?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
repository-version.txt: partialClone casing change
[git/debian.git] / packfile.c
blobc0d7dd93f46dfbc07ea0f17839d4a2d1a9d43622
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)
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 static void close_pack_revindex(struct packed_git *p)
328 {
329 if (!p->revindex_map)
330 return;
331
332 munmap((void *)p->revindex_map, p->revindex_size);
333 p->revindex_map = NULL;
334 p->revindex_data = NULL;
335 }
336
337 static void close_pack_mtimes(struct packed_git *p)
338 {
339 if (!p->mtimes_map)
340 return;
341
342 munmap((void *)p->mtimes_map, p->mtimes_size);
343 p->mtimes_map = NULL;
344 }
345
346 void close_pack(struct packed_git *p)
347 {
348 close_pack_windows(p);
349 close_pack_fd(p);
350 close_pack_index(p);
351 close_pack_revindex(p);
352 close_pack_mtimes(p);
353 oidset_clear(&p->bad_objects);
354 }
355
356 void close_object_store(struct raw_object_store *o)
357 {
358 struct packed_git *p;
359
360 for (p = o->packed_git; p; p = p->next)
361 if (p->do_not_close)
362 BUG("want to close pack marked 'do-not-close'");
363 else
364 close_pack(p);
365
366 if (o->multi_pack_index) {
367 close_midx(o->multi_pack_index);
368 o->multi_pack_index = NULL;
369 }
370
371 close_commit_graph(o);
372 }
373
374 void unlink_pack_path(const char *pack_name, int force_delete)
375 {
376 static const char *exts[] = {".pack", ".idx", ".rev", ".keep", ".bitmap", ".promisor", ".mtimes"};
377 int i;
378 struct strbuf buf = STRBUF_INIT;
379 size_t plen;
380
381 strbuf_addstr(&buf, pack_name);
382 strip_suffix_mem(buf.buf, &buf.len, ".pack");
383 plen = buf.len;
384
385 if (!force_delete) {
386 strbuf_addstr(&buf, ".keep");
387 if (!access(buf.buf, F_OK)) {
388 strbuf_release(&buf);
389 return;
390 }
391 }
392
393 for (i = 0; i < ARRAY_SIZE(exts); i++) {
394 strbuf_setlen(&buf, plen);
395 strbuf_addstr(&buf, exts[i]);
396 unlink(buf.buf);
397 }
398
399 strbuf_release(&buf);
400 }
401
402 /*
403 * The LRU pack is the one with the oldest MRU window, preferring packs
404 * with no used windows, or the oldest mtime if it has no windows allocated.
405 */
406 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
407 {
408 struct pack_window *w, *this_mru_w;
409 int has_windows_inuse = 0;
410
411 /*
412 * Reject this pack if it has windows and the previously selected
413 * one does not. If this pack does not have windows, reject
414 * it if the pack file is newer than the previously selected one.
415 */
416 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
417 return;
418
419 for (w = this_mru_w = p->windows; w; w = w->next) {
420 /*
421 * Reject this pack if any of its windows are in use,
422 * but the previously selected pack did not have any
423 * inuse windows. Otherwise, record that this pack
424 * has windows in use.
425 */
426 if (w->inuse_cnt) {
427 if (*accept_windows_inuse)
428 has_windows_inuse = 1;
429 else
430 return;
431 }
432
433 if (w->last_used > this_mru_w->last_used)
434 this_mru_w = w;
435
436 /*
437 * Reject this pack if it has windows that have been
438 * used more recently than the previously selected pack.
439 * If the previously selected pack had windows inuse and
440 * we have not encountered a window in this pack that is
441 * inuse, skip this check since we prefer a pack with no
442 * inuse windows to one that has inuse windows.
443 */
444 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
445 this_mru_w->last_used > (*mru_w)->last_used)
446 return;
447 }
448
449 /*
450 * Select this pack.
