_base_sha1);
2636
2637 /*
2638 * If we don't care about type or size, then we don't
2639 * need to look inside the object at all. Note that we
2640 * do not optimize out the stat call, even if the
2641 * caller doesn't care about the disk-size, since our
2642 * return value implicitly indicates whether the
2643 * object even exists.
2644 */
2645 if (!oi->typep && !oi->typename && !oi->sizep) {
2646 struct stat st;
2647 if (stat_sha1_file(sha1, &st) < 0)
2648 return -1;
2649 if (oi->disk_sizep)
2650 *oi->disk_sizep = st.st_size;
2651 return 0;
2652 }
2653
2654 map = map_sha1_file(sha1, &mapsize);
2655 if (!map)
2656 return -1;
2657 if (oi->disk_sizep)
2658 *oi->disk_sizep = mapsize;
2659 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2660 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2661 status = error("unable to unpack %s header with --allow-unknown-type",
2662 sha1_to_hex(sha1));
2663 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2664 status = error("unable to unpack %s header",
2665 sha1_to_hex(sha1));
2666 if (status < 0)
2667 ; /* Do nothing */
2668 else if (hdrbuf.len) {
2669 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2670 status = error("unable to parse %s header with --allow-unknown-type",
2671 sha1_to_hex(sha1));
2672 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2673 status = error("unable to parse %s header", sha1_to_hex(sha1));
2674 git_inflate_end(&stream);
2675 munmap(map, mapsize);
2676 if (status && oi->typep)
2677 *oi->typep = status;
2678 strbuf_release(&hdrbuf);
2679 return 0;
2680}
2681
2682int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2683{
2684 struct cached_object *co;
2685 struct pack_entry e;
2686 int rtype;
2687 enum object_type real_type;
2688 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2689
2690 co = find_cached_object(real);
2691 if (co) {
2692 if (oi->typep)
2693 *(oi->typep) = co->type;
2694 if (oi->sizep)
2695 *(oi->sizep) = co->size;
2696 if (oi->disk_sizep)
2697 *(oi->disk_sizep) = 0;
2698 if (oi->delta_base_sha1)
2699 hashclr(oi->delta_base_sha1);
2700 if (oi->typename)
2701 strbuf_addstr(oi->typename, typename(co->type));
2702 oi->whence = OI_CACHED;
2703 return 0;
2704 }
2705
2706 if (!find_pack_entry(real, &e)) {
2707 /* Most likely it's a loose object. */
2708 if (!sha1_loose_object_info(real, oi, flags)) {
2709 oi->whence = OI_LOOSE;
2710 return 0;
2711 }
2712
2713 /* Not a loose object; someone else may have just packed it. */
2714 reprepare_packed_git();
2715 if (!find_pack_entry(real, &e))
2716 return -1;
2717 }
2718
2719 /*
2720 * packed_object_info() does not follow the delta chain to
2721 * find out the real type, unless it is given oi->typep.
2722 */
2723 if (oi->typename && !oi->typep)
2724 oi->typep = &real_type;
2725
2726 rtype = packed_object_info(e.p, e.offset, oi);
2727 if (rtype < 0) {
2728 mark_bad_packed_object(e.p, real);
2729 if (oi->typep == &real_type)
2730 oi->typep = NULL;
2731 return sha1_object_info_extended(real, oi, 0);
2732 } else if (in_delta_base_cache(e.p, e.offset)) {
2733 oi->whence = OI_DBCACHED;
2734 } else {
2735 oi->whence = OI_PACKED;
2736 oi->u.packed.offset = e.offset;
2737 oi->u.packed.pack = e.p;
2738 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2739 rtype == OBJ_OFS_DELTA);
2740 }
2741 if (oi->typename)
2742 strbuf_addstr(oi->typename, typename(*oi->typep));
2743 if (oi->typep == &real_type)
2744 oi->typep = NULL;
2745
2746 return 0;
2747}
2748
2749/* returns enum object_type or negative */
2750int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2751{
2752 enum object_type type;
2753 struct object_info oi = {NULL};
2754
2755 oi.typep = &type;
2756 oi.sizep = sizep;
2757 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2758 return -1;
2759 return type;
2760}
2761
2762static void *read_packed_sha1(const unsigned char *sha1,
2763 enum object_type *type, unsigned long *size)
2764{
2765 struct pack_entry e;
2766 void *data;
2767
2768 if (!find_pack_entry(sha1, &e))
2769 return NULL;
2770 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2771 if (!data) {
2772 /*
2773 * We're probably in deep shit, but let's try to fetch
2774 * the required object anyway from another pack or loose.
