?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Sync with 2.33.8
[git/debian.git] / packfile.c
blob89402cfc69cd0f03b268f26bd6c8c7ad459b3e1b
1 #include "cache.h"
2 #include "list.h"
3 #include "pack.h"
4 #include "repository.h"
5 #include "dir.h"
6 #include "mergesort.h"
7 #include "packfile.h"
8 #include "delta.h"
9 #include "streaming.h"
10 #include "hash-lookup.h"
11 #include "commit.h"
12 #include "object.h"
13 #include "tag.h"
14 #include "tree-walk.h"
15 #include "tree.h"
16 #include "object-store.h"
17 #include "midx.h"
18 #include "commit-graph.h"
19 #include "promisor-remote.h"
20
21 char *odb_pack_name(struct strbuf *buf,
22 const unsigned char *hash,
23 const char *ext)
24 {
25 strbuf_reset(buf);
26 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27 hash_to_hex(hash), ext);
28 return buf->buf;
29 }
30
31 char *sha1_pack_name(const unsigned char *sha1)
32 {
33 static struct strbuf buf = STRBUF_INIT;
34 return odb_pack_name(&buf, sha1, "pack");
35 }
36
37 char *sha1_pack_index_name(const unsigned char *sha1)
38 {
39 static struct strbuf buf = STRBUF_INIT;
40 return odb_pack_name(&buf, sha1, "idx");
41 }
42
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
51
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
54
55 void pack_report(void)
56 {
57 fprintf(stderr,
58 "pack_report: getpagesize() = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size),
63 sz_fmt(packed_git_limit));
64 fprintf(stderr,
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT " / %10" SZ_FMT "\n",
70 pack_used_ctr,
71 pack_mmap_calls,
72 pack_open_windows, peak_pack_open_windows,
73 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
74 }
75
76 /*
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
79 * success.
80 */
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
82 {
83 void *idx_map;
84 size_t idx_size;
85 int fd = git_open(path), ret;
86 struct stat st;
87 const unsigned int hashsz = the_hash_algo->rawsz;
88
89 if (fd < 0)
90 return -1;
91 if (fstat(fd, &st)) {
92 close(fd);
93 return -1;
94 }
95 idx_size = xsize_t(st.st_size);
96 if (idx_size < 4 * 256 + hashsz + hashsz) {
97 close(fd);
98 return error("index file %s is too small", path);
99 }
100 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
101 close(fd);
102
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
104
105 if (ret)
106 munmap(idx_map, idx_size);
107
108 return ret;
109 }
110
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112 size_t idx_size, struct packed_git *p)
113 {
114 struct pack_idx_header *hdr = idx_map;
115 uint32_t version, nr, i, *index;
116
117 if (idx_size < 4 * 256 + hashsz + hashsz)
118 return error("index file %s is too small", path);
119 if (idx_map == NULL)
120 return error("empty data");
121
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123 version = ntohl(hdr->idx_version);
124 if (version < 2 || version > 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
128 path, version);
129 } else
130 version = 1;
131
132 nr = 0;
133 index = idx_map;
134 if (version > 1)
135 index += 2; /* skip index header */
136 for (i = 0; i < 256; i++) {
137 uint32_t n = ntohl(index[i]);
138 if (n < nr)
139 return error("non-monotonic index %s", path);
140 nr = n;
141 }
142
143 if (version == 1) {
144 /*
145 * Total size:
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
149 * - file checksum
150 */
151 if (idx_size != st_add(4 * 256 + hashsz + hashsz, st_mult(nr, hashsz + 4)))
152 return error("wrong index v1 file size in %s", path);
153 } else if (version == 2) {
154 /*
155 * Minimum size:
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
162 * - file checksum
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
166 */
167 size_t min_size = st_add(8 + 4*256 + hashsz + hashsz, st_mult(nr, hashsz + 4 + 4));
168 size_t max_size = min_size;
169 if (nr)
170 max_size = st_add(max_size, st_mult(nr - 1, 8));
171 if (idx_size < min_size || idx_size > max_size)
172 return error("wrong index v2 file size in %s", path);
173 if (idx_size != min_size &&
174 /*
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
178 */
179 (sizeof(off_t) <= 4))
180 return error("pack too large for current definition of off_t in %s", path);
181 p->crc_offset = 8 + 4 * 256 + nr * hashsz;
182 }
183
184 p->index_version = version;
185 p->index_data = idx_map;
186 p->index_size = idx_size;
187 p->num_objects = nr;
188 return 0;
189 }
190
191 int open_pack_index(struct packed_git *p)
192 {
193 char *idx_name;
194 size_t len;
195 int ret;
196
197 if (p->index_data)
198 return 0;
199
200 if (!strip_suffix(p->pack_name, ".pack", &len))
201 BUG("pack_name does not end in .pack");
202 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
203 ret = check_packed_git_idx(idx_name, p);
204 free(idx_name);
205 return ret;
206 }
207
208 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
209 {
210 const uint32_t *level1_ofs = p->index_data;
211
212 if (!level1_ofs) {
213 if (open_pack_index(p))
214 return 0;
215 level1_ofs = p->index_data;
216 }
217
218 if (p->index_version > 1) {
219 level1_ofs += 2;
220 }
221
222 return ntohl(level1_ofs[value]);
223 }
224
225 static struct packed_git *alloc_packed_git(int extra)
226 {
227 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
228 memset(p, 0, sizeof(*p));
229 p->pack_fd = -1;
230 return p;
231 }
232
233 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
234 {
235 const char *path = sha1_pack_name(sha1);
236 size_t alloc = st_add(strlen(path), 1);
237 struct packed_git *p = alloc_packed_git(alloc);
238
239 memcpy(p->pack_name, path, alloc); /* includes NUL */
240 hashcpy(p->hash, sha1);
241 if (check_packed_git_idx(idx_path, p)) {
242 free(p);
243 return NULL;
244 }
245
246 return p;
247 }
248
249 static void scan_windows(struct packed_git *p,
250 struct packed_git **lru_p,
251 struct pack_window **lru_w,
252 struct pack_window **lru_l)
253 {
254 struct pack_window *w, *w_l;
255
256 for (w_l = NULL, w = p->windows; w; w = w->next) {
257 if (!w->inuse_cnt) {
258 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
259 *lru_p = p;
260 *lru_w = w;
261 *lru_l = w_l;
262 }
263 }
264 w_l = w;
265 }
266 }
267
268 static int unuse_one_window(struct packed_git *current)
269 {
270 struct packed_git *p, *lru_p = NULL;
271 struct pack_window *lru_w = NULL, *lru_l = NULL;
272
273 if (current)
274 scan_windows(current, &lru_p, &lru_w, &lru_l);
275 for (p = the_repository->objects->packed_git; p; p = p->next)
276 scan_windows(p, &lru_p, &lru_w, &lru_l);
277 if (lru_p) {
278 munmap(lru_w->base, lru_w->len);
279 pack_mapped -= lru_w->len;
280 if (lru_l)
281 lru_l->next = lru_w->next;
282 else
283 lru_p->windows = lru_w->next;
284 free(lru_w);
285 pack_open_windows--;
286 return 1;
287 }
288 return 0;
289 }
290
291 void close_pack_windows(struct packed_git *p)
292 {
293 while (p->windows) {
294 struct pack_window *w = p->windows;
295
296 if (w->inuse_cnt)
297 die("pack '%s' still has open windows to it",
298 p->pack_name);
299 munmap(w->base, w->len);
300 pack_mapped -= w->len;
301 pack_open_windows--;
302 p->windows = w->next;
303 free(w);
304 }
305 }
306
307 int close_pack_fd(struct packed_git *p)
308 {
309 if (p->pack_fd < 0)
310 return 0;
311
312 close(p->pack_fd);
313 pack_open_fds--;
314 p->pack_fd = -1;
315
316 return 1;
317 }
318
319 void close_pack_index(struct packed_git *p)
320 {
321 if (p->index_data) {
322 munmap((void *)p->index_data, p->index_size);
323 p->index_data = NULL;
324 }
325 }
326
327 void close_pack_revindex(struct packed_git *p) {
328 if (!p->revindex_map)
329 return;
330
331 munmap((void *)p->revindex_map, p->revindex_size);
332 p->revindex_map = NULL;
333 p->revindex_data = NULL;
334 }
335
336 void close_pack(struct packed_git *p)
337 {
338 close_pack_windows(p);
339 close_pack_fd(p);
340 close_pack_index(p);
341 close_pack_revindex(p);
342 oidset_clear(&p->bad_objects);
343 }
344
345 void close_object_store(struct raw_object_store *o)
346 {
347 struct packed_git *p;
348
349 for (p = o->packed_git; p; p = p->next)
350 if (p->do_not_close)
351 BUG("want to close pack marked 'do-not-close'");
352 else
353 close_pack(p);
354
355 if (o->multi_pack_index) {
356 close_midx(o->multi_pack_index);
357 o->multi_pack_index = NULL;
358 }
359
360 close_commit_graph(o);
361 }
362
363 void unlink_pack_path(const char *pack_name, int force_delete)
364 {
365 static const char *exts[] = {".pack", ".idx", ".rev", ".keep", ".bitmap", ".promisor"};
366 int i;
367 struct strbuf buf = STRBUF_INIT;
368 size_t plen;
369
370 strbuf_addstr(&buf, pack_name);
371 strip_suffix_mem(buf.buf, &buf.len, ".pack");
372 plen = buf.len;
373
374 if (!force_delete) {
375 strbuf_addstr(&buf, ".keep");
376 if (!access(buf.buf, F_OK)) {
377 strbuf_release(&buf);
378 return;
379 }
380 }
381
382 for (i = 0; i < ARRAY_SIZE(exts); i++) {
383 strbuf_setlen(&buf, plen);
384 strbuf_addstr(&buf, exts[i]);
385 unlink(buf.buf);
386 }
387
388 strbuf_release(&buf);
389 }
390
391 /*
392 * The LRU pack is the one with the oldest MRU window, preferring packs
393 * with no used windows, or the oldest mtime if it has no windows allocated.
