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 size_t 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 = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent = xmalloc(st_add(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, '\n') != 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(st_add(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 = st_add3(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 size_t alloc = st_add(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 + GIT_SHA1_RAWSZ * i))
1417 return;
1418 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1419 st_mult(GIT_SHA1_RAWSZ,
1420 st_add(p->num_bad_objects, 1)));
1421 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1422 p->num_bad_objects++;
1423}
1424
1425static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1426{
1427 struct packed_git *p;
1428 unsigned i;
1429
1430 for (p = packed_git; p; p = p->next)
1431 for (i = 0; i < p->num_bad_objects; i++)
1432 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1433 return p;
1434 return NULL;
1435}
1436
1437/*
1438 * With an in-core object data in "map", rehash it to make sure the
1439 * object name actually matches "sha1" to detect object corruption.
1440 * With "map" == NULL, try reading the object named with "sha1" using
1441 * the streaming interface and rehash it to do the same.
1442 */
1443int check_sha1_signature(const unsigned char *sha1, void *map,
1444 unsigned long size, const char *type)
1445{
1446 unsigned char real_sha1[20];
1447 enum object_type obj_type;
1448 struct git_istream *st;
1449 git_SHA_CTX c;
1450 char hdr[32];
1451 int hdrlen;
1452
1453 if (map) {
1454 hash_sha1_file(map, size, type, real_sha1);
1455 return hashcmp(sha1, real_sha1) ? -1 : 0;
1456 }
1457
1458 st = open_istream(sha1, &obj_type, &size, NULL);
1459 if (!st)
1460 return -1;
1461
1462 /* Generate the header */
1463 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1464
1465 /* Sha1.. */
1466 git_SHA1_Init(&c);
1467 git_SHA1_Update(&c, hdr, hdrlen);
1468 for (;;) {
1469 char buf[1024 * 16];
1470 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1471
1472 if (readlen < 0) {
1473 close_istream(st);
1474 return -1;
1475 }
1476 if (!readlen)
1477 break;
1478 git_SHA1_Update(&c, buf, readlen);
1479 }
1480 git_SHA1_Final(real_sha1, &c);
1481 close_istream(st);
1482 return hashcmp(sha1, real_sha1) ? -1 : 0;
1483}
1484
1485int git_open_noatime(const char *name)
1486{
1487 static int sha1_file_open_flag = O_NOATIME;
1488
1489 for (;;) {
1490 int fd;
1491
1492 errno = 0;
1493 fd = open(name, O_RDONLY | sha1_file_open_flag);
1494 if (fd >= 0)
1495 return fd;
1496
1497 /* Might the failure be due to O_NOATIME? */
1498 if (errno != ENOENT && sha1_file_open_flag) {
1499 sha1_file_open_flag = 0;
1500 continue;
1501 }
1502
1503 return -1;
1504 }
1505}
1506
1507static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1508{
1509 struct alternate_object_database *alt;
1510
1511 if (!lstat(sha1_file_name(sha1), st))
1512 return 0;
1513
1514 prepare_alt_odb();
1515 errno = ENOENT;
1516 for (alt = alt_odb_list; alt; alt = alt->next) {
1517 fill_sha1_path(alt->name, sha1);
1518 if (!lstat(alt->base, st))
1519 return 0;
1520 }
1521
1522 return -1;
1523}
1524
1525static int open_sha1_file(const unsigned char *sha1)
1526{
1527 int fd;
1528 struct alternate_object_database *alt;
1529 int most_interesting_errno;
1530
1531 fd = git_open_noatime(sha1_file_name(sha1));
1532 if (fd >= 0)
1533 return fd;
1534 most_interesting_errno = errno;
1535
1536 prepare_alt_odb();
1537 for (alt = alt_odb_list; alt; alt = alt->next) {
1538 fill_sha1_path(alt->name, sha1);
1539 fd = git_open_noatime(alt->base);
1540 if (fd >= 0)
1541 return fd;
1542 if (most_interesting_errno == ENOENT)
1543 most_interesting_errno = errno;
1544 }
1545 errno = most_interesting_errno;
1546 return -1;
1547}
1548
1549void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1550{
1551 void *map;
1552 int fd;
1553
1554 fd = open_sha1_file(sha1);
1555 map = NULL;
1556 if (fd >= 0) {
1557 struct stat st;
1558
1559 if (!fstat(fd, &st)) {
1560 *size = xsize_t(st.st_size);
1561 if (!*size) {
1562 /* mmap() is forbidden on empty files */
1563 error("object file %s is empty", sha1_file_name(sha1));
1564 return NULL;
1565 }
1566 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1567 }
1568 close(fd);
1569 }
1570 return map;
1571}
1572
1573unsigned long unpack_object_header_buffer(const unsigned char *buf,
1574 unsigned long len, enum object_type *type, unsigned long *sizep)
1575{
1576 unsigned shift;
1577 unsigned long size, c;
1578 unsigned long used = 0;
1579
1580 c = buf[used++];
1581 *type = (c >> 4) & 7;
1582 size = c & 15;
1583 shift = 4;
1584 while (c & 0x80) {
1585 if (len <= used || bitsizeof(long) <= shift) {
1586 error("bad object header");
1587 size = used = 0;
1588 break;
1589 }
1590 c = buf[used++];
1591 size += (c & 0x7f) << shift;
1592 shift += 7;
1593 }
1594 *sizep = size;
1595 return used;
1596}
1597
1598int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1599{
1600 /* Get the data stream */
1601 memset(stream, 0, sizeof(*stream));
1602 stream->next_in = map;
1603 stream->avail_in = mapsize;
1604 stream->next_out = buffer;
1605 stream->avail_out = bufsiz;
1606
1607 git_inflate_init(stream);
1608 return git_inflate(stream, 0);
1609}
1610
1611static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1612 unsigned long mapsize, void *buffer,
1613 unsigned long bufsiz, struct strbuf *header)
1614{
1615 int status;
1616
1617 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1618
1619 /*
1620 * Check if entire header is unpacked in the first iteration.
1621 */
1622 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1623 return 0;
1624
1625 /*
1626 * buffer[0..bufsiz] was not large enough. Copy the partial
1627 * result out to header, and then append the result of further
1628 * reading the stream.
1629 */
1630 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1631 stream->next_out = buffer;
1632 stream->avail_out = bufsiz;
1633
1634 do {
1635 status = git_inflate(stream, 0);
1636 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1637 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1638 return 0;
1639 stream->next_out = buffer;
1640 stream->avail_out = bufsiz;
1641 } while (status != Z_STREAM_END);
1642 return -1;
1643}
1644
1645static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1646{
1647 int bytes = strlen(buffer) + 1;
1648 unsigned char *buf = xmallocz(size);
1649 unsigned long n;
1650 int status = Z_OK;
1651
1652 n = stream->total_out - bytes;
1653 if (n > size)
1654 n = size;
1655 memcpy(buf, (char *) buffer + bytes, n);
1656 bytes = n;
1657 if (bytes <= size) {
1658 /*
1659 * The above condition must be (bytes <= size), not
1660 * (bytes < size). In other words, even though we
1661 * expect no more output and set avail_out to zero,
1662 * the input zlib stream may have bytes that express
1663 * "this concludes the stream", and we *do* want to
1664 * eat that input.
1665 *
1666 * Otherwise we would not be able to test that we
1667 * consumed all the input to reach the expected size;
1668 * we also want to check that zlib tells us that all
1669 * went well with status == Z_STREAM_END at the end.
1670 */
1671 stream->next_out = buf + bytes;
1672 stream->avail_out = size - bytes;
1673 while (status == Z_OK)
1674 status = git_inflate(stream, Z_FINISH);
1675 }
1676 if (status == Z_STREAM_END && !stream->avail_in) {
1677 git_inflate_end(stream);
1678 return buf;
1679 }
1680
1681 if (status < 0)
1682 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1683 else if (stream->avail_in)
1684 error("garbage at end of loose object '%s'",
1685 sha1_to_hex(sha1));
1686 free(buf);
1687 return NULL;
1688}
1689
1690/*
1691 * We used to just use "sscanf()", but that's actually way
1692 * too permissive for what we want to check. So do an anal
1693 * object header parse by hand.