2775 * This should happen only in the presence of a corrupted
2776 * pack, and is better than failing outright.
2777 */
2778 error("failed to read object %s at offset %"PRIuMAX" from %s",
2779 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2780 mark_bad_packed_object(e.p, sha1);
2781 data = read_object(sha1, type, size);
2782 }
2783 return data;
2784}
2785
2786int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2787 unsigned char *sha1)
2788{
2789 struct cached_object *co;
2790
2791 hash_sha1_file(buf, len, typename(type), sha1);
2792 if (has_sha1_file(sha1) || find_cached_object(sha1))
2793 return 0;
2794 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2795 co = &cached_objects[cached_object_nr++];
2796 co->size = len;
2797 co->type = type;
2798 co->buf = xmalloc(len);
2799 memcpy(co->buf, buf, len);
2800 hashcpy(co->sha1, sha1);
2801 return 0;
2802}
2803
2804static void *read_object(const unsigned char *sha1, enum object_type *type,
2805 unsigned long *size)
2806{
2807 unsigned long mapsize;
2808 void *map, *buf;
2809 struct cached_object *co;
2810
2811 co = find_cached_object(sha1);
2812 if (co) {
2813 *type = co->type;
2814 *size = co->size;
2815 return xmemdupz(co->buf, co->size);
2816 }
2817
2818 buf = read_packed_sha1(sha1, type, size);
2819 if (buf)
2820 return buf;
2821 map = map_sha1_file(sha1, &mapsize);
2822 if (map) {
2823 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2824 munmap(map, mapsize);
2825 return buf;
2826 }
2827 reprepare_packed_git();
2828 return read_packed_sha1(sha1, type, size);
2829}
2830
2831/*
2832 * This function dies on corrupt objects; the callers who want to
2833 * deal with them should arrange to call read_object() and give error
2834 * messages themselves.
2835 */
2836void *read_sha1_file_extended(const unsigned char *sha1,
2837 enum object_type *type,
2838 unsigned long *size,
2839 unsigned flag)
2840{
2841 void *data;
2842 const struct packed_git *p;
2843 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2844
2845 errno = 0;
2846 data = read_object(repl, type, size);
2847 if (data)
2848 return data;
2849
2850 if (errno && errno != ENOENT)
2851 die_errno("failed to read object %s", sha1_to_hex(sha1));
2852
2853 /* die if we replaced an object with one that does not exist */
2854 if (repl != sha1)
2855 die("replacement %s not found for %s",
2856 sha1_to_hex(repl), sha1_to_hex(sha1));
2857
2858 if (has_loose_object(repl)) {
2859 const char *path = sha1_file_name(sha1);
2860
2861 die("loose object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl), path);
2863 }
2864
2865 if ((p = has_packed_and_bad(repl)) != NULL)
2866 die("packed object %s (stored in %s) is corrupt",
2867 sha1_to_hex(repl), p->pack_name);
2868
2869 return NULL;
2870}
2871
2872void *read_object_with_reference(const unsigned char *sha1,
2873 const char *required_type_name,
2874 unsigned long *size,
2875 unsigned char *actual_sha1_return)
2876{
2877 enum object_type type, required_type;
2878 void *buffer;
2879 unsigned long isize;
2880 unsigned char actual_sha1[20];
2881
2882 required_type = type_from_string(required_type_name);
2883 hashcpy(actual_sha1, sha1);
2884 while (1) {
2885 int ref_length = -1;
2886 const char *ref_type = NULL;
2887
2888 buffer = read_sha1_file(actual_sha1, &type, &isize);
2889 if (!buffer)
2890 return NULL;
2891 if (type == required_type) {
2892 *size = isize;
2893 if (actual_sha1_return)
2894 hashcpy(actual_sha1_return, actual_sha1);
2895 return buffer;
2896 }
2897 /* Handle references */
2898 else if (type == OBJ_COMMIT)
2899 ref_type = "tree ";
2900 else if (type == OBJ_TAG)
2901 ref_type = "object ";
2902 else {
2903 free(buffer);
2904 return NULL;
2905 }
2906 ref_length = strlen(ref_type);
2907
2908 if (ref_length + 40 > isize ||
2909 memcmp(buffer, ref_type, ref_length) ||
2910 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2911 free(buffer);
2912 return NULL;
2913 }
2914 free(buffer);
2915 /* Now we have the ID of the referred-to object in
2916 * actual_sha1. Check again. */
2917 }
2918}
2919
2920static void write_sha1_file_prepare(const void *buf, unsigned long len,
2921 const char *type, unsigned char *sha1,
2922 char *hdr, int *hdrlen)
2923{
2924 git_SHA_CTX c;
2925
2926 /* Generate the header */
2927 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2928
2929 /* Sha1.. */
2930 git_SHA1_Init(&c);
2931 git_SHA1_Update(&c, hdr, *hdrlen);
2932 git_SHA1_Update(&c, buf, len);
2933 git_SHA1_Final(sha1, &c);
2934}
2935
2936/*
2937 * Move the just written object into its final resting place.