394 */
395 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
396 {
397 struct pack_window *w, *this_mru_w;
398 int has_windows_inuse = 0;
399
400 /*
401 * Reject this pack if it has windows and the previously selected
402 * one does not. If this pack does not have windows, reject
403 * it if the pack file is newer than the previously selected one.
404 */
405 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
406 return;
407
408 for (w = this_mru_w = p->windows; w; w = w->next) {
409 /*
410 * Reject this pack if any of its windows are in use,
411 * but the previously selected pack did not have any
412 * inuse windows. Otherwise, record that this pack
413 * has windows in use.
414 */
415 if (w->inuse_cnt) {
416 if (*accept_windows_inuse)
417 has_windows_inuse = 1;
418 else
419 return;
420 }
421
422 if (w->last_used > this_mru_w->last_used)
423 this_mru_w = w;
424
425 /*
426 * Reject this pack if it has windows that have been
427 * used more recently than the previously selected pack.
428 * If the previously selected pack had windows inuse and
429 * we have not encountered a window in this pack that is
430 * inuse, skip this check since we prefer a pack with no
431 * inuse windows to one that has inuse windows.
432 */
433 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
434 this_mru_w->last_used > (*mru_w)->last_used)
435 return;
436 }
437
438 /*
439 * Select this pack.
440 */
441 *mru_w = this_mru_w;
442 *lru_p = p;
443 *accept_windows_inuse = has_windows_inuse;
444 }
445
446 static int close_one_pack(void)
447 {
448 struct packed_git *p, *lru_p = NULL;
449 struct pack_window *mru_w = NULL;
450 int accept_windows_inuse = 1;
451
452 for (p = the_repository->objects->packed_git; p; p = p->next) {
453 if (p->pack_fd == -1)
454 continue;
455 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
456 }
457
458 if (lru_p)
459 return close_pack_fd(lru_p);
460
461 return 0;
462 }
463
464 static unsigned int get_max_fd_limit(void)
465 {
466 #ifdef RLIMIT_NOFILE
467 {
468 struct rlimit lim;
469
470 if (!getrlimit(RLIMIT_NOFILE, &lim))
471 return lim.rlim_cur;
472 }
473 #endif
474
475 #ifdef _SC_OPEN_MAX
476 {
477 long open_max = sysconf(_SC_OPEN_MAX);
478 if (0 < open_max)
479 return open_max;
480 /*
481 * Otherwise, we got -1 for one of the two
482 * reasons:
483 *
484 * (1) sysconf() did not understand _SC_OPEN_MAX
485 * and signaled an error with -1; or
486 * (2) sysconf() said there is no limit.
487 *
488 * We _could_ clear errno before calling sysconf() to
489 * tell these two cases apart and return a huge number
490 * in the latter case to let the caller cap it to a
491 * value that is not so selfish, but letting the
492 * fallback OPEN_MAX codepath take care of these cases
493 * is a lot simpler.
494 */
495 }
496 #endif
497
498 #ifdef OPEN_MAX
499 return OPEN_MAX;
500 #else
501 return 1; /* see the caller ;-) */
502 #endif
503 }
504
505 const char *pack_basename(struct packed_git *p)
506 {
507 const char *ret = strrchr(p->pack_name, '/');
508 if (ret)
509 ret = ret + 1; /* skip past slash */
510 else
511 ret = p->pack_name; /* we only have a base */
512 return ret;
513 }
514
515 /*
516 * Do not call this directly as this leaks p->pack_fd on error return;
517 * call open_packed_git() instead.
518 */
519 static int open_packed_git_1(struct packed_git *p)
520 {
521 struct stat st;
522 struct pack_header hdr;
523 unsigned char hash[GIT_MAX_RAWSZ];
524 unsigned char *idx_hash;
525 ssize_t read_result;
526 const unsigned hashsz = the_hash_algo->rawsz;
527
528 if (open_pack_index(p))
529 return error("packfile %s index unavailable", p->pack_name);
530
531 if (!pack_max_fds) {
532 unsigned int max_fds = get_max_fd_limit();
533
534 /* Save 3 for stdin/stdout/stderr, 22 for work */
535 if (25 < max_fds)
536 pack_max_fds = max_fds - 25;
537 else
538 pack_max_fds = 1;
539 }
540
541 while (pack_max_fds <= pack_open_fds && close_one_pack())
542 ; /* nothing */
543
544 p->pack_fd = git_open(p->pack_name);
545 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
546 return -1;
547 pack_open_fds++;
548
549 /* If we created the struct before we had the pack we lack size. */
550 if (!p->pack_size) {
551 if (!S_ISREG(st.st_mode))
552 return error("packfile %s not a regular file", p->pack_name);
553 p->pack_size = st.st_size;
554 } else if (p->pack_size != st.st_size)
555 return error("packfile %s size changed", p->pack_name);
556
557 /* Verify we recognize this pack file format. */
558 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
559 if (read_result < 0)
560 return error_errno("error reading from %s", p->pack_name);
561 if (read_result != sizeof(hdr))
562 return error("file %s is far too short to be a packfile", p->pack_name);
563 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
564 return error("file %s is not a GIT packfile", p->pack_name);
565 if (!pack_version_ok(hdr.hdr_version))
566 return error("packfile %s is version %"PRIu32" and not"
567 " supported (try upgrading GIT to a newer version)",
568 p->pack_name, ntohl(hdr.hdr_version));
569
570 /* Verify the pack matches its index. */
571 if (p->num_objects != ntohl(hdr.hdr_entries))
572 return error("packfile %s claims to have %"PRIu32" objects"
573 " while index indicates %"PRIu32" objects",
574 p->pack_name, ntohl(hdr.hdr_entries),
575 p->num_objects);
576 read_result = pread_in_full(p->pack_fd, hash, hashsz,
577 p->pack_size - hashsz);
578 if (read_result < 0)
579 return error_errno("error reading from %s", p->pack_name);
580 if (read_result != hashsz)
581 return error("packfile %s signature is unavailable", p->pack_name);
582 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
583 if (!hasheq(hash, idx_hash))
584 return error("packfile %s does not match index", p->pack_name);
585 return 0;
586 }
587
588 static int open_packed_git(struct packed_git *p)
589 {
590 if (!open_packed_git_1(p))
591 return 0;
592 close_pack_fd(p);
593 return -1;
594 }
595
596 static int in_window(struct pack_window *win, off_t offset)
597 {
598 /* We must promise at least one full hash after the
599 * offset is available from this window, otherwise the offset
600 * is not actually in this window and a different window (which
601 * has that one hash excess) must be used. This is to support
602 * the object header and delta base parsing routines below.
603 */
604 off_t win_off = win->offset;
605 return win_off <= offset
606 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
607 }
608
609 unsigned char *use_pack(struct packed_git *p,
610 struct pack_window **w_cursor,
611 off_t offset,
612 unsigned long *left)
613 {
614 struct pack_window *win = *w_cursor;
615
616 /* Since packfiles end in a hash of their content and it's
617 * pointless to ask for an offset into the middle of that
618 * hash, and the in_window function above wouldn't match
619 * don't allow an offset too close to the end of the file.