1694 */
1695static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1696 unsigned int flags)
1697{
1698 const char *type_buf = hdr;
1699 unsigned long size;
1700 int type, type_len = 0;
1701
1702 /*
1703 * The type can be of any size but is followed by
1704 * a space.
1705 */
1706 for (;;) {
1707 char c = *hdr++;
1708 if (c == ' ')
1709 break;
1710 type_len++;
1711 }
1712
1713 type = type_from_string_gently(type_buf, type_len, 1);
1714 if (oi->typename)
1715 strbuf_add(oi->typename, type_buf, type_len);
1716 /*
1717 * Set type to 0 if its an unknown object and
1718 * we're obtaining the type using '--allow-unkown-type'
1719 * option.
1720 */
1721 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1722 type = 0;
1723 else if (type < 0)
1724 die("invalid object type");
1725 if (oi->typep)
1726 *oi->typep = type;
1727
1728 /*
1729 * The length must follow immediately, and be in canonical
1730 * decimal format (ie "010" is not valid).
1731 */
1732 size = *hdr++ - '0';
1733 if (size > 9)
1734 return -1;
1735 if (size) {
1736 for (;;) {
1737 unsigned long c = *hdr - '0';
1738 if (c > 9)
1739 break;
1740 hdr++;
1741 size = size * 10 + c;
1742 }
1743 }
1744
1745 if (oi->sizep)
1746 *oi->sizep = size;
1747
1748 /*
1749 * The length must be followed by a zero byte
1750 */
1751 return *hdr ? -1 : type;
1752}
1753
1754int parse_sha1_header(const char *hdr, unsigned long *sizep)
1755{
1756 struct object_info oi;
1757
1758 oi.sizep = sizep;
1759 oi.typename = NULL;
1760 oi.typep = NULL;
1761 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1762}
1763
1764static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1765{
1766 int ret;
1767 git_zstream stream;
1768 char hdr[8192];
1769
1770 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1771 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1772 return NULL;
1773
1774 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1775}
1776
1777unsigned long get_size_from_delta(struct packed_git *p,
1778 struct pack_window **w_curs,
1779 off_t curpos)
1780{
1781 const unsigned char *data;
1782 unsigned char delta_head[20], *in;
1783 git_zstream stream;
1784 int st;
1785
1786 memset(&stream, 0, sizeof(stream));
1787 stream.next_out = delta_head;
1788 stream.avail_out = sizeof(delta_head);
1789
1790 git_inflate_init(&stream);
1791 do {
1792 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1793 stream.next_in = in;
1794 st = git_inflate(&stream, Z_FINISH);
1795 curpos += stream.next_in - in;
1796 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1797 stream.total_out < sizeof(delta_head));
1798 git_inflate_end(&stream);
1799 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1800 error("delta data unpack-initial failed");
1801 return 0;
1802 }
1803
1804 /* Examine the initial part of the delta to figure out
1805 * the result size.
1806 */
1807 data = delta_head;
1808
1809 /* ignore base size */
1810 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1811
1812 /* Read the result size */
1813 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1814}
1815
1816static off_t get_delta_base(struct packed_git *p,
1817 struct pack_window **w_curs,
1818 off_t *curpos,
1819 enum object_type type,
1820 off_t delta_obj_offset)
1821{
1822 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1823 off_t base_offset;
1824
1825 /* use_pack() assured us we have [base_info, base_info + 20)
1826 * as a range that we can look at without walking off the
1827 * end of the mapped window. Its actually the hash size
1828 * that is assured. An OFS_DELTA longer than the hash size
1829 * is stupid, as then a REF_DELTA would be smaller to store.
1830 */
1831 if (type == OBJ_OFS_DELTA) {
1832 unsigned used = 0;
1833 unsigned char c = base_info[used++];
1834 base_offset = c & 127;
1835 while (c & 128) {
1836 base_offset += 1;
1837 if (!base_offset || MSB(base_offset, 7))
1838 return 0; /* overflow */
1839 c = base_info[used++];
1840 base_offset = (base_offset << 7) + (c & 127);
1841 }
1842 base_offset = delta_obj_offset - base_offset;
1843 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1844 return 0; /* out of bound */
1845 *curpos += used;
1846 } else if (type == OBJ_REF_DELTA) {
1847 /* The base entry _must_ be in the same pack */
1848 base_offset = find_pack_entry_one(base_info, p);
1849 *curpos += 20;
1850 } else
1851 die("I am totally screwed");
1852 return base_offset;
1853}
1854
1855/*
1856 * Like get_delta_base above, but we return the sha1 instead of the pack
1857 * offset. This means it is cheaper for REF deltas (we do not have to do
1858 * the final object lookup), but more expensive for OFS deltas (we
1859 * have to load the revidx to convert the offset back into a sha1).
1860 */
1861static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1862 struct pack_window **w_curs,
1863 off_t curpos,
1864 enum object_type type,
1865 off_t delta_obj_offset)
1866{
1867 if (type == OBJ_REF_DELTA) {
1868 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1869 return base;
1870 } else if (type == OBJ_OFS_DELTA) {
1871 struct revindex_entry *revidx;
1872 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1873 type, delta_obj_offset);
1874
1875 if (!base_offset)
1876 return NULL;
1877
1878 revidx = find_pack_revindex(p, base_offset);
1879 if (!revidx)
1880 return NULL;
1881
1882 return nth_packed_object_sha1(p, revidx->nr);
1883 } else
1884 return NULL;
1885}
1886
1887int unpack_object_header(struct packed_git *p,
1888 struct pack_window **w_curs,
1889 off_t *curpos,
1890 unsigned long *sizep)
1891{
1892 unsigned char *base;
1893 unsigned long left;
1894 unsigned long used;
1895 enum object_type type;
1896
1897 /* use_pack() assures us we have [base, base + 20) available
1898 * as a range that we can look at. (Its actually the hash
1899 * size that is assured.) With our object header encoding
1900 * the maximum deflated object size is 2^137, which is just
1901 * insane, so we know won't exceed what we have been given.
1902 */
1903 base = use_pack(p, w_curs, *curpos, &left);
1904 used = unpack_object_header_buffer(base, left, &type, sizep);
1905 if (!used) {
1906 type = OBJ_BAD;
1907 } else
1908 *curpos += used;
1909
1910 return type;
1911}
1912
1913static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1914{
1915 int type;
1916 struct revindex_entry *revidx;
1917 const unsigned char *sha1;
1918 revidx = find_pack_revindex(p, obj_offset);
1919 if (!revidx)
1920 return OBJ_BAD;
1921 sha1 = nth_packed_object_sha1(p, revidx->nr);
1922 mark_bad_packed_object(p, sha1);
1923 type = sha1_object_info(sha1, NULL);
1924 if (type <= OBJ_NONE)
1925 return OBJ_BAD;
1926 return type;
1927}
1928
1929#define POI_STACK_PREALLOC 64
1930
1931static enum object_type packed_to_object_type(struct packed_git *p,
1932 off_t obj_offset,
1933 enum object_type type,
1934 struct pack_window **w_curs,
1935 off_t curpos)
1936{
1937 off_t small_poi_stack[POI_STACK_PREALLOC];
1938 off_t *poi_stack = small_poi_stack;
1939 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1940
1941 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1942 off_t base_offset;
1943 unsigned long size;
1944 /* Push the object we're going to leave behind */
1945 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1946 poi_stack_alloc = alloc_nr(poi_stack_nr);
1947 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1948 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1949 } else {
1950 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1951 }
1952 poi_stack[poi_stack_nr++] = obj_offset;
1953 /* If parsing the base offset fails, just unwind */
1954 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1955 if (!base_offset)
1956 goto unwind;
1957 curpos = obj_offset = base_offset;
1958 type = unpack_object_header(p, w_curs, &curpos, &size);
1959 if (type <= OBJ_NONE) {
1960 /* If getting the base itself fails, we first
1961 * retry the base, otherwise unwind */
1962 type = retry_bad_packed_offset(p, base_offset);
1963 if (type > OBJ_NONE)
1964 goto out;
1965 goto unwind;
1966 }
1967 }
1968
1969 switch (type) {
1970 case OBJ_BAD:
1971 case OBJ_COMMIT:
1972 case OBJ_TREE:
1973 case OBJ_BLOB:
1974 case OBJ_TAG:
1975 break;
1976 default:
1977 error("unknown object type %i at offset %"PRIuMAX" in %s",
1978 type, (uintmax_t)obj_offset, p->pack_name);
1979 type = OBJ_BAD;
1980 }
1981
1982out:
1983 if (poi_stack != small_poi_stack)
1984 free(poi_stack);
1985 return type;
1986
1987unwind:
1988 while (poi_stack_nr) {
1989 obj_offset = poi_stack[--poi_stack_nr];
1990 type = retry_bad_packed_offset(p, obj_offset);
1991 if (type > OBJ_NONE)
1992 goto out;
1993 }
1994 type = OBJ_BAD;
1995 goto out;
1996}
1997
1998static int packed_object_info(struct packed_git *p, off_t obj_offset,
1999 struct object_info *oi)
2000{
2001 struct pack_window *w_curs = NULL;
2002 unsigned long size;
2003 off_t curpos = obj_offset;
2004 enum object_type type;
2005
2006 /*
2007 * We always get the representation type, but only convert it to
2008 * a "real" type later if the caller is interested.