451 */
452 *mru_w = this_mru_w;
453 *lru_p = p;
454 *accept_windows_inuse = has_windows_inuse;
455 }
456
457 static int close_one_pack(void)
458 {
459 struct packed_git *p, *lru_p = NULL;
460 struct pack_window *mru_w = NULL;
461 int accept_windows_inuse = 1;
462
463 for (p = the_repository->objects->packed_git; p; p = p->next) {
464 if (p->pack_fd == -1)
465 continue;
466 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
467 }
468
469 if (lru_p)
470 return close_pack_fd(lru_p);
471
472 return 0;
473 }
474
475 static unsigned int get_max_fd_limit(void)
476 {
477 #ifdef RLIMIT_NOFILE
478 {
479 struct rlimit lim;
480
481 if (!getrlimit(RLIMIT_NOFILE, &lim))
482 return lim.rlim_cur;
483 }
484 #endif
485
486 #ifdef _SC_OPEN_MAX
487 {
488 long open_max = sysconf(_SC_OPEN_MAX);
489 if (0 < open_max)
490 return open_max;
491 /*
492 * Otherwise, we got -1 for one of the two
493 * reasons:
494 *
495 * (1) sysconf() did not understand _SC_OPEN_MAX
496 * and signaled an error with -1; or
497 * (2) sysconf() said there is no limit.
498 *
499 * We _could_ clear errno before calling sysconf() to
500 * tell these two cases apart and return a huge number
501 * in the latter case to let the caller cap it to a
502 * value that is not so selfish, but letting the
503 * fallback OPEN_MAX codepath take care of these cases
504 * is a lot simpler.
505 */
506 }
507 #endif
508
509 #ifdef OPEN_MAX
510 return OPEN_MAX;
511 #else
512 return 1; /* see the caller ;-) */
513 #endif
514 }
515
516 const char *pack_basename(struct packed_git *p)
517 {
518 const char *ret = strrchr(p->pack_name, '/');
519 if (ret)
520 ret = ret + 1; /* skip past slash */
521 else
522 ret = p->pack_name; /* we only have a base */
523 return ret;
524 }
525
526 /*
527 * Do not call this directly as this leaks p->pack_fd on error return;
528 * call open_packed_git() instead.
529 */
530 static int open_packed_git_1(struct packed_git *p)
531 {
532 struct stat st;
533 struct pack_header hdr;
534 unsigned char hash[GIT_MAX_RAWSZ];
535 unsigned char *idx_hash;
536 ssize_t read_result;
537 const unsigned hashsz = the_hash_algo->rawsz;
538
539 if (open_pack_index(p))
540 return error("packfile %s index unavailable", p->pack_name);
541
542 if (!pack_max_fds) {
543 unsigned int max_fds = get_max_fd_limit();
544
545 /* Save 3 for stdin/stdout/stderr, 22 for work */
546 if (25 < max_fds)
547 pack_max_fds = max_fds - 25;
548 else
549 pack_max_fds = 1;
550 }
551
552 while (pack_max_fds <= pack_open_fds && close_one_pack())
553 ; /* nothing */
554
555 p->pack_fd = git_open(p->pack_name);
556 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
557 return -1;
558 pack_open_fds++;
559
560 /* If we created the struct before we had the pack we lack size. */
561 if (!p->pack_size) {
562 if (!S_ISREG(st.st_mode))
563 return error("packfile %s not a regular file", p->pack_name);
564 p->pack_size = st.st_size;
565 } else if (p->pack_size != st.st_size)
566 return error("packfile %s size changed", p->pack_name);
567
568 /* Verify we recognize this pack file format. */
569 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
570 if (read_result < 0)
571 return error_errno("error reading from %s", p->pack_name);
572 if (read_result != sizeof(hdr))
573 return error("file %s is far too short to be a packfile", p->pack_name);
574 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
575 return error("file %s is not a GIT packfile", p->pack_name);
576 if (!pack_version_ok(hdr.hdr_version))
577 return error("packfile %s is version %"PRIu32" and not"
578 " supported (try upgrading GIT to a newer version)",
579 p->pack_name, ntohl(hdr.hdr_version));
580
581 /* Verify the pack matches its index. */
582 if (p->num_objects != ntohl(hdr.hdr_entries))
583 return error("packfile %s claims to have %"PRIu32" objects"
584 " while index indicates %"PRIu32" objects",
585 p->pack_name, ntohl(hdr.hdr_entries),
586 p->num_objects);
587 read_result = pread_in_full(p->pack_fd, hash, hashsz,
588 p->pack_size - hashsz);
589 if (read_result < 0)
590 return error_errno("error reading from %s", p->pack_name);
591 if (read_result != hashsz)
592 return error("packfile %s signature is unavailable", p->pack_name);
593 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
594 if (!hasheq(hash, idx_hash))
595 return error("packfile %s does not match index", p->pack_name);
596 return 0;
597 }
598
599 static int open_packed_git(struct packed_git *p)
600 {
601 if (!open_packed_git_1(p))
602 return 0;
603 close_pack_fd(p);
604 return -1;
605 }
606
607 static int in_window(struct pack_window *win, off_t offset)
608 {
609 /* We must promise at least one full hash after the
610 * offset is available from this window, otherwise the offset
611 * is not actually in this window and a different window (which
612 * has that one hash excess) must be used. This is to support
613 * the object header and delta base parsing routines below.