620 */
621 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
622 die("packfile %s cannot be accessed", p->pack_name);
623 if (offset > (p->pack_size - the_hash_algo->rawsz))
624 die("offset beyond end of packfile (truncated pack?)");
625 if (offset < 0)
626 die(_("offset before end of packfile (broken .idx?)"));
627
628 if (!win || !in_window(win, offset)) {
629 if (win)
630 win->inuse_cnt--;
631 for (win = p->windows; win; win = win->next) {
632 if (in_window(win, offset))
633 break;
634 }
635 if (!win) {
636 size_t window_align = packed_git_window_size / 2;
637 off_t len;
638
639 if (p->pack_fd == -1 && open_packed_git(p))
640 die("packfile %s cannot be accessed", p->pack_name);
641
642 CALLOC_ARRAY(win, 1);
643 win->offset = (offset / window_align) * window_align;
644 len = p->pack_size - win->offset;
645 if (len > packed_git_window_size)
646 len = packed_git_window_size;
647 win->len = (size_t)len;
648 pack_mapped += win->len;
649 while (packed_git_limit < pack_mapped
650 && unuse_one_window(p))
651 ; /* nothing */
652 win->base = xmmap_gently(NULL, win->len,
653 PROT_READ, MAP_PRIVATE,
654 p->pack_fd, win->offset);
655 if (win->base == MAP_FAILED)
656 die_errno(_("packfile %s cannot be mapped%s"),
657 p->pack_name, mmap_os_err());
658 if (!win->offset && win->len == p->pack_size
659 && !p->do_not_close)
660 close_pack_fd(p);
661 pack_mmap_calls++;
662 pack_open_windows++;
663 if (pack_mapped > peak_pack_mapped)
664 peak_pack_mapped = pack_mapped;
665 if (pack_open_windows > peak_pack_open_windows)
666 peak_pack_open_windows = pack_open_windows;
667 win->next = p->windows;
668 p->windows = win;
669 }
670 }
671 if (win != *w_cursor) {
672 win->last_used = pack_used_ctr++;
673 win->inuse_cnt++;
674 *w_cursor = win;
675 }
676 offset -= win->offset;
677 if (left)
678 *left = win->len - xsize_t(offset);
679 return win->base + offset;
680 }
681
682 void unuse_pack(struct pack_window **w_cursor)
683 {
684 struct pack_window *w = *w_cursor;
685 if (w) {
686 w->inuse_cnt--;
687 *w_cursor = NULL;
688 }
689 }
690
691 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
692 {
693 struct stat st;
694 size_t alloc;
695 struct packed_git *p;
696
697 /*
698 * Make sure a corresponding .pack file exists and that
699 * the index looks sane.
700 */
701 if (!strip_suffix_mem(path, &path_len, ".idx"))
702 return NULL;
703
704 /*
705 * ".promisor" is long enough to hold any suffix we're adding (and
706 * the use xsnprintf double-checks that)
707 */
708 alloc = st_add3(path_len, strlen(".promisor"), 1);
709 p = alloc_packed_git(alloc);
710 memcpy(p->pack_name, path, path_len);
711
712 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
713 if (!access(p->pack_name, F_OK))
714 p->pack_keep = 1;
715
716 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
717 if (!access(p->pack_name, F_OK))
718 p->pack_promisor = 1;
719
720 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
721 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
722 free(p);
723 return NULL;
724 }
725
726 /* ok, it looks sane as far as we can check without
727 * actually mapping the pack file.
728 */
729 p->pack_size = st.st_size;
730 p->pack_local = local;
731 p->mtime = st.st_mtime;
732 if (path_len < the_hash_algo->hexsz ||
733 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
734 hashclr(p->hash);
735 return p;
736 }
737
738 void install_packed_git(struct repository *r, struct packed_git *pack)
739 {
740 if (pack->pack_fd != -1)
741 pack_open_fds++;
742
743 pack->next = r->objects->packed_git;
744 r->objects->packed_git = pack;
745
746 hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name));
747 hashmap_add(&r->objects->pack_map, &pack->packmap_ent);
748 }
749
750 void (*report_garbage)(unsigned seen_bits, const char *path);
751
752 static void report_helper(const struct string_list *list,
753 int seen_bits, int first, int last)
754 {
755 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
756 return;
757
758 for (; first < last; first++)
759 report_garbage(seen_bits, list->items[first].string);
760 }
761
762 static void report_pack_garbage(struct string_list *list)
763 {
764 int i, baselen = -1, first = 0, seen_bits = 0;
765
766 if (!report_garbage)
767 return;
768
769 string_list_sort(list);
770
771 for (i = 0; i < list->nr; i++) {
772 const char *path = list->items[i].string;
773 if (baselen != -1 &&
774 strncmp(path, list->items[first].string, baselen)) {
775 report_helper(list, seen_bits, first, i);
776 baselen = -1;
777 seen_bits = 0;
778 }
779 if (baselen == -1) {
780 const char *dot = strrchr(path, '.');
781 if (!dot) {
782 report_garbage(PACKDIR_FILE_GARBAGE, path);
783 continue;
784 }
785 baselen = dot - path + 1;
786 first = i;
787 }
788 if (!strcmp(path + baselen, "pack"))
789 seen_bits |= 1;
790 else if (!strcmp(path + baselen, "idx"))
791 seen_bits |= 2;
792 }
793 report_helper(list, seen_bits, first, list->nr);
794 }
795
796 void for_each_file_in_pack_dir(const char *objdir,
797 each_file_in_pack_dir_fn fn,
798 void *data)
799 {
800 struct strbuf path = STRBUF_INIT;
801 size_t dirnamelen;
802 DIR *dir;
803 struct dirent *de;
804
805 strbuf_addstr(&path, objdir);
806 strbuf_addstr(&path, "/pack");
807 dir = opendir(path.buf);
808 if (!dir) {
809 if (errno != ENOENT)
810 error_errno("unable to open object pack directory: %s",
811 path.buf);
812 strbuf_release(&path);
813 return;
814 }
815 strbuf_addch(&path, '/');
816 dirnamelen = path.len;
817 while ((de = readdir_skip_dot_and_dotdot(dir)) != NULL) {
818 strbuf_setlen(&path, dirnamelen);
819 strbuf_addstr(&path, de->d_name);
820
821 fn(path.buf, path.len, de->d_name, data);
822 }
823
824 closedir(dir);
825 strbuf_release(&path);
826 }
827
828 struct prepare_pack_data {
829 struct repository *r;
830 struct string_list *garbage;
831 int local;
832 struct multi_pack_index *m;
833 };
834
835 static void prepare_pack(const char *full_name, size_t full_name_len,
836 const char *file_name, void *_data)
837 {
838 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
839 struct packed_git *p;
840 size_t base_len = full_name_len;
841
842 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
843 !(data->m && midx_contains_pack(data->m, file_name))) {
844 struct hashmap_entry hent;
845 char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name);
846 unsigned int hash = strhash(pack_name);
847 hashmap_entry_init(&hent, hash);
848
849 /* Don't reopen a pack we already have. */
850 if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) {
851 p = add_packed_git(full_name, full_name_len, data->local);
852 if (p)
853 install_packed_git(data->r, p);
854 }
855 free(pack_name);
856 }
857
858 if (!report_garbage)
859 return;
860
861 if (!strcmp(file_name, "multi-pack-index"))
862 return;
863 if (starts_with(file_name, "multi-pack-index") &&
864 (ends_with(file_name, ".bitmap") || ends_with(file_name, ".rev")))
865 return;
866 if (ends_with(file_name, ".idx") ||
867 ends_with(file_name, ".rev") ||
868 ends_with(file_name, ".pack") ||
869 ends_with(file_name, ".bitmap") ||
870 ends_with(file_name, ".keep") ||
871 ends_with(file_name, ".promisor"))
872 string_list_append(data->garbage, full_name);
873 else
874 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
875 }
876
877 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
878 {
879 struct prepare_pack_data data;
880 struct string_list garbage = STRING_LIST_INIT_DUP;
881
882 data.m = r->objects->multi_pack_index;
883
884 /* look for the multi-pack-index for this object directory */
885 while (data.m && strcmp(data.m->object_dir, objdir))
886 data.m = data.m->next;
887
888 data.r = r;
889 data.garbage = &garbage;
890 data.local = local;
891
892 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
893
894 report_pack_garbage(data.garbage);
895 string_list_clear(data.garbage, 0);
896 }
897
898 static void prepare_packed_git(struct repository *r);
899 /*
900 * Give a fast, rough count of the number of objects in the repository. This
901 * ignores loose objects completely. If you have a lot of them, then either
902 * you should repack because your performance will be awful, or they are
903 * all unreachable objects about to be pruned, in which case they're not really
904 * interesting as a measure of repo size in the first place.