2009 */
2010 type = unpack_object_header(p, &w_curs, &curpos, &size);
2011
2012 if (oi->sizep) {
2013 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2014 off_t tmp_pos = curpos;
2015 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2016 type, obj_offset);
2017 if (!base_offset) {
2018 type = OBJ_BAD;
2019 goto out;
2020 }
2021 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2022 if (*oi->sizep == 0) {
2023 type = OBJ_BAD;
2024 goto out;
2025 }
2026 } else {
2027 *oi->sizep = size;
2028 }
2029 }
2030
2031 if (oi->disk_sizep) {
2032 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2033 *oi->disk_sizep = revidx[1].offset - obj_offset;
2034 }
2035
2036 if (oi->typep) {
2037 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2038 if (*oi->typep < 0) {
2039 type = OBJ_BAD;
2040 goto out;
2041 }
2042 }
2043
2044 if (oi->delta_base_sha1) {
2045 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2046 const unsigned char *base;
2047
2048 base = get_delta_base_sha1(p, &w_curs, curpos,
2049 type, obj_offset);
2050 if (!base) {
2051 type = OBJ_BAD;
2052 goto out;
2053 }
2054
2055 hashcpy(oi->delta_base_sha1, base);
2056 } else
2057 hashclr(oi->delta_base_sha1);
2058 }
2059
2060out:
2061 unuse_pack(&w_curs);
2062 return type;
2063}
2064
2065static void *unpack_compressed_entry(struct packed_git *p,
2066 struct pack_window **w_curs,
2067 off_t curpos,
2068 unsigned long size)
2069{
2070 int st;
2071 git_zstream stream;
2072 unsigned char *buffer, *in;
2073
2074 buffer = xmallocz_gently(size);
2075 if (!buffer)
2076 return NULL;
2077 memset(&stream, 0, sizeof(stream));
2078 stream.next_out = buffer;
2079 stream.avail_out = size + 1;
2080
2081 git_inflate_init(&stream);
2082 do {
2083 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2084 stream.next_in = in;
2085 st = git_inflate(&stream, Z_FINISH);
2086 if (!stream.avail_out)
2087 break; /* the payload is larger than it should be */
2088 curpos += stream.next_in - in;
2089 } while (st == Z_OK || st == Z_BUF_ERROR);
2090 git_inflate_end(&stream);
2091 if ((st != Z_STREAM_END) || stream.total_out != size) {
2092 free(buffer);
2093 return NULL;
2094 }
2095
2096 return buffer;
2097}
2098
2099#define MAX_DELTA_CACHE (256)
2100
2101static size_t delta_base_cached;
2102
2103static struct delta_base_cache_lru_list {
2104 struct delta_base_cache_lru_list *prev;
2105 struct delta_base_cache_lru_list *next;
2106} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2107
2108static struct delta_base_cache_entry {
2109 struct delta_base_cache_lru_list lru;
2110 void *data;
2111 struct packed_git *p;
2112 off_t base_offset;
2113 unsigned long size;
2114 enum object_type type;
2115} delta_base_cache[MAX_DELTA_CACHE];
2116
2117static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2118{
2119 unsigned long hash;
2120
2121 hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
2122 hash += (hash >> 8) + (hash >> 16);
2123 return hash % MAX_DELTA_CACHE;
2124}
2125
2126static struct delta_base_cache_entry *
2127get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2128{
2129 unsigned long hash = pack_entry_hash(p, base_offset);
2130 return delta_base_cache + hash;
2131}
2132
2133static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2134 struct packed_git *p, off_t base_offset)
2135{
2136 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2137}
2138
2139static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2140{
2141 struct delta_base_cache_entry *ent;
2142 ent = get_delta_base_cache_entry(p, base_offset);
2143 return eq_delta_base_cache_entry(ent, p, base_offset);
2144}
2145
2146static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2147{
2148 ent->data = NULL;
2149 ent->lru.next->prev = ent->lru.prev;
2150 ent->lru.prev->next = ent->lru.next;
2151 delta_base_cached -= ent->size;
2152}
2153
2154static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2155 unsigned long *base_size, enum object_type *type, int keep_cache)
2156{
2157 struct delta_base_cache_entry *ent;
2158 void *ret;
2159
2160 ent = get_delta_base_cache_entry(p, base_offset);
2161
2162 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2163 return unpack_entry(p, base_offset, type, base_size);
2164
2165 ret = ent->data;
2166
2167 if (!keep_cache)
2168 clear_delta_base_cache_entry(ent);
2169 else
2170 ret = xmemdupz(ent->data, ent->size);
2171 *type = ent->type;
2172 *base_size = ent->size;
2173 return ret;
2174}
2175
2176static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2177{
2178 if (ent->data) {
2179 free(ent->data);
2180 ent->data = NULL;
2181 ent->lru.next->prev = ent->lru.prev;
2182 ent->lru.prev->next = ent->lru.next;
2183 delta_base_cached -= ent->size;
2184 }
2185}
2186
2187void clear_delta_base_cache(void)
2188{
2189 unsigned long p;
2190 for (p = 0; p < MAX_DELTA_CACHE; p++)
2191 release_delta_base_cache(&delta_base_cache[p]);
2192}
2193
2194static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2195 void *base, unsigned long base_size, enum object_type type)
2196{
2197 unsigned long hash = pack_entry_hash(p, base_offset);
2198 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2199 struct delta_base_cache_lru_list *lru;
2200
2201 release_delta_base_cache(ent);
2202 delta_base_cached += base_size;
2203
2204 for (lru = delta_base_cache_lru.next;
2205 delta_base_cached > delta_base_cache_limit
2206 && lru != &delta_base_cache_lru;
2207 lru = lru->next) {
2208 struct delta_base_cache_entry *f = (void *)lru;
2209 if (f->type == OBJ_BLOB)
2210 release_delta_base_cache(f);
2211 }
2212 for (lru = delta_base_cache_lru.next;
2213 delta_base_cached > delta_base_cache_limit
2214 && lru != &delta_base_cache_lru;
2215 lru = lru->next) {
2216 struct delta_base_cache_entry *f = (void *)lru;
2217 release_delta_base_cache(f);
2218 }
2219
2220 ent->p = p;
2221 ent->base_offset = base_offset;
2222 ent->type = type;
2223 ent->data = base;
2224 ent->size = base_size;
2225 ent->lru.next = &delta_base_cache_lru;
2226 ent->lru.prev = delta_base_cache_lru.prev;
2227 delta_base_cache_lru.prev->next = &ent->lru;
2228 delta_base_cache_lru.prev = &ent->lru;
2229}
2230
2231static void *read_object(const unsigned char *sha1, enum object_type *type,
2232 unsigned long *size);
2233
2234static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2235{
2236 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2237 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2238 p->pack_name, (uintmax_t)obj_offset);
2239}
2240
2241int do_check_packed_object_crc;
2242
2243#define UNPACK_ENTRY_STACK_PREALLOC 64
2244struct unpack_entry_stack_ent {
2245 off_t obj_offset;
2246 off_t curpos;
2247 unsigned long size;
2248};
2249
2250void *unpack_entry(struct packed_git *p, off_t obj_offset,
2251 enum object_type *final_type, unsigned long *final_size)
2252{
2253 struct pack_window *w_curs = NULL;
2254 off_t curpos = obj_offset;
2255 void *data = NULL;
2256 unsigned long size;
2257 enum object_type type;
2258 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2259 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2260 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2261 int base_from_cache = 0;
2262
2263 write_pack_access_log(p, obj_offset);
2264
2265 /* PHASE 1: drill down to the innermost base object */
2266 for (;;) {
2267 off_t base_offset;
2268 int i;
2269 struct delta_base_cache_entry *ent;
2270
2271 ent = get_delta_base_cache_entry(p, curpos);
2272 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2273 type = ent->type;