2938 */
2939int finalize_object_file(const char *tmpfile, const char *filename)
2940{
2941 int ret = 0;
2942
2943 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2944 goto try_rename;
2945 else if (link(tmpfile, filename))
2946 ret = errno;
2947
2948 /*
2949 * Coda hack - coda doesn't like cross-directory links,
2950 * so we fall back to a rename, which will mean that it
2951 * won't be able to check collisions, but that's not a
2952 * big deal.
2953 *
2954 * The same holds for FAT formatted media.
2955 *
2956 * When this succeeds, we just return. We have nothing
2957 * left to unlink.
2958 */
2959 if (ret && ret != EEXIST) {
2960 try_rename:
2961 if (!rename(tmpfile, filename))
2962 goto out;
2963 ret = errno;
2964 }
2965 unlink_or_warn(tmpfile);
2966 if (ret) {
2967 if (ret != EEXIST) {
2968 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2969 }
2970 /* FIXME!!! Collision check here ? */
2971 }
2972
2973out:
2974 if (adjust_shared_perm(filename))
2975 return error("unable to set permission to '%s'", filename);
2976 return 0;
2977}
2978
2979static int write_buffer(int fd, const void *buf, size_t len)
2980{
2981 if (write_in_full(fd, buf, len) < 0)
2982 return error("file write error (%s)", strerror(errno));
2983 return 0;
2984}
2985
2986int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2987 unsigned char *sha1)
2988{
2989 char hdr[32];
2990 int hdrlen = sizeof(hdr);
2991 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2992 return 0;
2993}
2994
2995/* Finalize a file on disk, and close it. */
2996static void close_sha1_file(int fd)
2997{
2998 if (fsync_object_files)
2999 fsync_or_die(fd, "sha1 file");
3000 if (close(fd) != 0)
3001 die_errno("error when closing sha1 file");
3002}
3003
3004/* Size of directory component, including the ending '/' */
3005static inline int directory_size(const char *filename)
3006{
3007 const char *s = strrchr(filename, '/');
3008 if (!s)
3009 return 0;
3010 return s - filename + 1;
3011}
3012
3013/*
3014 * This creates a temporary file in the same directory as the final
3015 * 'filename'
3016 *
3017 * We want to avoid cross-directory filename renames, because those
3018 * can have problems on various filesystems (FAT, NFS, Coda).
3019 */
3020static int create_tmpfile(struct strbuf *tmp, const char *filename)
3021{
3022 int fd, dirlen = directory_size(filename);
3023
3024 strbuf_reset(tmp);
3025 strbuf_add(tmp, filename, dirlen);
3026 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3027 fd = git_mkstemp_mode(tmp->buf, 0444);
3028 if (fd < 0 && dirlen && errno == ENOENT) {
3029 /*
3030 * Make sure the directory exists; note that the contents
3031 * of the buffer are undefined after mkstemp returns an
3032 * error, so we have to rewrite the whole buffer from
3033 * scratch.