905 */
906 unsigned long repo_approximate_object_count(struct repository *r)
907 {
908 if (!r->objects->approximate_object_count_valid) {
909 unsigned long count;
910 struct multi_pack_index *m;
911 struct packed_git *p;
912
913 prepare_packed_git(r);
914 count = 0;
915 for (m = get_multi_pack_index(r); m; m = m->next)
916 count += m->num_objects;
917 for (p = r->objects->packed_git; p; p = p->next) {
918 if (open_pack_index(p))
919 continue;
920 count += p->num_objects;
921 }
922 r->objects->approximate_object_count = count;
923 r->objects->approximate_object_count_valid = 1;
924 }
925 return r->objects->approximate_object_count;
926 }
927
928 static void *get_next_packed_git(const void *p)
929 {
930 return ((const struct packed_git *)p)->next;
931 }
932
933 static void set_next_packed_git(void *p, void *next)
934 {
935 ((struct packed_git *)p)->next = next;
936 }
937
938 static int sort_pack(const void *a_, const void *b_)
939 {
940 const struct packed_git *a = a_;
941 const struct packed_git *b = b_;
942 int st;
943
944 /*
945 * Local packs tend to contain objects specific to our
946 * variant of the project than remote ones. In addition,
947 * remote ones could be on a network mounted filesystem.
948 * Favor local ones for these reasons.
949 */
950 st = a->pack_local - b->pack_local;
951 if (st)
952 return -st;
953
954 /*
955 * Younger packs tend to contain more recent objects,
956 * and more recent objects tend to get accessed more
957 * often.
958 */
959 if (a->mtime < b->mtime)
960 return 1;
961 else if (a->mtime == b->mtime)
962 return 0;
963 return -1;
964 }
965
966 static void rearrange_packed_git(struct repository *r)
967 {
968 r->objects->packed_git = llist_mergesort(
969 r->objects->packed_git, get_next_packed_git,
970 set_next_packed_git, sort_pack);
971 }
972
973 static void prepare_packed_git_mru(struct repository *r)
974 {
975 struct packed_git *p;
976
977 INIT_LIST_HEAD(&r->objects->packed_git_mru);
978
979 for (p = r->objects->packed_git; p; p = p->next)
980 list_add_tail(&p->mru, &r->objects->packed_git_mru);
981 }
982
983 static void prepare_packed_git(struct repository *r)
984 {
985 struct object_directory *odb;
986
987 if (r->objects->packed_git_initialized)
988 return;
989
990 prepare_alt_odb(r);
991 for (odb = r->objects->odb; odb; odb = odb->next) {
992 int local = (odb == r->objects->odb);
993 prepare_multi_pack_index_one(r, odb->path, local);
994 prepare_packed_git_one(r, odb->path, local);
995 }
996 rearrange_packed_git(r);
997
998 prepare_packed_git_mru(r);
999 r->objects->packed_git_initialized = 1;
1000 }
1001
1002 void reprepare_packed_git(struct repository *r)
1003 {
1004 struct object_directory *odb;
1005
1006 obj_read_lock();
1007 for (odb = r->objects->odb; odb; odb = odb->next)
1008 odb_clear_loose_cache(odb);
1009
1010 r->objects->approximate_object_count_valid = 0;
1011 r->objects->packed_git_initialized = 0;
1012 prepare_packed_git(r);
1013 obj_read_unlock();
1014 }
1015
1016 struct packed_git *get_packed_git(struct repository *r)
1017 {
1018 prepare_packed_git(r);
1019 return r->objects->packed_git;
1020 }
1021
1022 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1023 {
1024 prepare_packed_git(r);
1025 return r->objects->multi_pack_index;
1026 }
1027
1028 struct multi_pack_index *get_local_multi_pack_index(struct repository *r)
1029 {
1030 struct multi_pack_index *m = get_multi_pack_index(r);
1031
1032 /* no need to iterate; we always put the local one first (if any) */
1033 if (m && m->local)
1034 return m;
1035
1036 return NULL;
1037 }
1038
1039 struct packed_git *get_all_packs(struct repository *r)
1040 {
1041 struct multi_pack_index *m;
1042
1043 prepare_packed_git(r);
1044 for (m = r->objects->multi_pack_index; m; m = m->next) {
1045 uint32_t i;
1046 for (i = 0; i < m->num_packs; i++)
1047 prepare_midx_pack(r, m, i);
1048 }
1049
1050 return r->objects->packed_git;
1051 }
1052
1053 struct list_head *get_packed_git_mru(struct repository *r)
1054 {
1055 prepare_packed_git(r);
1056 return &r->objects->packed_git_mru;
1057 }
1058
1059 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1060 unsigned long len, enum object_type *type, unsigned long *sizep)
1061 {
1062 unsigned shift;
1063 unsigned long size, c;
1064 unsigned long used = 0;
1065
1066 c = buf[used++];
1067 *type = (c >> 4) & 7;
1068 size = c & 15;
1069 shift = 4;
1070 while (c & 0x80) {
1071 if (len <= used || bitsizeof(long) <= shift) {
1072 error("bad object header");
1073 size = used = 0;
1074 break;
1075 }
1076 c = buf[used++];
1077 size += (c & 0x7f) << shift;
1078 shift += 7;
1079 }
1080 *sizep = size;
1081 return used;
1082 }
1083
1084 unsigned long get_size_from_delta(struct packed_git *p,
1085 struct pack_window **w_curs,
1086 off_t curpos)
1087 {
1088 const unsigned char *data;
1089 unsigned char delta_head[20], *in;
1090 git_zstream stream;
1091 int st;
1092
1093 memset(&stream, 0, sizeof(stream));
1094 stream.next_out = delta_head;
1095 stream.avail_out = sizeof(delta_head);
1096
1097 git_inflate_init(&stream);
1098 do {
1099 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1100 stream.next_in = in;
1101 /*
1102 * Note: the window section returned by use_pack() must be
1103 * available throughout git_inflate()'s unlocked execution. To
1104 * ensure no other thread will modify the window in the
1105 * meantime, we rely on the packed_window.inuse_cnt. This
1106 * counter is incremented before window reading and checked
1107 * before window disposal.
1108 *
1109 * Other worrying sections could be the call to close_pack_fd(),
1110 * which can close packs even with in-use windows, and to
1111 * reprepare_packed_git(). Regarding the former, mmap doc says:
1112 * "closing the file descriptor does not unmap the region". And
1113 * for the latter, it won't re-open already available packs.
1114 */
1115 obj_read_unlock();
1116 st = git_inflate(&stream, Z_FINISH);
1117 obj_read_lock();
1118 curpos += stream.next_in - in;
1119 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1120 stream.total_out < sizeof(delta_head));
1121 git_inflate_end(&stream);
1122 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1123 error("delta data unpack-initial failed");
1124 return 0;
1125 }
1126
1127 /* Examine the initial part of the delta to figure out
1128 * the result size.
1129 */
1130 data = delta_head;
1131
1132 /* ignore base size */
1133 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1134
1135 /* Read the result size */
1136 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1137 }
1138
1139 int unpack_object_header(struct packed_git *p,
1140 struct pack_window **w_curs,
1141 off_t *curpos,
1142 unsigned long *sizep)
1143 {
1144 unsigned char *base;
1145 unsigned long left;
1146 unsigned long used;
1147 enum object_type type;
1148
1149 /* use_pack() assures us we have [base, base + 20) available
1150 * as a range that we can look at. (Its actually the hash
1151 * size that is assured.) With our object header encoding
1152 * the maximum deflated object size is 2^137, which is just
1153 * insane, so we know won't exceed what we have been given.
1154 */
1155 base = use_pack(p, w_curs, *curpos, &left);
1156 used = unpack_object_header_buffer(base, left, &type, sizep);
1157 if (!used) {
1158 type = OBJ_BAD;
1159 } else
1160 *curpos += used;
1161
1162 return type;
1163 }
1164
1165 void mark_bad_packed_object(struct packed_git *p, const struct object_id *oid)
1166 {
1167 oidset_insert(&p->bad_objects, oid);
1168 }
1169
1170 const struct packed_git *has_packed_and_bad(struct repository *r,
1171 const struct object_id *oid)
1172 {
1173 struct packed_git *p;
1174
1175 for (p = r->objects->packed_git; p; p = p->next)
1176 if (oidset_contains(&p->bad_objects, oid))
1177 return p;
1178 return NULL;
1179 }
1180
1181 off_t get_delta_base(struct packed_git *p,
1182 struct pack_window **w_curs,
1183 off_t *curpos,
1184 enum object_type type,
1185 off_t delta_obj_offset)
1186 {
1187 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1188 off_t base_offset;
1189
1190 /* use_pack() assured us we have [base_info, base_info + 20)
1191 * as a range that we can look at without walking off the
1192 * end of the mapped window. Its actually the hash size
1193 * that is assured. An OFS_DELTA longer than the hash size
1194 * is stupid, as then a REF_DELTA would be smaller to store.