2274 data = ent->data;
2275 size = ent->size;
2276 clear_delta_base_cache_entry(ent);
2277 base_from_cache = 1;
2278 break;
2279 }
2280
2281 if (do_check_packed_object_crc && p->index_version > 1) {
2282 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2283 unsigned long len = revidx[1].offset - obj_offset;
2284 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2285 const unsigned char *sha1 =
2286 nth_packed_object_sha1(p, revidx->nr);
2287 error("bad packed object CRC for %s",
2288 sha1_to_hex(sha1));
2289 mark_bad_packed_object(p, sha1);
2290 unuse_pack(&w_curs);
2291 return NULL;
2292 }
2293 }
2294
2295 type = unpack_object_header(p, &w_curs, &curpos, &size);
2296 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2297 break;
2298
2299 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2300 if (!base_offset) {
2301 error("failed to validate delta base reference "
2302 "at offset %"PRIuMAX" from %s",
2303 (uintmax_t)curpos, p->pack_name);
2304 /* bail to phase 2, in hopes of recovery */
2305 data = NULL;
2306 break;
2307 }
2308
2309 /* push object, proceed to base */
2310 if (delta_stack_nr >= delta_stack_alloc
2311 && delta_stack == small_delta_stack) {
2312 delta_stack_alloc = alloc_nr(delta_stack_nr);
2313 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2314 memcpy(delta_stack, small_delta_stack,
2315 sizeof(*delta_stack)*delta_stack_nr);
2316 } else {
2317 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2318 }
2319 i = delta_stack_nr++;
2320 delta_stack[i].obj_offset = obj_offset;
2321 delta_stack[i].curpos = curpos;
2322 delta_stack[i].size = size;
2323
2324 curpos = obj_offset = base_offset;
2325 }
2326
2327 /* PHASE 2: handle the base */
2328 switch (type) {
2329 case OBJ_OFS_DELTA:
2330 case OBJ_REF_DELTA:
2331 if (data)
2332 die("BUG in unpack_entry: left loop at a valid delta");
2333 break;
2334 case OBJ_COMMIT:
2335 case OBJ_TREE:
2336 case OBJ_BLOB:
2337 case OBJ_TAG:
2338 if (!base_from_cache)
2339 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2340 break;
2341 default:
2342 data = NULL;
2343 error("unknown object type %i at offset %"PRIuMAX" in %s",
2344 type, (uintmax_t)obj_offset, p->pack_name);
2345 }
2346
2347 /* PHASE 3: apply deltas in order */
2348
2349 /* invariants:
2350 * 'data' holds the base data, or NULL if there was corruption
2351 */
2352 while (delta_stack_nr) {
2353 void *delta_data;
2354 void *base = data;
2355 unsigned long delta_size, base_size = size;
2356 int i;
2357
2358 data = NULL;
2359
2360 if (base)
2361 add_delta_base_cache(p, obj_offset, base, base_size, type);
2362
2363 if (!base) {
2364 /*
2365 * We're probably in deep shit, but let's try to fetch
2366 * the required base anyway from another pack or loose.
2367 * This is costly but should happen only in the presence
2368 * of a corrupted pack, and is better than failing outright.
2369 */
2370 struct revindex_entry *revidx;
2371 const unsigned char *base_sha1;
2372 revidx = find_pack_revindex(p, obj_offset);
2373 if (revidx) {
2374 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2375 error("failed to read delta base object %s"
2376 " at offset %"PRIuMAX" from %s",
2377 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2378 p->pack_name);
2379 mark_bad_packed_object(p, base_sha1);
2380 base = read_object(base_sha1, &type, &base_size);
2381 }
2382 }
2383
2384 i = --delta_stack_nr;
2385 obj_offset = delta_stack[i].obj_offset;
2386 curpos = delta_stack[i].curpos;
2387 delta_size = delta_stack[i].size;
2388
2389 if (!base)
2390 continue;
2391
2392 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2393
2394 if (!delta_data) {
2395 error("failed to unpack compressed delta "
2396 "at offset %"PRIuMAX" from %s",
2397 (uintmax_t)curpos, p->pack_name);
2398 data = NULL;
2399 continue;
2400 }
2401
2402 data = patch_delta(base, base_size,
2403 delta_data, delta_size,
2404 &size);
2405
2406 /*
2407 * We could not apply the delta; warn the user, but keep going.
2408 * Our failure will be noticed either in the next iteration of
2409 * the loop, or if this is the final delta, in the caller when
2410 * we return NULL. Those code paths will take care of making
2411 * a more explicit warning and retrying with another copy of
2412 * the object.
2413 */
2414 if (!data)
2415 error("failed to apply delta");
2416
2417 free(delta_data);
2418 }
2419
2420 *final_type = type;
2421 *final_size = size;
2422
2423 unuse_pack(&w_curs);
2424
2425 if (delta_stack != small_delta_stack)
2426 free(delta_stack);
2427
2428 return data;
2429}
2430
2431const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2432 uint32_t n)
2433{
2434 const unsigned char *index = p->index_data;
2435 if (!index) {
2436 if (open_pack_index(p))
2437 return NULL;
2438 index = p->index_data;
2439 }
2440 if (n >= p->num_objects)
2441 return NULL;
2442 index += 4 * 256;
2443 if (p->index_version == 1) {
2444 return index + 24 * n + 4;
2445 } else {
2446 index += 8;
2447 return index + 20 * n;
2448 }
2449}
2450
2451off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2452{
2453 const unsigned char *index = p->index_data;
2454 index += 4 * 256;
2455 if (p->index_version == 1) {
2456 return ntohl(*((uint32_t *)(index + 24 * n)));
2457 } else {
2458 uint32_t off;
2459 index += 8 + p->num_objects * (20 + 4);
2460 off = ntohl(*((uint32_t *)(index + 4 * n)));
2461 if (!(off & 0x80000000))
2462 return off;
2463 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2464 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2465 ntohl(*((uint32_t *)(index + 4)));
2466 }
2467}
2468
2469off_t find_pack_entry_one(const unsigned char *sha1,
2470 struct packed_git *p)
2471{
2472 const uint32_t *level1_ofs = p->index_data;
2473 const unsigned char *index = p->index_data;
2474 unsigned hi, lo, stride;
2475 static int use_lookup = -1;
2476 static int debug_lookup = -1;
2477
2478 if (debug_lookup < 0)
2479 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2480
2481 if (!index) {
2482 if (open_pack_index(p))
2483 return 0;
2484 level1_ofs = p->index_data;
2485 index = p->index_data;
2486 }
2487 if (p->index_version > 1) {
2488 level1_ofs += 2;
2489 index += 8;
2490 }
2491 index += 4 * 256;
2492 hi = ntohl(level1_ofs[*sha1]);
2493 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2494 if (p->index_version > 1) {
2495 stride = 20;
2496 } else {
2497 stride = 24;
2498 index += 4;
2499 }
2500
2501 if (debug_lookup)
2502 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2503 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2504
2505 if (use_lookup < 0)
2506 use_lookup = !!getenv("GIT_USE_LOOKUP");
2507 if (use_lookup) {
2508 int pos = sha1_entry_pos(index, stride, 0,
2509 lo, hi, p->num_objects, sha1);
2510 if (pos < 0)
2511 return 0;
2512 return nth_packed_object_offset(p, pos);
2513 }
2514
2515 do {
2516 unsigned mi = (lo + hi) / 2;
2517 int cmp = hashcmp(index + mi * stride, sha1);
2518
2519 if (debug_lookup)
2520 printf("lo %u hi %u rg %u mi %u\n",
2521 lo, hi, hi - lo, mi);
2522 if (!cmp)
2523 return nth_packed_object_offset(p, mi);
2524 if (cmp > 0)
2525 hi = mi;
2526 else
2527 lo = mi+1;
2528 } while (lo < hi);
2529 return 0;
2530}
2531
2532int is_pack_valid(struct packed_git *p)
2533{
2534 /* An already open pack is known to be valid. */
2535 if (p->pack_fd != -1)
2536 return 1;
2537
2538 /* If the pack has one window completely covering the
2539 * file size, the pack is known to be valid even if
2540 * the descriptor is not currently open.