3034 */
3035 strbuf_reset(tmp);
3036 strbuf_add(tmp, filename, dirlen - 1);
3037 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3038 return -1;
3039 if (adjust_shared_perm(tmp->buf))
3040 return -1;
3041
3042 /* Try again */
3043 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3044 fd = git_mkstemp_mode(tmp->buf, 0444);
3045 }
3046 return fd;
3047}
3048
3049static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3050 const void *buf, unsigned long len, time_t mtime)
3051{
3052 int fd, ret;
3053 unsigned char compressed[4096];
3054 git_zstream stream;
3055 git_SHA_CTX c;
3056 unsigned char parano_sha1[20];
3057 static struct strbuf tmp_file = STRBUF_INIT;
3058 const char *filename = sha1_file_name(sha1);
3059
3060 fd = create_tmpfile(&tmp_file, filename);
3061 if (fd < 0) {
3062 if (errno == EACCES)
3063 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3064 else
3065 return error("unable to create temporary file: %s", strerror(errno));
3066 }
3067
3068 /* Set it up */
3069 git_deflate_init(&stream, zlib_compression_level);
3070 stream.next_out = compressed;
3071 stream.avail_out = sizeof(compressed);
3072 git_SHA1_Init(&c);
3073
3074 /* First header.. */
3075 stream.next_in = (unsigned char *)hdr;
3076 stream.avail_in = hdrlen;
3077 while (git_deflate(&stream, 0) == Z_OK)
3078 ; /* nothing */
3079 git_SHA1_Update(&c, hdr, hdrlen);
3080
3081 /* Then the data itself.. */
3082 stream.next_in = (void *)buf;
3083 stream.avail_in = len;
3084 do {
3085 unsigned char *in0 = stream.next_in;
3086 ret = git_deflate(&stream, Z_FINISH);
3087 git_SHA1_Update(&c, in0, stream.next_in - in0);
3088 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3089 die("unable to write sha1 file");
3090 stream.next_out = compressed;
3091 stream.avail_out = sizeof(compressed);
3092 } while (ret == Z_OK);
3093
3094 if (ret != Z_STREAM_END)
3095 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3096 ret = git_deflate_end_gently(&stream);
3097 if (ret != Z_OK)
3098 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3099 git_SHA1_Final(parano_sha1, &c);
3100 if (hashcmp(sha1, parano_sha1) != 0)
3101 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3102
3103 close_sha1_file(fd);
3104
3105 if (mtime) {
3106 struct utimbuf utb;
3107 utb.actime = mtime;
3108 utb.modtime = mtime;
3109 if (utime(tmp_file.buf, &utb) < 0)
3110 warning("failed utime() on %s: %s",
3111 tmp_file.buf, strerror(errno));
3112 }
3113
3114 return finalize_object_file(tmp_file.buf, filename);
3115}
3116
3117static int freshen_loose_object(const unsigned char *sha1)
3118{
3119 return check_and_freshen(sha1, 1);
3120}
3121
3122static int freshen_packed_object(const unsigned char *sha1)
3123{
3124 struct pack_entry e;
3125 if (!find_pack_entry(sha1, &e))
3126 return 0;
3127 if (e.p->freshened)
3128 return 1;
3129 if (!freshen_file(e.p->pack_name))
3130 return 0;
3131 e.p->freshened = 1;
3132 return 1;
3133}
3134
3135int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3136{
3137 char hdr[32];
3138 int hdrlen = sizeof(hdr);
3139
3140 /* Normally if we have it in the pack then we do not bother writing
3141 * it out into .git/objects/??/?{38} file.