1195 */
1196 if (type == OBJ_OFS_DELTA) {
1197 unsigned used = 0;
1198 unsigned char c = base_info[used++];
1199 base_offset = c & 127;
1200 while (c & 128) {
1201 base_offset += 1;
1202 if (!base_offset || MSB(base_offset, 7))
1203 return 0; /* overflow */
1204 c = base_info[used++];
1205 base_offset = (base_offset << 7) + (c & 127);
1206 }
1207 base_offset = delta_obj_offset - base_offset;
1208 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1209 return 0; /* out of bound */
1210 *curpos += used;
1211 } else if (type == OBJ_REF_DELTA) {
1212 /* The base entry _must_ be in the same pack */
1213 base_offset = find_pack_entry_one(base_info, p);
1214 *curpos += the_hash_algo->rawsz;
1215 } else
1216 die("I am totally screwed");
1217 return base_offset;
1218 }
1219
1220 /*
1221 * Like get_delta_base above, but we return the sha1 instead of the pack
1222 * offset. This means it is cheaper for REF deltas (we do not have to do
1223 * the final object lookup), but more expensive for OFS deltas (we
1224 * have to load the revidx to convert the offset back into a sha1).
1225 */
1226 static int get_delta_base_oid(struct packed_git *p,
1227 struct pack_window **w_curs,
1228 off_t curpos,
1229 struct object_id *oid,
1230 enum object_type type,
1231 off_t delta_obj_offset)
1232 {
1233 if (type == OBJ_REF_DELTA) {
1234 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1235 oidread(oid, base);
1236 return 0;
1237 } else if (type == OBJ_OFS_DELTA) {
1238 uint32_t base_pos;
1239 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1240 type, delta_obj_offset);
1241
1242 if (!base_offset)
1243 return -1;
1244
1245 if (offset_to_pack_pos(p, base_offset, &base_pos) < 0)
1246 return -1;
1247
1248 return nth_packed_object_id(oid, p,
1249 pack_pos_to_index(p, base_pos));
1250 } else
1251 return -1;
1252 }
1253
1254 static int retry_bad_packed_offset(struct repository *r,
1255 struct packed_git *p,
1256 off_t obj_offset)
1257 {
1258 int type;
1259 uint32_t pos;
1260 struct object_id oid;
1261 if (offset_to_pack_pos(p, obj_offset, &pos) < 0)
1262 return OBJ_BAD;
1263 nth_packed_object_id(&oid, p, pack_pos_to_index(p, pos));
1264 mark_bad_packed_object(p, &oid);
1265 type = oid_object_info(r, &oid, NULL);
1266 if (type <= OBJ_NONE)
1267 return OBJ_BAD;
1268 return type;
1269 }
1270
1271 #define POI_STACK_PREALLOC 64
1272
1273 static enum object_type packed_to_object_type(struct repository *r,
1274 struct packed_git *p,
1275 off_t obj_offset,
1276 enum object_type type,
1277 struct pack_window **w_curs,
1278 off_t curpos)
1279 {
1280 off_t small_poi_stack[POI_STACK_PREALLOC];
1281 off_t *poi_stack = small_poi_stack;
1282 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1283
1284 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1285 off_t base_offset;
1286 unsigned long size;
1287 /* Push the object we're going to leave behind */
1288 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1289 poi_stack_alloc = alloc_nr(poi_stack_nr);
1290 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1291 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1292 } else {
1293 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1294 }
1295 poi_stack[poi_stack_nr++] = obj_offset;
1296 /* If parsing the base offset fails, just unwind */
1297 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1298 if (!base_offset)
1299 goto unwind;
1300 curpos = obj_offset = base_offset;
1301 type = unpack_object_header(p, w_curs, &curpos, &size);
1302 if (type <= OBJ_NONE) {
1303 /* If getting the base itself fails, we first
1304 * retry the base, otherwise unwind */
1305 type = retry_bad_packed_offset(r, p, base_offset);
1306 if (type > OBJ_NONE)
1307 goto out;
1308 goto unwind;
1309 }
1310 }
1311
1312 switch (type) {
1313 case OBJ_BAD:
1314 case OBJ_COMMIT:
1315 case OBJ_TREE:
1316 case OBJ_BLOB:
1317 case OBJ_TAG:
1318 break;
1319 default:
1320 error("unknown object type %i at offset %"PRIuMAX" in %s",
1321 type, (uintmax_t)obj_offset, p->pack_name);
1322 type = OBJ_BAD;
1323 }
1324
1325 out:
1326 if (poi_stack != small_poi_stack)
1327 free(poi_stack);
1328 return type;
1329
1330 unwind:
1331 while (poi_stack_nr) {
1332 obj_offset = poi_stack[--poi_stack_nr];
1333 type = retry_bad_packed_offset(r, p, obj_offset);
1334 if (type > OBJ_NONE)
1335 goto out;
1336 }
1337 type = OBJ_BAD;
1338 goto out;
1339 }
1340
1341 static struct hashmap delta_base_cache;
1342 static size_t delta_base_cached;
1343
1344 static LIST_HEAD(delta_base_cache_lru);
1345
1346 struct delta_base_cache_key {
1347 struct packed_git *p;
1348 off_t base_offset;
1349 };
1350
1351 struct delta_base_cache_entry {
1352 struct hashmap_entry ent;
1353 struct delta_base_cache_key key;
1354 struct list_head lru;
1355 void *data;
1356 unsigned long size;
1357 enum object_type type;
1358 };
1359
1360 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1361 {
1362 unsigned int hash;
1363
1364 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1365 hash += (hash >> 8) + (hash >> 16);
1366 return hash;
1367 }
1368
1369 static struct delta_base_cache_entry *
1370 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1371 {
1372 struct hashmap_entry entry, *e;
1373 struct delta_base_cache_key key;
1374
1375 if (!delta_base_cache.cmpfn)
1376 return NULL;
1377
1378 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1379 key.p = p;
1380 key.base_offset = base_offset;
1381 e = hashmap_get(&delta_base_cache, &entry, &key);
1382 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1383 }
1384
1385 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1386 const struct delta_base_cache_key *b)
1387 {
1388 return a->p == b->p && a->base_offset == b->base_offset;
1389 }
1390
1391 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1392 const struct hashmap_entry *va,
1393 const struct hashmap_entry *vb,
1394 const void *vkey)
1395 {
1396 const struct delta_base_cache_entry *a, *b;
1397 const struct delta_base_cache_key *key = vkey;
1398
1399 a = container_of(va, const struct delta_base_cache_entry, ent);
1400 b = container_of(vb, const struct delta_base_cache_entry, ent);
1401
1402 if (key)
1403 return !delta_base_cache_key_eq(&a->key, key);
1404 else
1405 return !delta_base_cache_key_eq(&a->key, &b->key);
1406 }
1407
1408 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1409 {
1410 return !!get_delta_base_cache_entry(p, base_offset);
1411 }
1412
1413 /*
1414 * Remove the entry from the cache, but do _not_ free the associated
1415 * entry data. The caller takes ownership of the "data" buffer, and
1416 * should copy out any fields it wants before detaching.
1417 */
1418 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1419 {
1420 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1421 list_del(&ent->lru);
1422 delta_base_cached -= ent->size;
1423 free(ent);
1424 }
1425
1426 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1427 off_t base_offset, unsigned long *base_size,
1428 enum object_type *type)
1429 {
1430 struct delta_base_cache_entry *ent;
1431
1432 ent = get_delta_base_cache_entry(p, base_offset);
1433 if (!ent)
1434 return unpack_entry(r, p, base_offset, type, base_size);
1435
1436 if (type)
1437 *type = ent->type;
1438 if (base_size)
1439 *base_size = ent->size;
1440 return xmemdupz(ent->data, ent->size);
1441 }
1442
1443 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1444 {
1445 free(ent->data);
1446 detach_delta_base_cache_entry(ent);
1447 }
1448
1449 void clear_delta_base_cache(void)
1450 {
1451 struct list_head *lru, *tmp;
1452 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1453 struct delta_base_cache_entry *entry =
1454 list_entry(lru, struct delta_base_cache_entry, lru);
1455 release_delta_base_cache(entry);
1456 }
1457 }
1458
1459 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1460 void *base, unsigned long base_size, enum object_type type)
1461 {
1462 struct delta_base_cache_entry *ent;
1463 struct list_head *lru, *tmp;
1464
1465 /*
1466 * Check required to avoid redundant entries when more than one thread
1467 * is unpacking the same object, in unpack_entry() (since its phases I
1468 * and III might run concurrently across multiple threads).