2541 */
2542 if (p->windows) {
2543 struct pack_window *w = p->windows;
2544
2545 if (!w->offset && w->len == p->pack_size)
2546 return 1;
2547 }
2548
2549 /* Force the pack to open to prove its valid. */
2550 return !open_packed_git(p);
2551}
2552
2553static int fill_pack_entry(const unsigned char *sha1,
2554 struct pack_entry *e,
2555 struct packed_git *p)
2556{
2557 off_t offset;
2558
2559 if (p->num_bad_objects) {
2560 unsigned i;
2561 for (i = 0; i < p->num_bad_objects; i++)
2562 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2563 return 0;
2564 }
2565
2566 offset = find_pack_entry_one(sha1, p);
2567 if (!offset)
2568 return 0;
2569
2570 /*
2571 * We are about to tell the caller where they can locate the
2572 * requested object. We better make sure the packfile is
2573 * still here and can be accessed before supplying that
2574 * answer, as it may have been deleted since the index was
2575 * loaded!
2576 */
2577 if (!is_pack_valid(p))
2578 return 0;
2579 e->offset = offset;
2580 e->p = p;
2581 hashcpy(e->sha1, sha1);
2582 return 1;
2583}
2584
2585/*
2586 * Iff a pack file contains the object named by sha1, return true and
2587 * store its location to e.
2588 */
2589static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2590{
2591 struct packed_git *p;
2592
2593 prepare_packed_git();
2594 if (!packed_git)
2595 return 0;
2596
2597 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2598 return 1;
2599
2600 for (p = packed_git; p; p = p->next) {
2601 if (p == last_found_pack)
2602 continue; /* we already checked this one */
2603
2604 if (fill_pack_entry(sha1, e, p)) {
2605 last_found_pack = p;
2606 return 1;
2607 }
2608 }
2609 return 0;
2610}
2611
2612struct packed_git *find_sha1_pack(const unsigned char *sha1,
2613 struct packed_git *packs)
2614{
2615 struct packed_git *p;
2616
2617 for (p = packs; p; p = p->next) {
2618 if (find_pack_entry_one(sha1, p))
2619 return p;
2620 }
2621 return NULL;
2622
2623}
2624
2625static int sha1_loose_object_info(const unsigned char *sha1,
2626 struct object_info *oi,
2627 int flags)
2628{
2629 int status = 0;
2630 unsigned long mapsize;
2631 void *map;
2632 git_zstream stream;
2633 char hdr[32];
2634 struct strbuf hdrbuf = STRBUF_INIT;
2635
2636 if (oi->delta_base_sha1)
2637 hashclr(oi->delta_base_sha1);
2638
2639 /*
2640 * If we don't care about type or size, then we don't
2641 * need to look inside the object at all. Note that we
2642 * do not optimize out the stat call, even if the
2643 * caller doesn't care about the disk-size, since our
2644 * return value implicitly indicates whether the
2645 * object even exists.
2646 */
2647 if (!oi->typep && !oi->typename && !oi->sizep) {
2648 struct stat st;
2649 if (stat_sha1_file(sha1, &st) < 0)
2650 return -1;
2651 if (oi->disk_sizep)
2652 *oi->disk_sizep = st.st_size;
2653 return 0;
2654 }
2655
2656 map = map_sha1_file(sha1, &mapsize);
2657 if (!map)
2658 return -1;
2659 if (oi->disk_sizep)
2660 *oi->disk_sizep = mapsize;
2661 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2662 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2663 status = error("unable to unpack %s header with --allow-unknown-type",
2664 sha1_to_hex(sha1));
2665 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2666 status = error("unable to unpack %s header",
2667 sha1_to_hex(sha1));
2668 if (status < 0)
2669 ; /* Do nothing */
2670 else if (hdrbuf.len) {
2671 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2672 status = error("unable to parse %s header with --allow-unknown-type",
2673 sha1_to_hex(sha1));
2674 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2675 status = error("unable to parse %s header", sha1_to_hex(sha1));
2676 git_inflate_end(&stream);
2677 munmap(map, mapsize);
2678 if (status && oi->typep)
2679 *oi->typep = status;
2680 strbuf_release(&hdrbuf);
2681 return 0;
2682}
2683
2684int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2685{
2686 struct cached_object *co;
2687 struct pack_entry e;
2688 int rtype;
2689 enum object_type real_type;
2690 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2691
2692 co = find_cached_object(real);
2693 if (co) {
2694 if (oi->typep)
2695 *(oi->typep) = co->type;
2696 if (oi->sizep)
2697 *(oi->sizep) = co->size;
2698 if (oi->disk_sizep)
2699 *(oi->disk_sizep) = 0;
2700 if (oi->delta_base_sha1)
2701 hashclr(oi->delta_base_sha1);
2702 if (oi->typename)
2703 strbuf_addstr(oi->typename, typename(co->type));
2704 oi->whence = OI_CACHED;
2705 return 0;
2706 }
2707
2708 if (!find_pack_entry(real, &e)) {
2709 /* Most likely it's a loose object. */
2710 if (!sha1_loose_object_info(real, oi, flags)) {
2711 oi->whence = OI_LOOSE;
2712 return 0;
2713 }
2714
2715 /* Not a loose object; someone else may have just packed it. */
2716 reprepare_packed_git();
2717 if (!find_pack_entry(real, &e))
2718 return -1;
2719 }
2720
2721 /*
2722 * packed_object_info() does not follow the delta chain to
2723 * find out the real type, unless it is given oi->typep.
2724 */
2725 if (oi->typename && !oi->typep)
2726 oi->typep = &real_type;
2727
2728 rtype = packed_object_info(e.p, e.offset, oi);
2729 if (rtype < 0) {
2730 mark_bad_packed_object(e.p, real);
2731 if (oi->typep == &real_type)
2732 oi->typep = NULL;
2733 return sha1_object_info_extended(real, oi, 0);
2734 } else if (in_delta_base_cache(e.p, e.offset)) {
2735 oi->whence = OI_DBCACHED;
2736 } else {
2737 oi->whence = OI_PACKED;
2738 oi->u.packed.offset = e.offset;
2739 oi->u.packed.pack = e.p;
2740 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2741 rtype == OBJ_OFS_DELTA);
2742 }
2743 if (oi->typename)
2744 strbuf_addstr(oi->typename, typename(*oi->typep));
2745 if (oi->typep == &real_type)
2746 oi->typep = NULL;
2747
2748 return 0;
2749}
2750
2751/* returns enum object_type or negative */
2752int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2753{
2754 enum object_type type;
2755 struct object_info oi = {NULL};
2756
2757 oi.typep = &type;
2758 oi.sizep = sizep;
2759 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2760 return -1;
2761 return type;
2762}
2763
2764static void *read_packed_sha1(const unsigned char *sha1,
2765 enum object_type *type, unsigned long *size)
2766{
2767 struct pack_entry e;
2768 void *data;
2769
2770 if (!find_pack_entry(sha1, &e))
2771 return NULL;
2772 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2773 if (!data) {
2774 /*
2775 * We're probably in deep shit, but let's try to fetch
2776 * the required object anyway from another pack or loose.
2777 * This should happen only in the presence of a corrupted
2778 * pack, and is better than failing outright.