614 */
615 off_t win_off = win->offset;
616 return win_off <= offset
617 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
618 }
619
620 unsigned char *use_pack(struct packed_git *p,
621 struct pack_window **w_cursor,
622 off_t offset,
623 unsigned long *left)
624 {
625 struct pack_window *win = *w_cursor;
626
627 /* Since packfiles end in a hash of their content and it's
628 * pointless to ask for an offset into the middle of that
629 * hash, and the in_window function above wouldn't match
630 * don't allow an offset too close to the end of the file.
631 */
632 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
633 die("packfile %s cannot be accessed", p->pack_name);
634 if (offset > (p->pack_size - the_hash_algo->rawsz))
635 die("offset beyond end of packfile (truncated pack?)");
636 if (offset < 0)
637 die(_("offset before end of packfile (broken .idx?)"));
638
639 if (!win || !in_window(win, offset)) {
640 if (win)
641 win->inuse_cnt--;
642 for (win = p->windows; win; win = win->next) {
643 if (in_window(win, offset))
644 break;
645 }
646 if (!win) {
647 size_t window_align = packed_git_window_size / 2;
648 off_t len;
649
650 if (p->pack_fd == -1 && open_packed_git(p))
651 die("packfile %s cannot be accessed", p->pack_name);
652
653 CALLOC_ARRAY(win, 1);
654 win->offset = (offset / window_align) * window_align;
655 len = p->pack_size - win->offset;
656 if (len > packed_git_window_size)
657 len = packed_git_window_size;
658 win->len = (size_t)len;
659 pack_mapped += win->len;
660 while (packed_git_limit < pack_mapped
661 && unuse_one_window(p))
662 ; /* nothing */
663 win->base = xmmap_gently(NULL, win->len,
664 PROT_READ, MAP_PRIVATE,
665 p->pack_fd, win->offset);
666 if (win->base == MAP_FAILED)
667 die_errno(_("packfile %s cannot be mapped%s"),
668 p->pack_name, mmap_os_err());
669 if (!win->offset && win->len == p->pack_size
670 && !p->do_not_close)
671 close_pack_fd(p);
672 pack_mmap_calls++;
673 pack_open_windows++;
674 if (pack_mapped > peak_pack_mapped)
675 peak_pack_mapped = pack_mapped;
676 if (pack_open_windows > peak_pack_open_windows)
677 peak_pack_open_windows = pack_open_windows;
678 win->next = p->windows;
679 p->windows = win;
680 }
681 }
682 if (win != *w_cursor) {
683 win->last_used = pack_used_ctr++;
684 win->inuse_cnt++;
685 *w_cursor = win;
686 }
687 offset -= win->offset;
688 if (left)
689 *left = win->len - xsize_t(offset);
690 return win->base + offset;
691 }
692
693 void unuse_pack(struct pack_window **w_cursor)
694 {
695 struct pack_window *w = *w_cursor;
696 if (w) {
697 w->inuse_cnt--;
698 *w_cursor = NULL;
699 }
700 }
701
702 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
703 {
704 struct stat st;
705 size_t alloc;
706 struct packed_git *p;
707
708 /*
709 * Make sure a corresponding .pack file exists and that
710 * the index looks sane.