1469 */
1470 if (in_delta_base_cache(p, base_offset)) {
1471 free(base);
1472 return;
1473 }
1474
1475 delta_base_cached += base_size;
1476
1477 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1478 struct delta_base_cache_entry *f =
1479 list_entry(lru, struct delta_base_cache_entry, lru);
1480 if (delta_base_cached <= delta_base_cache_limit)
1481 break;
1482 release_delta_base_cache(f);
1483 }
1484
1485 ent = xmalloc(sizeof(*ent));
1486 ent->key.p = p;
1487 ent->key.base_offset = base_offset;
1488 ent->type = type;
1489 ent->data = base;
1490 ent->size = base_size;
1491 list_add_tail(&ent->lru, &delta_base_cache_lru);
1492
1493 if (!delta_base_cache.cmpfn)
1494 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1495 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1496 hashmap_add(&delta_base_cache, &ent->ent);
1497 }
1498
1499 int packed_object_info(struct repository *r, struct packed_git *p,
1500 off_t obj_offset, struct object_info *oi)
1501 {
1502 struct pack_window *w_curs = NULL;
1503 unsigned long size;
1504 off_t curpos = obj_offset;
1505 enum object_type type;
1506
1507 /*
1508 * We always get the representation type, but only convert it to
1509 * a "real" type later if the caller is interested.
1510 */
1511 if (oi->contentp) {
1512 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1513 &type);
1514 if (!*oi->contentp)
1515 type = OBJ_BAD;
1516 } else {
1517 type = unpack_object_header(p, &w_curs, &curpos, &size);
1518 }
1519
1520 if (!oi->contentp && oi->sizep) {
1521 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1522 off_t tmp_pos = curpos;
1523 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1524 type, obj_offset);
1525 if (!base_offset) {
1526 type = OBJ_BAD;
1527 goto out;
1528 }
1529 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1530 if (*oi->sizep == 0) {
1531 type = OBJ_BAD;
1532 goto out;
1533 }
1534 } else {
1535 *oi->sizep = size;
1536 }
1537 }
1538
1539 if (oi->disk_sizep) {
1540 uint32_t pos;
1541 if (offset_to_pack_pos(p, obj_offset, &pos) < 0) {
1542 error("could not find object at offset %"PRIuMAX" "
1543 "in pack %s", (uintmax_t)obj_offset, p->pack_name);
1544 type = OBJ_BAD;
1545 goto out;
1546 }
1547
1548 *oi->disk_sizep = pack_pos_to_offset(p, pos + 1) - obj_offset;
1549 }
1550
1551 if (oi->typep || oi->type_name) {
1552 enum object_type ptot;
1553 ptot = packed_to_object_type(r, p, obj_offset,
1554 type, &w_curs, curpos);
1555 if (oi->typep)
1556 *oi->typep = ptot;
1557 if (oi->type_name) {
1558 const char *tn = type_name(ptot);
1559 if (tn)
1560 strbuf_addstr(oi->type_name, tn);
1561 }
1562 if (ptot < 0) {
1563 type = OBJ_BAD;
1564 goto out;
1565 }
1566 }
1567
1568 if (oi->delta_base_oid) {
1569 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1570 if (get_delta_base_oid(p, &w_curs, curpos,
1571 oi->delta_base_oid,
1572 type, obj_offset) < 0) {
1573 type = OBJ_BAD;
1574 goto out;
1575 }
1576 } else
1577 oidclr(oi->delta_base_oid);
1578 }
1579
1580 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1581 OI_PACKED;
1582
1583 out:
1584 unuse_pack(&w_curs);
1585 return type;
1586 }
1587
1588 static void *unpack_compressed_entry(struct packed_git *p,
1589 struct pack_window **w_curs,
1590 off_t curpos,
1591 unsigned long size)
1592 {
1593 int st;
1594 git_zstream stream;
1595 unsigned char *buffer, *in;
1596
1597 buffer = xmallocz_gently(size);
1598 if (!buffer)
1599 return NULL;
1600 memset(&stream, 0, sizeof(stream));
1601 stream.next_out = buffer;
1602 stream.avail_out = size + 1;
1603
1604 git_inflate_init(&stream);
1605 do {
1606 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1607 stream.next_in = in;
1608 /*
1609 * Note: we must ensure the window section returned by
1610 * use_pack() will be available throughout git_inflate()'s
1611 * unlocked execution. Please refer to the comment at
1612 * get_size_from_delta() to see how this is done.
1613 */
1614 obj_read_unlock();
1615 st = git_inflate(&stream, Z_FINISH);
1616 obj_read_lock();
1617 if (!stream.avail_out)
1618 break; /* the payload is larger than it should be */
1619 curpos += stream.next_in - in;
1620 } while (st == Z_OK || st == Z_BUF_ERROR);
1621 git_inflate_end(&stream);
1622 if ((st != Z_STREAM_END) || stream.total_out != size) {
1623 free(buffer);
1624 return NULL;
1625 }
1626
1627 /* versions of zlib can clobber unconsumed portion of outbuf */
1628 buffer[size] = '\0';
1629
1630 return buffer;
1631 }
1632
1633 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1634 {
1635 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1636 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1637 p->pack_name, (uintmax_t)obj_offset);
1638 }
1639
1640 int do_check_packed_object_crc;
1641
1642 #define UNPACK_ENTRY_STACK_PREALLOC 64
1643 struct unpack_entry_stack_ent {
1644 off_t obj_offset;
1645 off_t curpos;
1646 unsigned long size;
1647 };
1648
1649 static void *read_object(struct repository *r,
1650 const struct object_id *oid,
1651 enum object_type *type,
1652 unsigned long *size)
1653 {
1654 struct object_info oi = OBJECT_INFO_INIT;
1655 void *content;
1656 oi.typep = type;
1657 oi.sizep = size;
1658 oi.contentp = &content;
1659
1660 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1661 return NULL;
1662 return content;
1663 }
1664
1665 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1666 enum object_type *final_type, unsigned long *final_size)
1667 {
1668 struct pack_window *w_curs = NULL;
1669 off_t curpos = obj_offset;
1670 void *data = NULL;
1671 unsigned long size;
1672 enum object_type type;
1673 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1674 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1675 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1676 int base_from_cache = 0;
1677
1678 write_pack_access_log(p, obj_offset);
1679
1680 /* PHASE 1: drill down to the innermost base object */
1681 for (;;) {
1682 off_t base_offset;
1683 int i;
1684 struct delta_base_cache_entry *ent;
1685
1686 ent = get_delta_base_cache_entry(p, curpos);
1687 if (ent) {
1688 type = ent->type;
1689 data = ent->data;
1690 size = ent->size;
1691 detach_delta_base_cache_entry(ent);
1692 base_from_cache = 1;
1693 break;
1694 }
1695
1696 if (do_check_packed_object_crc && p->index_version > 1) {
1697 uint32_t pack_pos, index_pos;
1698 off_t len;
1699
1700 if (offset_to_pack_pos(p, obj_offset, &pack_pos) < 0) {
1701 error("could not find object at offset %"PRIuMAX" in pack %s",
1702 (uintmax_t)obj_offset, p->pack_name);
1703 data = NULL;
1704 goto out;
1705 }
1706
1707 len = pack_pos_to_offset(p, pack_pos + 1) - obj_offset;
1708 index_pos = pack_pos_to_index(p, pack_pos);
1709 if (check_pack_crc(p, &w_curs, obj_offset, len, index_pos)) {
1710 struct object_id oid;
1711 nth_packed_object_id(&oid, p, index_pos);
1712 error("bad packed object CRC for %s",
1713 oid_to_hex(&oid));
1714 mark_bad_packed_object(p, &oid);
1715 data = NULL;
1716 goto out;
1717 }
1718 }
1719
1720 type = unpack_object_header(p, &w_curs, &curpos, &size);
1721 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1722 break;
1723
1724 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1725 if (!base_offset) {
1726 error("failed to validate delta base reference "
1727 "at offset %"PRIuMAX" from %s",
1728 (uintmax_t)curpos, p->pack_name);
1729 /* bail to phase 2, in hopes of recovery */
1730 data = NULL;
1731 break;
1732 }
1733
1734 /* push object, proceed to base */
1735 if (delta_stack_nr >= delta_stack_alloc
1736 && delta_stack == small_delta_stack) {
1737 delta_stack_alloc = alloc_nr(delta_stack_nr);
1738 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1739 COPY_ARRAY(delta_stack, small_delta_stack,
1740 delta_stack_nr);
1741 } else {
1742 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1743 }
1744 i = delta_stack_nr++;
1745 delta_stack[i].obj_offset = obj_offset;
1746 delta_stack[i].curpos = curpos;
1747 delta_stack[i].size = size;
1748
1749 curpos = obj_offset = base_offset;
1750 }
1751
1752 /* PHASE 2: handle the base */
1753 switch (type) {
1754 case OBJ_OFS_DELTA:
1755 case OBJ_REF_DELTA:
1756 if (data)
1757 BUG("unpack_entry: left loop at a valid delta");
1758 break;
1759 case OBJ_COMMIT:
1760 case OBJ_TREE:
1761 case OBJ_BLOB:
1762 case OBJ_TAG:
1763 if (!base_from_cache)
1764 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1765 break;
1766 default:
1767 data = NULL;
1768 error("unknown object type %i at offset %"PRIuMAX" in %s",
1769 type, (uintmax_t)obj_offset, p->pack_name);
1770 }
1771
1772 /* PHASE 3: apply deltas in order */
1773
1774 /* invariants:
1775 * 'data' holds the base data, or NULL if there was corruption
1776 */
1777 while (delta_stack_nr) {
1778 void *delta_data;
1779 void *base = data;
1780 void *external_base = NULL;
1781 unsigned long delta_size, base_size = size;
1782 int i;
1783 off_t base_obj_offset = obj_offset;
1784
1785 data = NULL;
1786
1787 if (!base) {
1788 /*
1789 * We're probably in deep shit, but let's try to fetch
1790 * the required base anyway from another pack or loose.