2779 */
2780 error("failed to read object %s at offset %"PRIuMAX" from %s",
2781 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2782 mark_bad_packed_object(e.p, sha1);
2783 data = read_object(sha1, type, size);
2784 }
2785 return data;
2786}
2787
2788int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2789 unsigned char *sha1)
2790{
2791 struct cached_object *co;
2792
2793 hash_sha1_file(buf, len, typename(type), sha1);
2794 if (has_sha1_file(sha1) || find_cached_object(sha1))
2795 return 0;
2796 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2797 co = &cached_objects[cached_object_nr++];
2798 co->size = len;
2799 co->type = type;
2800 co->buf = xmalloc(len);
2801 memcpy(co->buf, buf, len);
2802 hashcpy(co->sha1, sha1);
2803 return 0;
2804}
2805
2806static void *read_object(const unsigned char *sha1, enum object_type *type,
2807 unsigned long *size)
2808{
2809 unsigned long mapsize;
2810 void *map, *buf;
2811 struct cached_object *co;
2812
2813 co = find_cached_object(sha1);
2814 if (co) {
2815 *type = co->type;
2816 *size = co->size;
2817 return xmemdupz(co->buf, co->size);
2818 }
2819
2820 buf = read_packed_sha1(sha1, type, size);
2821 if (buf)
2822 return buf;
2823 map = map_sha1_file(sha1, &mapsize);
2824 if (map) {
2825 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2826 munmap(map, mapsize);
2827 return buf;
2828 }
2829 reprepare_packed_git();
2830 return read_packed_sha1(sha1, type, size);
2831}
2832
2833/*
2834 * This function dies on corrupt objects; the callers who want to
2835 * deal with them should arrange to call read_object() and give error
2836 * messages themselves.
2837 */
2838void *read_sha1_file_extended(const unsigned char *sha1,
2839 enum object_type *type,
2840 unsigned long *size,
2841 unsigned flag)
2842{
2843 void *data;
2844 const struct packed_git *p;
2845 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2846
2847 errno = 0;
2848 data = read_object(repl, type, size);
2849 if (data)
2850 return data;
2851
2852 if (errno && errno != ENOENT)
2853 die_errno("failed to read object %s", sha1_to_hex(sha1));
2854
2855 /* die if we replaced an object with one that does not exist */
2856 if (repl != sha1)
2857 die("replacement %s not found for %s",
2858 sha1_to_hex(repl), sha1_to_hex(sha1));
2859
2860 if (has_loose_object(repl)) {
2861 const char *path = sha1_file_name(sha1);
2862
2863 die("loose object %s (stored in %s) is corrupt",
2864 sha1_to_hex(repl), path);
2865 }
2866
2867 if ((p = has_packed_and_bad(repl)) != NULL)
2868 die("packed object %s (stored in %s) is corrupt",
2869 sha1_to_hex(repl), p->pack_name);
2870
2871 return NULL;
2872}
2873
2874void *read_object_with_reference(const unsigned char *sha1,
2875 const char *required_type_name,
2876 unsigned long *size,
2877 unsigned char *actual_sha1_return)
2878{
2879 enum object_type type, required_type;
2880 void *buffer;
2881 unsigned long isize;
2882 unsigned char actual_sha1[20];
2883
2884 required_type = type_from_string(required_type_name);
2885 hashcpy(actual_sha1, sha1);
2886 while (1) {
2887 int ref_length = -1;
2888 const char *ref_type = NULL;
2889
2890 buffer = read_sha1_file(actual_sha1, &type, &isize);
2891 if (!buffer)
2892 return NULL;
2893 if (type == required_type) {
2894 *size = isize;
2895 if (actual_sha1_return)
2896 hashcpy(actual_sha1_return, actual_sha1);
2897 return buffer;
2898 }
2899 /* Handle references */
2900 else if (type == OBJ_COMMIT)
2901 ref_type = "tree ";
2902 else if (type == OBJ_TAG)
2903 ref_type = "object ";
2904 else {
2905 free(buffer);
2906 return NULL;
2907 }
2908 ref_length = strlen(ref_type);
2909
2910 if (ref_length + 40 > isize ||
2911 memcmp(buffer, ref_type, ref_length) ||
2912 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2913 free(buffer);
2914 return NULL;
2915 }
2916 free(buffer);
2917 /* Now we have the ID of the referred-to object in
2918 * actual_sha1. Check again. */
2919 }
2920}
2921
2922static void write_sha1_file_prepare(const void *buf, unsigned long len,
2923 const char *type, unsigned char *sha1,
2924 char *hdr, int *hdrlen)
2925{
2926 git_SHA_CTX c;
2927
2928 /* Generate the header */
2929 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2930
2931 /* Sha1.. */
2932 git_SHA1_Init(&c);
2933 git_SHA1_Update(&c, hdr, *hdrlen);
2934 git_SHA1_Update(&c, buf, len);
2935 git_SHA1_Final(sha1, &c);
2936}
2937
2938/*
2939 * Move the just written object into its final resting place.
2940 */
2941int finalize_object_file(const char *tmpfile, const char *filename)
2942{
2943 int ret = 0;
2944
2945 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2946 goto try_rename;
2947 else if (link(tmpfile, filename))
2948 ret = errno;
2949
2950 /*
2951 * Coda hack - coda doesn't like cross-directory links,
2952 * so we fall back to a rename, which will mean that it
2953 * won't be able to check collisions, but that's not a
2954 * big deal.
2955 *
2956 * The same holds for FAT formatted media.
2957 *
2958 * When this succeeds, we just return. We have nothing
2959 * left to unlink.
2960 */
2961 if (ret && ret != EEXIST) {
2962 try_rename:
2963 if (!rename(tmpfile, filename))
2964 goto out;
2965 ret = errno;
2966 }
2967 unlink_or_warn(tmpfile);
2968 if (ret) {
2969 if (ret != EEXIST) {
2970 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2971 }
2972 /* FIXME!!! Collision check here ? */
2973 }
2974
2975out:
2976 if (adjust_shared_perm(filename))
2977 return error("unable to set permission to '%s'", filename);
2978 return 0;
2979}
2980
2981static int write_buffer(int fd, const void *buf, size_t len)
2982{
2983 if (write_in_full(fd, buf, len) < 0)
2984 return error("file write error (%s)", strerror(errno));
2985 return 0;
2986}
2987
2988int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2989 unsigned char *sha1)
2990{
2991 char hdr[32];
2992 int hdrlen = sizeof(hdr);
2993 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2994 return 0;
2995}
2996
2997/* Finalize a file on disk, and close it. */
2998static void close_sha1_file(int fd)
2999{
3000 if (fsync_object_files)
3001 fsync_or_die(fd, "sha1 file");
3002 if (close(fd) != 0)
3003 die_errno("error when closing sha1 file");
3004}
3005
3006/* Size of directory component, including the ending '/' */
3007static inline int directory_size(const char *filename)
3008{
3009 const char *s = strrchr(filename, '/');
3010 if (!s)
3011 return 0;
3012 return s - filename + 1;
3013}
3014
3015/*
3016 * This creates a temporary file in the same directory as the final
3017 * 'filename'
3018 *
3019 * We want to avoid cross-directory filename renames, because those
3020 * can have problems on various filesystems (FAT, NFS, Coda).
3021 */
3022static int create_tmpfile(struct strbuf *tmp, const char *filename)
3023{
3024 int fd, dirlen = directory_size(filename);
3025
3026 strbuf_reset(tmp);
3027 strbuf_add(tmp, filename, dirlen);
3028 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3029 fd = git_mkstemp_mode(tmp->buf, 0444);
3030 if (fd < 0 && dirlen && errno == ENOENT) {
3031 /*
3032 * Make sure the directory exists; note that the contents
3033 * of the buffer are undefined after mkstemp returns an
3034 * error, so we have to rewrite the whole buffer from
3035 * scratch.