711 */
712 if (!strip_suffix_mem(path, &path_len, ".idx"))
713 return NULL;
714
715 /*
716 * ".promisor" is long enough to hold any suffix we're adding (and
717 * the use xsnprintf double-checks that)
718 */
719 alloc = st_add3(path_len, strlen(".promisor"), 1);
720 p = alloc_packed_git(alloc);
721 memcpy(p->pack_name, path, path_len);
722
723 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
724 if (!access(p->pack_name, F_OK))
725 p->pack_keep = 1;
726
727 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
728 if (!access(p->pack_name, F_OK))
729 p->pack_promisor = 1;
730
731 xsnprintf(p->pack_name + path_len, alloc - path_len, ".mtimes");
732 if (!access(p->pack_name, F_OK))
733 p->is_cruft = 1;
734
735 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
736 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
737 free(p);
738 return NULL;
739 }
740
741 /* ok, it looks sane as far as we can check without
742 * actually mapping the pack file.
743 */
744 p->pack_size = st.st_size;
745 p->pack_local = local;
746 p->mtime = st.st_mtime;
747 if (path_len < the_hash_algo->hexsz ||
748 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
749 hashclr(p->hash);
750 return p;
751 }
752
753 void install_packed_git(struct repository *r, struct packed_git *pack)
754 {
755 if (pack->pack_fd != -1)
756 pack_open_fds++;
757
758 pack->next = r->objects->packed_git;
759 r->objects->packed_git = pack;
760
761 hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name));
762 hashmap_add(&r->objects->pack_map, &pack->packmap_ent);
763 }
764
765 void (*report_garbage)(unsigned seen_bits, const char *path);
766
767 static void report_helper(const struct string_list *list,
768 int seen_bits, int first, int last)
769 {
770 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
771 return;
772
773 for (; first < last; first++)
774 report_garbage(seen_bits, list->items[first].string);
775 }
776
777 static void report_pack_garbage(struct string_list *list)
778 {
779 int i, baselen = -1, first = 0, seen_bits = 0;
780
781 if (!report_garbage)
782 return;
783
784 string_list_sort(list);
785
786 for (i = 0; i < list->nr; i++) {
787 const char *path = list->items[i].string;
788 if (baselen != -1 &&
789 strncmp(path, list->items[first].string, baselen)) {
790 report_helper(list, seen_bits, first, i);
791 baselen = -1;
792 seen_bits = 0;
793 }
794 if (baselen == -1) {
795 const char *dot = strrchr(path, '.');
796 if (!dot) {
797 report_garbage(PACKDIR_FILE_GARBAGE, path);
798 continue;
799 }
800 baselen = dot - path + 1;
801 first = i;
802 }
803 if (!strcmp(path + baselen, "pack"))
804 seen_bits |= 1;
805 else if (!strcmp(path + baselen, "idx"))
806 seen_bits |= 2;
807 }
808 report_helper(list, seen_bits, first, list->nr);
809 }
810
811 void for_each_file_in_pack_dir(const char *objdir,
812 each_file_in_pack_dir_fn fn,
813 void *data)
814 {
815 struct strbuf path = STRBUF_INIT;
816 size_t dirnamelen;
817 DIR *dir;
818 struct dirent *de;
819
820 strbuf_addstr(&path, objdir);
821 strbuf_addstr(&path, "/pack");
822 dir = opendir(path.buf);
823 if (!dir) {
824 if (errno != ENOENT)
825 error_errno("unable to open object pack directory: %s",
826 path.buf);
827 strbuf_release(&path);
828 return;
829 }
830 strbuf_addch(&path, '/');
831 dirnamelen = path.len;
832 while ((de = readdir_skip_dot_and_dotdot(dir)) != NULL) {
833 strbuf_setlen(&path, dirnamelen);
834 strbuf_addstr(&path, de->d_name);
835
836 fn(path.buf, path.len, de->d_name, data);
837 }
838
839 closedir(dir);
840 strbuf_release(&path);
841 }
842
843 struct prepare_pack_data {
844 struct repository *r;
845 struct string_list *garbage;
846 int local;
847 struct multi_pack_index *m;
848 };
849