1791 * This is costly but should happen only in the presence
1792 * of a corrupted pack, and is better than failing outright.
1793 */
1794 uint32_t pos;
1795 struct object_id base_oid;
1796 if (!(offset_to_pack_pos(p, obj_offset, &pos))) {
1797 nth_packed_object_id(&base_oid, p,
1798 pack_pos_to_index(p, pos));
1799 error("failed to read delta base object %s"
1800 " at offset %"PRIuMAX" from %s",
1801 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1802 p->pack_name);
1803 mark_bad_packed_object(p, &base_oid);
1804 base = read_object(r, &base_oid, &type, &base_size);
1805 external_base = base;
1806 }
1807 }
1808
1809 i = --delta_stack_nr;
1810 obj_offset = delta_stack[i].obj_offset;
1811 curpos = delta_stack[i].curpos;
1812 delta_size = delta_stack[i].size;
1813
1814 if (!base)
1815 continue;
1816
1817 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1818
1819 if (!delta_data) {
1820 error("failed to unpack compressed delta "
1821 "at offset %"PRIuMAX" from %s",
1822 (uintmax_t)curpos, p->pack_name);
1823 data = NULL;
1824 } else {
1825 data = patch_delta(base, base_size, delta_data,
1826 delta_size, &size);
1827
1828 /*
1829 * We could not apply the delta; warn the user, but
1830 * keep going. Our failure will be noticed either in
1831 * the next iteration of the loop, or if this is the
1832 * final delta, in the caller when we return NULL.
1833 * Those code paths will take care of making a more
1834 * explicit warning and retrying with another copy of
1835 * the object.
1836 */
1837 if (!data)
1838 error("failed to apply delta");
1839 }
1840
1841 /*
1842 * We delay adding `base` to the cache until the end of the loop
1843 * because unpack_compressed_entry() momentarily releases the
1844 * obj_read_mutex, giving another thread the chance to access
1845 * the cache. Therefore, if `base` was already there, this other
1846 * thread could free() it (e.g. to make space for another entry)
1847 * before we are done using it.
1848 */
1849 if (!external_base)
1850 add_delta_base_cache(p, base_obj_offset, base, base_size, type);
1851
1852 free(delta_data);
1853 free(external_base);
1854 }
1855
1856 if (final_type)
1857 *final_type = type;
1858 if (final_size)
1859 *final_size = size;
1860
1861 out:
1862 unuse_pack(&w_curs);
1863
1864 if (delta_stack != small_delta_stack)
1865 free(delta_stack);
1866
1867 return data;
1868 }
1869
1870 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1871 {
1872 const unsigned char *index_fanout = p->index_data;
1873 const unsigned char *index_lookup;
1874 const unsigned int hashsz = the_hash_algo->rawsz;
1875 int index_lookup_width;
1876
1877 if (!index_fanout)
1878 BUG("bsearch_pack called without a valid pack-index");
1879
1880 index_lookup = index_fanout + 4 * 256;
1881 if (p->index_version == 1) {
1882 index_lookup_width = hashsz + 4;
1883 index_lookup += 4;
1884 } else {
1885 index_lookup_width = hashsz;
1886 index_fanout += 8;
1887 index_lookup += 8;
1888 }
1889
1890 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1891 index_lookup, index_lookup_width, result);
1892 }
1893
1894 int nth_packed_object_id(struct object_id *oid,
1895 struct packed_git *p,
1896 uint32_t n)
1897 {
1898 const unsigned char *index = p->index_data;
1899 const unsigned int hashsz = the_hash_algo->rawsz;
1900 if (!index) {
1901 if (open_pack_index(p))
1902 return -1;
1903 index = p->index_data;
1904 }
1905 if (n >= p->num_objects)
1906 return -1;
1907 index += 4 * 256;
1908 if (p->index_version == 1) {
1909 oidread(oid, index + (hashsz + 4) * n + 4);
1910 } else {
1911 index += 8;
1912 oidread(oid, index + hashsz * n);
1913 }
1914 return 0;
1915 }
1916
1917 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1918 {
1919 const unsigned char *ptr = vptr;
1920 const unsigned char *start = p->index_data;
1921 const unsigned char *end = start + p->index_size;
1922 if (ptr < start)
1923 die(_("offset before start of pack index for %s (corrupt index?)"),
1924 p->pack_name);
1925 /* No need to check for underflow; .idx files must be at least 8 bytes */
1926 if (ptr >= end - 8)
1927 die(_("offset beyond end of pack index for %s (truncated index?)"),
1928 p->pack_name);
1929 }
1930
1931 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1932 {
1933 const unsigned char *index = p->index_data;
1934 const unsigned int hashsz = the_hash_algo->rawsz;
1935 index += 4 * 256;
1936 if (p->index_version == 1) {
1937 return ntohl(*((uint32_t *)(index + (hashsz + 4) * (size_t)n)));
1938 } else {
1939 uint32_t off;
1940 index += 8 + (size_t)p->num_objects * (hashsz + 4);
1941 off = ntohl(*((uint32_t *)(index + 4 * n)));
1942 if (!(off & 0x80000000))
1943 return off;
1944 index += (size_t)p->num_objects * 4 + (off & 0x7fffffff) * 8;
1945 check_pack_index_ptr(p, index);
1946 return get_be64(index);
1947 }
1948 }
1949
1950 off_t find_pack_entry_one(const unsigned char *sha1,
1951 struct packed_git *p)
1952 {
1953 const unsigned char *index = p->index_data;
1954 struct object_id oid;
1955 uint32_t result;
1956
1957 if (!index) {
1958 if (open_pack_index(p))
1959 return 0;
1960 }
1961
1962 hashcpy(oid.hash, sha1);
1963 if (bsearch_pack(&oid, p, &result))
1964 return nth_packed_object_offset(p, result);
1965 return 0;
1966 }
1967
1968 int is_pack_valid(struct packed_git *p)
1969 {
1970 /* An already open pack is known to be valid. */
1971 if (p->pack_fd != -1)
1972 return 1;
1973
1974 /* If the pack has one window completely covering the
1975 * file size, the pack is known to be valid even if
1976 * the descriptor is not currently open.