3036 */
3037 strbuf_reset(tmp);
3038 strbuf_add(tmp, filename, dirlen - 1);
3039 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3040 return -1;
3041 if (adjust_shared_perm(tmp->buf))
3042 return -1;
3043
3044 /* Try again */
3045 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3046 fd = git_mkstemp_mode(tmp->buf, 0444);
3047 }
3048 return fd;
3049}
3050
3051static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3052 const void *buf, unsigned long len, time_t mtime)
3053{
3054 int fd, ret;
3055 unsigned char compressed[4096];
3056 git_zstream stream;
3057 git_SHA_CTX c;
3058 unsigned char parano_sha1[20];
3059 static struct strbuf tmp_file = STRBUF_INIT;
3060 const char *filename = sha1_file_name(sha1);
3061
3062 fd = create_tmpfile(&tmp_file, filename);
3063 if (fd < 0) {
3064 if (errno == EACCES)
3065 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3066 else
3067 return error("unable to create temporary file: %s", strerror(errno));
3068 }
3069
3070 /* Set it up */
3071 git_deflate_init(&stream, zlib_compression_level);
3072 stream.next_out = compressed;
3073 stream.avail_out = sizeof(compressed);
3074 git_SHA1_Init(&c);
3075
3076 /* First header.. */
3077 stream.next_in = (unsigned char *)hdr;
3078 stream.avail_in = hdrlen;
3079 while (git_deflate(&stream, 0) == Z_OK)
3080 ; /* nothing */
3081 git_SHA1_Update(&c, hdr, hdrlen);
3082
3083 /* Then the data itself.. */
3084 stream.next_in = (void *)buf;
3085 stream.avail_in = len;
3086 do {
3087 unsigned char *in0 = stream.next_in;
3088 ret = git_deflate(&stream, Z_FINISH);
3089 git_SHA1_Update(&c, in0, stream.next_in - in0);
3090 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3091 die("unable to write sha1 file");
3092 stream.next_out = compressed;
3093 stream.avail_out = sizeof(compressed);
3094 } while (ret == Z_OK);
3095
3096 if (ret != Z_STREAM_END)
3097 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3098 ret = git_deflate_end_gently(&stream);
3099 if (ret != Z_OK)
3100 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3101 git_SHA1_Final(parano_sha1, &c);
3102 if (hashcmp(sha1, parano_sha1) != 0)
3103 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3104
3105 close_sha1_file(fd);
3106
3107 if (mtime) {
3108 struct utimbuf utb;
3109 utb.actime = mtime;
3110 utb.modtime = mtime;
3111 if (utime(tmp_file.buf, &utb) < 0)
3112 warning("failed utime() on %s: %s",
3113 tmp_file.buf, strerror(errno));
3114 }
3115
3116 return finalize_object_file(tmp_file.buf, filename);
3117}
3118
3119static int freshen_loose_object(const unsigned char *sha1)
3120{
3121 return check_and_freshen(sha1, 1);
3122}
3123
3124static int freshen_packed_object(const unsigned char *sha1)
3125{
3126 struct pack_entry e;
3127 if (!find_pack_entry(sha1, &e))
3128 return 0;
3129 if (e.p->freshened)
3130 return 1;
3131 if (!freshen_file(e.p->pack_name))
3132 return 0;
3133 e.p->freshened = 1;
3134 return 1;
3135}
3136
3137int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3138{
3139 char hdr[32];
3140 int hdrlen = sizeof(hdr);
3141
3142 /* Normally if we have it in the pack then we do not bother writing
3143 * it out into .git/objects/??/?{38} file.
3144 */
3145 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3146 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3147 return 0;
3148 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3149}
3150
3151int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3152 unsigned char *sha1, unsigned flags)
3153{
3154 char *header;
3155 int hdrlen, status = 0;
3156
3157 /* type string, SP, %lu of the length plus NUL must fit this */
3158 hdrlen = strlen(type) + 32;
3159 header = xmalloc(hdrlen);
3160 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3161
3162 if (!(flags & HASH_WRITE_OBJECT))
3163 goto cleanup;
3164 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3165 goto cleanup;
3166 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3167
3168cleanup:
3169 free(header);
3170 return status;
3171}
3172
3173int force_object_loose(const unsigned char *sha1, time_t mtime)
3174{
3175 void *buf;
3176 unsigned long len;
3177 enum object_type type;
3178 char hdr[32];
3179 int hdrlen;
3180 int ret;
3181
3182 if (has_loose_object(sha1))
3183 return 0;
3184 buf = read_packed_sha1(sha1, &type, &len);
3185 if (!buf)
3186 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3187 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3188 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3189 free(buf);
3190
3191 return ret;
3192}
3193
3194int has_pack_index(const unsigned char *sha1)
3195{
3196 struct stat st;
3197 if (stat(sha1_pack_index_name(sha1), &st))
3198 return 0;
3199 return 1;
3200}
3201
3202int has_sha1_pack(const unsigned char *sha1)
3203{
3204 struct pack_entry e;
3205 return find_pack_entry(sha1, &e);
3206}
3207
3208int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3209{
3210 struct pack_entry e;
3211
3212 if (find_pack_entry(sha1, &e))
3213 return 1;
3214 if (has_loose_object(sha1))
3215 return 1;
3216 if (flags & HAS_SHA1_QUICK)
3217 return 0;
3218 reprepare_packed_git();
3219 return find_pack_entry(sha1, &e);
3220}
3221
3222int has_object_file(const struct object_id *oid)
3223{
3224 return has_sha1_file(oid->hash);
3225}
3226
3227static void check_tree(const void *buf, size_t size)
3228{
3229 struct tree_desc desc;
3230 struct name_entry entry;
3231
3232 init_tree_desc(&desc, buf, size);
3233 while (tree_entry(&desc, &entry))
3234 /* do nothing
3235 * tree_entry() will die() on malformed entries */
3236 ;
3237}
3238
3239static void check_commit(const void *buf, size_t size)
3240{
3241 struct commit c;
3242 memset(&c, 0, sizeof(c));
3243 if (parse_commit_buffer(&c, buf, size))
3244 die("corrupt commit");
3245}
3246
3247static void check_tag(const void *buf, size_t size)
3248{
3249 struct tag t;
3250 memset(&t, 0, sizeof(t));
3251 if (parse_tag_buffer(&t, buf, size))
3252 die("corrupt tag");
3253}
3254
3255static int index_mem(unsigned char *sha1, void *buf, size_t size,
3256 enum object_type type,
3257 const char *path, unsigned flags)
3258{
3259 int ret, re_allocated = 0;
3260 int write_object = flags & HASH_WRITE_OBJECT;
3261
3262 if (!type)
3263 type = OBJ_BLOB;
3264
3265 /*
3266 * Convert blobs to git internal format
3267 */
3268 if ((type == OBJ_BLOB) && path) {
3269 struct strbuf nbuf = STRBUF_INIT;
3270 if (convert_to_git(path, buf, size, &nbuf,
3271 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3272 buf = strbuf_detach(&nbuf, &size);
3273 re_allocated = 1;
3274 }
3275 }
3276 if (flags & HASH_FORMAT_CHECK) {
3277 if (type == OBJ_TREE)
3278 check_tree(buf, size);
3279 if (type == OBJ_COMMIT)
3280 check_commit(buf, size);
3281 if (type == OBJ_TAG)
3282 check_tag(buf, size);
3283 }
3284
3285 if (write_object)
3286 ret = write_sha1_file(buf, size, typename(type), sha1);
3287 else
3288 ret = hash_sha1_file(buf, size, typename(type), sha1);
3289 if (re_allocated)
3290 free(buf);
3291 return ret;
3292}
3293
3294static int index_stream_convert_blob(unsigned char *sha1, int fd,
3295 const char *path, unsigned flags)
3296{
3297 int ret;
3298 const int write_object = flags & HASH_WRITE_OBJECT;
3299 struct strbuf sbuf = STRBUF_INIT;
3300
3301 assert(path);
3302 assert(would_convert_to_git_filter_fd(path));
3303
3304 convert_to_git_filter_fd(path, fd, &sbuf,
3305 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3306
3307 if (write_object)
3308 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3309 sha1);
3310 else
3311 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3312 sha1);
3313 strbuf_release(&sbuf);
3314 return ret;
3315}
3316
3317static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3318 const char *path, unsigned flags)
3319{
3320 struct strbuf sbuf = STRBUF_INIT;
3321 int ret;
3322
3323 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3324 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3325 else
3326 ret = -1;
3327 strbuf_release(&sbuf);
3328 return ret;
3329}
3330
3331#define SMALL_FILE_SIZE (32*1024)
3332
3333static int index_core(unsigned char *sha1, int fd, size_t size,
3334 enum object_type type, const char *path,
3335 unsigned flags)
3336{
3337 int ret;
3338
3339 if (!size) {
3340 ret = index_mem(sha1, "", size, type, path, flags);
3341 } else if (size <= SMALL_FILE_SIZE) {
3342 char *buf = xmalloc(size);
3343 if (size == read_in_full(fd, buf, size))
3344 ret = index_mem(sha1, buf, size, type, path, flags);
3345 else
3346 ret = error("short read %s", strerror(errno));
3347 free(buf);
3348 } else {
3349 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3350 ret = index_mem(sha1, buf, size, type, path, flags);
3351 munmap(buf, size);
3352 }
3353 return ret;
3354}
3355
3356/*
3357 * This creates one packfile per large blob unless bulk-checkin
3358 * machinery is "plugged".