850 static void prepare_pack(const char *full_name, size_t full_name_len,
851 const char *file_name, void *_data)
852 {
853 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
854 struct packed_git *p;
855 size_t base_len = full_name_len;
856
857 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
858 !(data->m && midx_contains_pack(data->m, file_name))) {
859 struct hashmap_entry hent;
860 char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name);
861 unsigned int hash = strhash(pack_name);
862 hashmap_entry_init(&hent, hash);
863
864 /* Don't reopen a pack we already have. */
865 if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) {
866 p = add_packed_git(full_name, full_name_len, data->local);
867 if (p)
868 install_packed_git(data->r, p);
869 }
870 free(pack_name);
871 }
872
873 if (!report_garbage)
874 return;
875
876 if (!strcmp(file_name, "multi-pack-index"))
877 return;
878 if (starts_with(file_name, "multi-pack-index") &&
879 (ends_with(file_name, ".bitmap") || ends_with(file_name, ".rev")))
880 return;
881 if (ends_with(file_name, ".idx") ||
882 ends_with(file_name, ".rev") ||
883 ends_with(file_name, ".pack") ||
884 ends_with(file_name, ".bitmap") ||
885 ends_with(file_name, ".keep") ||
886 ends_with(file_name, ".promisor") ||
887 ends_with(file_name, ".mtimes"))
888 string_list_append(data->garbage, full_name);
889 else
890 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
891 }
892
893 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
894 {
895 struct prepare_pack_data data;
896 struct string_list garbage = STRING_LIST_INIT_DUP;
897
898 data.m = r->objects->multi_pack_index;
899
900 /* look for the multi-pack-index for this object directory */
901 while (data.m && strcmp(data.m->object_dir, objdir))
902 data.m = data.m->next;
903
904 data.r = r;
905 data.garbage = &garbage;
906 data.local = local;
907
908 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
909
910 report_pack_garbage(data.garbage);
911 string_list_clear(data.garbage, 0);
912 }
913
914 static void prepare_packed_git(struct repository *r);
915 /*
916 * Give a fast, rough count of the number of objects in the repository. This
917 * ignores loose objects completely. If you have a lot of them, then either
918 * you should repack because your performance will be awful, or they are
919 * all unreachable objects about to be pruned, in which case they're not really
920 * interesting as a measure of repo size in the first place.
921 */
922 unsigned long repo_approximate_object_count(struct repository *r)
923 {
924 if (!r->objects->approximate_object_count_valid) {
925 unsigned long count;
926 struct multi_pack_index *m;
927 struct packed_git *p;
928
929 prepare_packed_git(r);
930 count = 0;
931 for (m = get_multi_pack_index(r); m; m = m->next)
932 count += m->num_objects;
933 for (p = r->objects->packed_git; p; p = p->next) {
934 if (open_pack_index(p))
935 continue;
936 count += p->num_objects;
937 }
938 r->objects->approximate_object_count = count;
939 r->objects->approximate_object_count_valid = 1;
940 }
941 return r->objects->approximate_object_count;
942 }
943
944 DEFINE_LIST_SORT(static, sort_packs, struct packed_git, next);
945
946 static int sort_pack(const struct packed_git *a, const struct packed_git *b)
947 {
948 int st;
949
950 /*
951 * Local packs tend to contain objects specific to our
952 * variant of the project than remote ones. In addition,
953 * remote ones could be on a network mounted filesystem.
954 * Favor local ones for these reasons.
955 */
956 st = a->pack_local - b->pack_local;
957 if (st)
958 return -st;
959
960 /*
961 * Younger packs tend to contain more recent objects,
962 * and more recent objects tend to get accessed more
963 * often.