1977 */
1978 if (p->windows) {
1979 struct pack_window *w = p->windows;
1980
1981 if (!w->offset && w->len == p->pack_size)
1982 return 1;
1983 }
1984
1985 /* Force the pack to open to prove its valid. */
1986 return !open_packed_git(p);
1987 }
1988
1989 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1990 struct packed_git *packs)
1991 {
1992 struct packed_git *p;
1993
1994 for (p = packs; p; p = p->next) {
1995 if (find_pack_entry_one(sha1, p))
1996 return p;
1997 }
1998 return NULL;
1999
2000 }
2001
2002 static int fill_pack_entry(const struct object_id *oid,
2003 struct pack_entry *e,
2004 struct packed_git *p)
2005 {
2006 off_t offset;
2007
2008 if (oidset_size(&p->bad_objects) &&
2009 oidset_contains(&p->bad_objects, oid))
2010 return 0;
2011
2012 offset = find_pack_entry_one(oid->hash, p);
2013 if (!offset)
2014 return 0;
2015
2016 /*
2017 * We are about to tell the caller where they can locate the
2018 * requested object. We better make sure the packfile is
2019 * still here and can be accessed before supplying that
2020 * answer, as it may have been deleted since the index was
2021 * loaded!
2022 */
2023 if (!is_pack_valid(p))
2024 return 0;
2025 e->offset = offset;
2026 e->p = p;
2027 return 1;
2028 }
2029
2030 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2031 {
2032 struct list_head *pos;
2033 struct multi_pack_index *m;
2034
2035 prepare_packed_git(r);
2036 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2037 return 0;
2038
2039 for (m = r->objects->multi_pack_index; m; m = m->next) {
2040 if (fill_midx_entry(r, oid, e, m))
2041 return 1;
2042 }
2043
2044 list_for_each(pos, &r->objects->packed_git_mru) {
2045 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2046 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2047 list_move(&p->mru, &r->objects->packed_git_mru);
2048 return 1;
2049 }
2050 }
2051 return 0;
2052 }
2053
2054 static void maybe_invalidate_kept_pack_cache(struct repository *r,
2055 unsigned flags)
2056 {
2057 if (!r->objects->kept_pack_cache.packs)
2058 return;
2059 if (r->objects->kept_pack_cache.flags == flags)
2060 return;
2061 FREE_AND_NULL(r->objects->kept_pack_cache.packs);
2062 r->objects->kept_pack_cache.flags = 0;
2063 }
2064
2065 static struct packed_git **kept_pack_cache(struct repository *r, unsigned flags)
2066 {
2067 maybe_invalidate_kept_pack_cache(r, flags);
2068
2069 if (!r->objects->kept_pack_cache.packs) {
2070 struct packed_git **packs = NULL;
2071 size_t nr = 0, alloc = 0;
2072 struct packed_git *p;
2073
2074 /*
2075 * We want "all" packs here, because we need to cover ones that
2076 * are used by a midx, as well. We need to look in every one of
2077 * them (instead of the midx itself) to cover duplicates. It's
2078 * possible that an object is found in two packs that the midx
2079 * covers, one kept and one not kept, but the midx returns only
2080 * the non-kept version.
2081 */
2082 for (p = get_all_packs(r); p; p = p->next) {
2083 if ((p->pack_keep && (flags & ON_DISK_KEEP_PACKS)) ||
2084 (p->pack_keep_in_core && (flags & IN_CORE_KEEP_PACKS))) {
2085 ALLOC_GROW(packs, nr + 1, alloc);
2086 packs[nr++] = p;
2087 }
2088 }
2089 ALLOC_GROW(packs, nr + 1, alloc);
2090 packs[nr] = NULL;
2091
2092 r->objects->kept_pack_cache.packs = packs;
2093 r->objects->kept_pack_cache.flags = flags;
2094 }
2095
2096 return r->objects->kept_pack_cache.packs;
2097 }
2098
2099 int find_kept_pack_entry(struct repository *r,
2100 const struct object_id *oid,
2101 unsigned flags,
2102 struct pack_entry *e)
2103 {
2104 struct packed_git **cache;
2105
2106 for (cache = kept_pack_cache(r, flags); *cache; cache++) {
2107 struct packed_git *p = *cache;
2108 if (fill_pack_entry(oid, e, p))
2109 return 1;
2110 }
2111
2112 return 0;
2113 }
2114
2115 int has_object_pack(const struct object_id *oid)
2116 {
2117 struct pack_entry e;
2118 return find_pack_entry(the_repository, oid, &e);
2119 }
2120
2121 int has_object_kept_pack(const struct object_id *oid, unsigned flags)
2122 {
2123 struct pack_entry e;
2124 return find_kept_pack_entry(the_repository, oid, flags, &e);
2125 }
2126
2127 int has_pack_index(const unsigned char *sha1)
2128 {
2129 struct stat st;
2130 if (stat(sha1_pack_index_name(sha1), &st))
2131 return 0;
2132 return 1;
2133 }
2134
2135 int for_each_object_in_pack(struct packed_git *p,
2136 each_packed_object_fn cb, void *data,
2137 enum for_each_object_flags flags)
2138 {
2139 uint32_t i;
2140 int r = 0;
2141
2142 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2143 if (load_pack_revindex(p))
2144 return -1;
2145 }
2146
2147 for (i = 0; i < p->num_objects; i++) {
2148 uint32_t index_pos;
2149 struct object_id oid;
2150
2151 /*
2152 * We are iterating "i" from 0 up to num_objects, but its
2153 * meaning may be different, depending on the requested output
2154 * order:
2155 *
2156 * - in object-name order, it is the same as the index order
2157 * used by nth_packed_object_id(), so we can pass it
2158 * directly
2159 *
2160 * - in pack-order, it is pack position, which we must
2161 * convert to an index position in order to get the oid.
2162 */
2163 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2164 index_pos = pack_pos_to_index(p, i);
2165 else
2166 index_pos = i;
2167
2168 if (nth_packed_object_id(&oid, p, index_pos) < 0)
2169 return error("unable to get sha1 of object %u in %s",
2170 index_pos, p->pack_name);
2171
2172 r = cb(&oid, p, index_pos, data);
2173 if (r)
2174 break;
2175 }
2176 return r;
2177 }
2178
2179 int for_each_packed_object(each_packed_object_fn cb, void *data,
2180 enum for_each_object_flags flags)
2181 {
2182 struct packed_git *p;
2183 int r = 0;
2184 int pack_errors = 0;
2185
2186 prepare_packed_git(the_repository);
2187 for (p = get_all_packs(the_repository); p; p = p->next) {
2188 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2189 continue;
2190 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2191 !p->pack_promisor)
2192 continue;
2193 if ((flags & FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS) &&
2194 p->pack_keep_in_core)
2195 continue;
2196 if ((flags & FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS) &&
2197 p->pack_keep)
2198 continue;
2199 if (open_pack_index(p)) {
2200 pack_errors = 1;
2201 continue;
2202 }
2203 r = for_each_object_in_pack(p, cb, data, flags);
2204 if (r)
2205 break;
2206 }
2207 return r ? r : pack_errors;
2208 }
2209
2210 static int add_promisor_object(const struct object_id *oid,
2211 struct packed_git *pack,
2212 uint32_t pos,
2213 void *set_)
2214 {
2215 struct oidset *set = set_;
2216 struct object *obj = parse_object(the_repository, oid);
2217 if (!obj)
2218 return 1;
2219
2220 oidset_insert(set, oid);
2221
2222 /*
2223 * If this is a tree, commit, or tag, the objects it refers
2224 * to are also promisor objects. (Blobs refer to no objects->)
2225 */
2226 if (obj->type == OBJ_TREE) {
2227 struct tree *tree = (struct tree *)obj;
2228 struct tree_desc desc;
2229 struct name_entry entry;
2230 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2231 /*
2232 * Error messages are given when packs are
2233 * verified, so do not print any here.
2234 */
2235 return 0;
2236 while (tree_entry_gently(&desc, &entry))
2237 oidset_insert(set, &entry.oid);
2238 free_tree_buffer(tree);
2239 } else if (obj->type == OBJ_COMMIT) {
2240 struct commit *commit = (struct commit *) obj;
2241 struct commit_list *parents = commit->parents;
2242
2243 oidset_insert(set, get_commit_tree_oid(commit));
2244 for (; parents; parents = parents->next)
2245 oidset_insert(set, &parents->item->object.oid);
2246 } else if (obj->type == OBJ_TAG) {
2247 struct tag *tag = (struct tag *) obj;
2248 oidset_insert(set, get_tagged_oid(tag));
2249 }
2250 return 0;
2251 }
2252
2253 int is_promisor_object(const struct object_id *oid)
2254 {
2255 static struct oidset promisor_objects;
2256 static int promisor_objects_prepared;
2257
2258 if (!promisor_objects_prepared) {
2259 if (has_promisor_remote()) {
2260 for_each_packed_object(add_promisor_object,
2261 &promisor_objects,
2262 FOR_EACH_OBJECT_PROMISOR_ONLY);
2263 }
2264 promisor_objects_prepared = 1;
2265 }
2266 return oidset_contains(&promisor_objects, oid);
2267 }
git for Debian