3359 *
3360 * This also bypasses the usual "convert-to-git" dance, and that is on
3361 * purpose. We could write a streaming version of the converting
3362 * functions and insert that before feeding the data to fast-import
3363 * (or equivalent in-core API described above). However, that is
3364 * somewhat complicated, as we do not know the size of the filter
3365 * result, which we need to know beforehand when writing a git object.
3366 * Since the primary motivation for trying to stream from the working
3367 * tree file and to avoid mmaping it in core is to deal with large
3368 * binary blobs, they generally do not want to get any conversion, and
3369 * callers should avoid this code path when filters are requested.
3370 */
3371static int index_stream(unsigned char *sha1, int fd, size_t size,
3372 enum object_type type, const char *path,
3373 unsigned flags)
3374{
3375 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3376}
3377
3378int index_fd(unsigned char *sha1, int fd, struct stat *st,
3379 enum object_type type, const char *path, unsigned flags)
3380{
3381 int ret;
3382
3383 /*
3384 * Call xsize_t() only when needed to avoid potentially unnecessary
3385 * die() for large files.
3386 */
3387 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3388 ret = index_stream_convert_blob(sha1, fd, path, flags);
3389 else if (!S_ISREG(st->st_mode))
3390 ret = index_pipe(sha1, fd, type, path, flags);
3391 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3392 (path && would_convert_to_git(path)))
3393 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3394 flags);
3395 else
3396 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3397 flags);
3398 close(fd);
3399 return ret;
3400}
3401
3402int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3403{
3404 int fd;
3405 struct strbuf sb = STRBUF_INIT;
3406
3407 switch (st->st_mode & S_IFMT) {
3408 case S_IFREG:
3409 fd = open(path, O_RDONLY);
3410 if (fd < 0)
3411 return error("open(\"%s\"): %s", path,
3412 strerror(errno));
3413 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3414 return error("%s: failed to insert into database",
3415 path);
3416 break;
3417 case S_IFLNK:
3418 if (strbuf_readlink(&sb, path, st->st_size)) {
3419 char *errstr = strerror(errno);
3420 return error("readlink(\"%s\"): %s", path,
3421 errstr);
3422 }
3423 if (!(flags & HASH_WRITE_OBJECT))
3424 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3425 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3426 return error("%s: failed to insert into database",
3427 path);
3428 strbuf_release(&sb);
3429 break;
3430 case S_IFDIR:
3431 return resolve_gitlink_ref(path, "HEAD", sha1);
3432 default:
3433 return error("%s: unsupported file type", path);
3434 }
3435 return 0;
3436}
3437
3438int read_pack_header(int fd, struct pack_header *header)
3439{
3440 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3441 /* "eof before pack header was fully read" */
3442 return PH_ERROR_EOF;
3443
3444 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3445 /* "protocol error (pack signature mismatch detected)" */
3446 return PH_ERROR_PACK_SIGNATURE;
3447 if (!pack_version_ok(header->hdr_version))
3448 /* "protocol error (pack version unsupported)" */
3449 return PH_ERROR_PROTOCOL;
3450 return 0;
3451}
3452
3453void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3454{
3455 enum object_type type = sha1_object_info(sha1, NULL);
3456 if (type < 0)
3457 die("%s is not a valid object", sha1_to_hex(sha1));
3458 if (type != expect)
3459 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3460 typename(expect));
3461}
3462
3463static int for_each_file_in_obj_subdir(int subdir_nr,
3464 struct strbuf *path,
3465 each_loose_object_fn obj_cb,
3466 each_loose_cruft_fn cruft_cb,
3467 each_loose_subdir_fn subdir_cb,
3468 void *data)
3469{
3470 size_t baselen = path->len;
3471 DIR *dir = opendir(path->buf);
3472 struct dirent *de;
3473 int r = 0;
3474
3475 if (!dir) {
3476 if (errno == ENOENT)
3477 return 0;
3478 return error("unable to open %s: %s", path->buf, strerror(errno));
3479 }
3480
3481 while ((de = readdir(dir))) {
3482 if (is_dot_or_dotdot(de->d_name))
3483 continue;
3484
3485 strbuf_setlen(path, baselen);
3486 strbuf_addf(path, "/%s", de->d_name);
3487
3488 if (strlen(de->d_name) == 38) {
3489 char hex[41];
3490 unsigned char sha1[20];
3491
3492 snprintf(hex, sizeof(hex), "%02x%s",
3493 subdir_nr, de->d_name);
3494 if (!get_sha1_hex(hex, sha1)) {
3495 if (obj_cb) {
3496 r = obj_cb(sha1, path->buf, data);
3497 if (r)
3498 break;
3499 }
3500 continue;
3501 }
3502 }
3503
3504 if (cruft_cb) {
3505 r = cruft_cb(de->d_name, path->buf, data);
3506 if (r)
3507 break;
3508 }
3509 }
3510 closedir(dir);
3511
3512 strbuf_setlen(path, baselen);
3513 if (!r && subdir_cb)
3514 r = subdir_cb(subdir_nr, path->buf, data);
3515
3516 return r;
3517}
3518
3519int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3520 each_loose_object_fn obj_cb,
3521 each_loose_cruft_fn cruft_cb,
3522 each_loose_subdir_fn subdir_cb,
3523 void *data)
3524{
3525 size_t baselen = path->len;
3526 int r = 0;
3527 int i;
3528
3529 for (i = 0; i < 256; i++) {
3530 strbuf_addf(path, "/%02x", i);
3531 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3532 subdir_cb, data);
3533 strbuf_setlen(path, baselen);
3534 if (r)
3535 break;
3536 }
3537
3538 return r;
3539}
3540
3541int for_each_loose_file_in_objdir(const char *path,
3542 each_loose_object_fn obj_cb,
3543 each_loose_cruft_fn cruft_cb,
3544 each_loose_subdir_fn subdir_cb,
3545 void *data)
3546{
3547 struct strbuf buf = STRBUF_INIT;
3548 int r;
3549
3550 strbuf_addstr(&buf, path);
3551 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3552 subdir_cb, data);
3553 strbuf_release(&buf);
3554
3555 return r;
3556}
3557
3558struct loose_alt_odb_data {
3559 each_loose_object_fn *cb;
3560 void *data;
3561};
3562
3563static int loose_from_alt_odb(struct alternate_object_database *alt,
3564 void *vdata)
3565{
3566 struct loose_alt_odb_data *data = vdata;
3567 struct strbuf buf = STRBUF_INIT;
3568 int r;
3569
3570 /* copy base not including trailing '/' */
3571 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3572 r = for_each_loose_file_in_objdir_buf(&buf,
3573 data->cb, NULL, NULL,
3574 data->data);
3575 strbuf_release(&buf);
3576 return r;
3577}
3578
3579int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3580{
3581 struct loose_alt_odb_data alt;
3582 int r;
3583
3584 r = for_each_loose_file_in_objdir(get_object_directory(),
3585 cb, NULL, NULL, data);
3586 if (r)
3587 return r;
3588
3589 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3590 return 0;
3591
3592 alt.cb = cb;
3593 alt.data = data;
3594 return foreach_alt_odb(loose_from_alt_odb, &alt);
3595}
3596
3597static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3598{
3599 uint32_t i;
3600 int r = 0;
3601
3602 for (i = 0; i < p->num_objects; i++) {
3603 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3604
3605 if (!sha1)
3606 return error("unable to get sha1 of object %u in %s",
3607 i, p->pack_name);
3608
3609 r = cb(sha1, p, i, data);
3610 if (r)
3611 break;
3612 }
3613 return r;
3614}
3615
3616int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3617{
3618 struct packed_git *p;
3619 int r = 0;
3620 int pack_errors = 0;
3621
3622 prepare_packed_git();
3623 for (p = packed_git; p; p = p->next) {
3624 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3625 continue;
3626 if (open_pack_index(p)) {
3627 pack_errors = 1;
3628 continue;
3629 }
3630 r = for_each_object_in_pack(p, cb, data);
3631 if (r)
3632 break;
3633 }
3634 return r ? r : pack_errors;
3635}