3142 */
3143 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3144 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3145 return 0;
3146 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3147}
3148
3149int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3150 unsigned char *sha1, unsigned flags)
3151{
3152 char *header;
3153 int hdrlen, status = 0;
3154
3155 /* type string, SP, %lu of the length plus NUL must fit this */
3156 hdrlen = strlen(type) + 32;
3157 header = xmalloc(hdrlen);
3158 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3159
3160 if (!(flags & HASH_WRITE_OBJECT))
3161 goto cleanup;
3162 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3163 goto cleanup;
3164 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3165
3166cleanup:
3167 free(header);
3168 return status;
3169}
3170
3171int force_object_loose(const unsigned char *sha1, time_t mtime)
3172{
3173 void *buf;
3174 unsigned long len;
3175 enum object_type type;
3176 char hdr[32];
3177 int hdrlen;
3178 int ret;
3179
3180 if (has_loose_object(sha1))
3181 return 0;
3182 buf = read_packed_sha1(sha1, &type, &len);
3183 if (!buf)
3184 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3185 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3186 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3187 free(buf);
3188
3189 return ret;
3190}
3191
3192int has_pack_index(const unsigned char *sha1)
3193{
3194 struct stat st;
3195 if (stat(sha1_pack_index_name(sha1), &st))
3196 return 0;
3197 return 1;
3198}
3199
3200int has_sha1_pack(const unsigned char *sha1)
3201{
3202 struct pack_entry e;
3203 return find_pack_entry(sha1, &e);
3204}
3205
3206int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3207{
3208 struct pack_entry e;
3209
3210 if (find_pack_entry(sha1, &e))
3211 return 1;
3212 if (has_loose_object(sha1))
3213 return 1;
3214 if (flags & HAS_SHA1_QUICK)
3215 return 0;
3216 reprepare_packed_git();
3217 return find_pack_entry(sha1, &e);
3218}
3219
3220int has_object_file(const struct object_id *oid)
3221{
3222 return has_sha1_file(oid->hash);
3223}
3224
3225static void check_tree(const void *buf, size_t size)
3226{
3227 struct tree_desc desc;
3228 struct name_entry entry;
3229
3230 init_tree_desc(&desc, buf, size);
3231 while (tree_entry(&desc, &entry))
3232 /* do nothing
3233 * tree_entry() will die() on malformed entries */
3234 ;
3235}
3236
3237static void check_commit(const void *buf, size_t size)
3238{
3239 struct commit c;
3240 memset(&c, 0, sizeof(c));
3241 if (parse_commit_buffer(&c, buf, size))
3242 die("corrupt commit");
3243}
3244
3245static void check_tag(const void *buf, size_t size)
3246{
3247 struct tag t;
3248 memset(&t, 0, sizeof(t));
3249 if (parse_tag_buffer(&t, buf, size))
3250 die("corrupt tag");
3251}
3252
3253static int index_mem(unsigned char *sha1, void *buf, size_t size,
3254 enum object_type type,
3255 const char *path, unsigned flags)
3256{
3257 int ret, re_allocated = 0;
3258 int write_object = flags & HASH_WRITE_OBJECT;
3259
3260 if (!type)
3261 type = OBJ_BLOB;
3262
3263 /*
3264 * Convert blobs to git internal format
3265 */
3266 if ((type == OBJ_BLOB) && path) {
3267 struct strbuf nbuf = STRBUF_INIT;
3268 if (convert_to_git(path, buf, size, &nbuf,
3269 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3270 buf = strbuf_detach(&nbuf, &size);
3271 re_allocated = 1;
3272 }
3273 }
3274 if (flags & HASH_FORMAT_CHECK) {
3275 if (type == OBJ_TREE)
3276 check_tree(buf, size);
3277 if (type == OBJ_COMMIT)
3278 check_commit(buf, size);
3279 if (type == OBJ_TAG)
3280 check_tag(buf, size);
3281 }
3282
3283 if (write_object)
3284 ret = write_sha1_file(buf, size, typename(type), sha1);
3285 else
3286 ret = hash_sha1_file(buf, size, typename(type), sha1);
3287 if (re_allocated)
3288 free(buf);
3289 return ret;
3290}
3291
3292static int index_stream_convert_blob(unsigned char *sha1, int fd,
3293 const char *path, unsigned flags)
3294{
3295 int ret;
3296 const int write_object = flags & HASH_WRITE_OBJECT;
3297 struct strbuf sbuf = STRBUF_INIT;
3298
3299 assert(path);
3300 assert(would_convert_to_git_filter_fd(path));
3301
3302 convert_to_git_filter_fd(path, fd, &sbuf,
3303 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3304
3305 if (write_object)
3306 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3307 sha1);
3308 else
3309 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3310 sha1);
3311 strbuf_release(&sbuf);
3312 return ret;
3313}
3314
3315static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3316 const char *path, unsigned flags)
3317{
3318 struct strbuf sbuf = STRBUF_INIT;
3319 int ret;
3320
3321 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3322 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3323 else
3324 ret = -1;
3325 strbuf_release(&sbuf);
3326 return ret;
3327}
3328
3329#define SMALL_FILE_SIZE (32*1024)
3330
3331static int index_core(unsigned char *sha1, int fd, size_t size,
3332 enum object_type type, const char *path,
3333 unsigned flags)
3334{
3335 int ret;
3336
3337 if (!size) {
3338 ret = index_mem(sha1, "", size, type, path, flags);
3339 } else if (size <= SMALL_FILE_SIZE) {
3340 char *buf = xmalloc(size);
3341 if (size == read_in_full(fd, buf, size))
3342 ret = index_mem(sha1, buf, size, type, path, flags);
3343 else
3344 ret = error("short read %s", strerror(errno));
3345 free(buf);
3346 } else {
3347 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3348 ret = index_mem(sha1, buf, size, type, path, flags);
3349 munmap(buf, size);
3350 }
3351 return ret;
3352}
3353
3354/*
3355 * This creates one packfile per large blob unless bulk-checkin
3356 * machinery is "plugged".
3357 *
3358 * This also bypasses the usual "convert-to-git" dance, and that is on
3359 * purpose. We could write a streaming version of the converting
3360 * functions and insert that before feeding the data to fast-import
3361 * (or equivalent in-core API described above). However, that is
3362 * somewhat complicated, as we do not know the size of the filter
3363 * result, which we need to know beforehand when writing a git object.
3364 * Since the primary motivation for trying to stream from the working
3365 * tree file and to avoid mmaping it in core is to deal with large
3366 * binary blobs, they generally do not want to get any conversion, and
3367 * callers should avoid this code path when filters are requested.
3368 */
3369static int index_stream(unsigned char *sha1, int fd, size_t size,
3370 enum object_type type, const char *path,
3371 unsigned flags)
3372{
3373 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3374}
3375
3376int index_fd(unsigned char *sha1, int fd, struct stat *st,
3377 enum object_type type, const char *path, unsigned flags)
3378{
3379 int ret;
3380
3381 /*
3382 * Call xsize_t() only when needed to avoid potentially unnecessary
3383 * die() for large files.
3384 */
3385 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3386 ret = index_stream_convert_blob(sha1, fd, path, flags);
3387 else if (!S_ISREG(st->st_mode))
3388 ret = index_pipe(sha1, fd, type, path, flags);
3389 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3390 (path && would_convert_to_git(path)))
3391 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3392 flags);
3393 else
3394 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3395 flags);
3396 close(fd);
3397 return ret;
3398}
3399
3400int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3401{
3402 int fd;
3403 struct strbuf sb = STRBUF_INIT;
3404
3405 switch (st->st_mode & S_IFMT) {
3406 case S_IFREG:
3407 fd = open(path, O_RDONLY);
3408 if (fd < 0)
3409 return error("open(\"%s\"): %s", path,
3410 strerror(errno));
3411 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3412 return error("%s: failed to insert into database",
3413 path);
3414 break;
3415 case S_IFLNK:
3416 if (strbuf_readlink(&sb, path, st->st_size)) {
3417 char *errstr = strerror(errno);
3418 return error("readlink(\"%s\"): %s", path,
3419 errstr);
3420 }
3421 if (!(flags & HASH_WRITE_OBJECT))
3422 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3423 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3424 return error("%s: failed to insert into database",
3425 path);
3426 strbuf_release(&sb);
3427 break;
3428 case S_IFDIR:
3429 return resolve_gitlink_ref(path, "HEAD", sha1);
3430 default:
3431 return error("%s: unsupported file type", path);
3432 }
3433 return 0;
3434}
3435
3436int read_pack_header(int fd, struct pack_header *header)
3437{
3438 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3439 /* "eof before pack header was fully read" */
3440 return PH_ERROR_EOF;
3441
3442 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3443 /* "protocol error (pack signature mismatch detected)" */
3444 return PH_ERROR_PACK_SIGNATURE;
3445 if (!pack_version_ok(header->hdr_version))
3446 /* "protocol error (pack version unsupported)" */
3447 return PH_ERROR_PROTOCOL;
3448 return 0;
3449}
3450
3451void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3452{
3453 enum object_type type = sha1_object_info(sha1, NULL);
3454 if (type < 0)
3455 die("%s is not a valid object", sha1_to_hex(sha1));
3456 if (type != expect)
3457 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3458 typename(expect));
3459}
3460
3461static int for_each_file_in_obj_subdir(int subdir_nr,
3462 struct strbuf *path,
3463 each_loose_object_fn obj_cb,
3464 each_loose_cruft_fn cruft_cb,
3465 each_loose_subdir_fn subdir_cb,
3466 void *data)
3467{
3468 size_t baselen = path->len;
3469 DIR *dir = opendir(path->buf);
3470 struct dirent *de;
3471 int r = 0;
3472
3473 if (!dir) {
3474 if (errno == ENOENT)
3475 return 0;
3476 return error("unable to open %s: %s", path->buf, strerror(errno));
3477 }
3478
3479 while ((de = readdir(dir))) {
3480 if (is_dot_or_dotdot(de->d_name))
3481 continue;
3482
3483 strbuf_setlen(path, baselen);
3484 strbuf_addf(path, "/%s", de->d_name);
3485
3486 if (strlen(de->d_name) == 38) {
3487 char hex[41];
3488 unsigned char sha1[20];
3489
3490 snprintf(hex, sizeof(hex), "%02x%s",
3491 subdir_nr, de->d_name);
3492 if (!get_sha1_hex(hex, sha1)) {
3493 if (obj_cb) {
3494 r = obj_cb(sha1, path->buf, data);
3495 if (r)
3496 break;
3497 }
3498 continue;
3499 }
3500 }
3501
3502 if (cruft_cb) {
3503 r = cruft_cb(de->d_name, path->buf, data);
3504 if (r)
3505 break;
3506 }
3507 }
3508 closedir(dir);
3509
3510 strbuf_setlen(path, baselen);
3511 if (!r && subdir_cb)
3512 r = subdir_cb(subdir_nr, path->buf, data);
3513
3514 return r;
3515}
3516
3517int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3518 each_loose_object_fn obj_cb,
3519 each_loose_cruft_fn cruft_cb,
3520 each_loose_subdir_fn subdir_cb,
3521 void *data)
3522{
3523 size_t baselen = path->len;
3524 int r = 0;
3525 int i;
3526
3527 for (i = 0; i < 256; i++) {
3528 strbuf_addf(path, "/%02x", i);
3529 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3530 subdir_cb, data);
3531 strbuf_setlen(path, baselen);
3532 if (r)
3533 break;
3534 }
3535
3536 return r;
3537}
3538
3539int for_each_loose_file_in_objdir(const char *path,
3540 each_loose_object_fn obj_cb,
3541 each_loose_cruft_fn cruft_cb,
3542 each_loose_subdir_fn subdir_cb,
3543 void *data)
3544{
3545 struct strbuf buf = STRBUF_INIT;
3546 int r;
3547
3548 strbuf_addstr(&buf, path);
3549 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3550 subdir_cb, data);
3551 strbuf_release(&buf);
3552
3553 return r;
3554}
3555
3556struct loose_alt_odb_data {
3557 each_loose_object_fn *cb;
3558 void *data;
3559};
3560
3561static int loose_from_alt_odb(struct alternate_object_database *alt,
3562 void *vdata)
3563{
3564 struct loose_alt_odb_data *data = vdata;
3565 struct strbuf buf = STRBUF_INIT;
3566 int r;
3567
3568 /* copy base not including trailing '/' */
3569 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3570 r = for_each_loose_file_in_objdir_buf(&buf,
3571 data->cb, NULL, NULL,
3572 data->data);
3573 strbuf_release(&buf);
3574 return r;
3575}
3576
3577int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3578{
3579 struct loose_alt_odb_data alt;
3580 int r;
3581
3582 r = for_each_loose_file_in_objdir(get_object_directory(),
3583 cb, NULL, NULL, data);
3584 if (r)
3585 return r;
3586
3587 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3588 return 0;
3589
3590 alt.cb = cb;
3591 alt.data = data;
3592 return foreach_alt_odb(loose_from_alt_odb, &alt);
3593}
3594
3595static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3596{
3597 uint32_t i;
3598 int r = 0;
3599
3600 for (i = 0; i < p->num_objects; i++) {
3601 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3602
3603 if (!sha1)
3604 return error("unable to get sha1 of object %u in %s",
3605 i, p->pack_name);
3606
3607 r = cb(sha1, p, i, data);
3608 if (r)
3609 break;
3610 }
3611 return r;
3612}
3613
3614int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3615{
3616 struct packed_git *p;
3617 int r = 0;
3618 int pack_errors = 0;
3619
3620 prepare_packed_git();
3621 for (p = packed_git; p; p = p->next) {
3622 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3623 continue;
3624 if (open_pack_index(p)) {
3625 pack_errors = 1;
3626 continue;
3627 }
3628 r = for_each_object_in_pack(p, cb, data);
3629 if (r)
3630 break;
3631 }
3632 return r ? r : pack_errors;
3633}