964 */
965 if (a->mtime < b->mtime)
966 return 1;
967 else if (a->mtime == b->mtime)
968 return 0;
969 return -1;
970 }
971
972 static void rearrange_packed_git(struct repository *r)
973 {
974 sort_packs(&r->objects->packed_git, sort_pack);
975 }
976
977 static void prepare_packed_git_mru(struct repository *r)
978 {
979 struct packed_git *p;
980
981 INIT_LIST_HEAD(&r->objects->packed_git_mru);
982
983 for (p = r->objects->packed_git; p; p = p->next)
984 list_add_tail(&p->mru, &r->objects->packed_git_mru);
985 }
986
987 static void prepare_packed_git(struct repository *r)
988 {
989 struct object_directory *odb;
990
991 if (r->objects->packed_git_initialized)
992 return;
993
994 prepare_alt_odb(r);
995 for (odb = r->objects->odb; odb; odb = odb->next) {
996 int local = (odb == r->objects->odb);
997 prepare_multi_pack_index_one(r, odb->path, local);
998 prepare_packed_git_one(r, odb->path, local);
999 }
1000 rearrange_packed_git(r);
1001
1002 prepare_packed_git_mru(r);
1003 r->objects->packed_git_initialized = 1;
1004 }
1005
1006 void reprepare_packed_git(struct repository *r)
1007 {
1008 struct object_directory *odb;
1009
1010 obj_read_lock();
1011 for (odb = r->objects->odb; odb; odb = odb->next)
1012 odb_clear_loose_cache(odb);
1013
1014 r->objects->approximate_object_count_valid = 0;
1015 r->objects->packed_git_initialized = 0;
1016 prepare_packed_git(r);
1017 obj_read_unlock();
1018 }
1019
1020 struct packed_git *get_packed_git(struct repository *r)
1021 {
1022 prepare_packed_git(r);
1023 return r->objects->packed_git;
1024 }
1025
1026 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1027 {
1028 prepare_packed_git(r);
1029 return r->objects->multi_pack_index;
1030 }
1031
1032 struct multi_pack_index *get_local_multi_pack_index(struct repository *r)
1033 {
1034 struct multi_pack_index *m = get_multi_pack_index(r);
1035
1036 /* no need to iterate; we always put the local one first (if any) */
1037 if (m && m->local)
1038 return m;
1039
1040 return NULL;
1041 }
1042
1043 struct packed_git *get_all_packs(struct repository *r)
1044 {
1045 struct multi_pack_index *m;
1046
1047 prepare_packed_git(r);
1048 for (m = r->objects->multi_pack_index; m; m = m->next) {
1049 uint32_t i;
1050 for (i = 0; i < m->num_packs; i++)
1051 prepare_midx_pack(r, m, i);
1052 }
1053
1054 return r->objects->packed_git;
1055 }
1056
1057 struct list_head *get_packed_git_mru(struct repository *r)
1058 {
1059 prepare_packed_git(r);
1060 return &r->objects->packed_git_mru;
1061 }
1062
1063 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1064 unsigned long len, enum object_type *type, unsigned long *sizep)
1065 {
1066 unsigned shift;
1067 size_t size, c;
1068 unsigned long used = 0;
1069
1070 c = buf[used++];
1071 *type = (c >> 4) & 7;
1072 size = c & 15;
1073 shift = 4;
1074 while (c & 0x80) {
1075 if (len <= used || (bitsizeof(long) - 7) < shift) {
1076 error("bad object header");
1077 size = used = 0;
1078 break;
1079 }
1080 c = buf[used++];
1081 size = st_add(size, st_left_shift(c & 0x7f, shift));
1082 shift += 7;
1083 }
1084 *sizep = cast_size_t_to_ulong(size);
1085 return used;
1086 }
1087
1088 unsigned long get_size_from_delta(struct packed_git *p,
1089 struct pack_window **w_curs,
1090 off_t curpos)
1091 {
1092 const unsigned char *data;
1093 unsigned char delta_head[20], *in;
1094 git_zstream stream;
1095 int st;
1096
1097 memset(&stream, 0, sizeof(stream));
1098 stream.next_out = delta_head;
1099 stream.avail_out = sizeof(delta_head);
1100
1101 git_inflate_init(&stream);
1102 do {
1103 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1104 stream.next_in = in;
1105 /*
1106 * Note: the window section returned by use_pack() must be
1107 * available throughout git_inflate()'s unlocked execution. To
1108 * ensure no other thread will modify the window in the
1109 * meantime, we rely on the packed_window.inuse_cnt. This
1110 * counter is incremented before window reading and checked
1111 * before window disposal.
1112 *
1113 * Other worrying sections could be the call to close_pack_fd(),
1114 * which can close packs even with in-use windows, and to
1115 * reprepare_packed_git(). Regarding the former, mmap doc says:
1116 * "closing the file descriptor does not unmap the region". And
1117 * for the latter, it won't re-open already available packs.
1118 */
1119 obj_read_unlock();
1120 st = git_inflate(&stream, Z_FINISH);
1121 obj_read_lock();
1122 curpos += stream.next_in - in;
1123 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1124 stream.total_out < sizeof(delta_head));
1125 git_inflate_end(&stream);
1126 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1127 error("delta data unpack-initial failed");
1128 return 0;
1129 }
1130
1131 /* Examine the initial part of the delta to figure out
1132 * the result size.
1133 */
1134 data = delta_head;
1135
1136 /* ignore base size */
1137 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1138
1139 /* Read the result size */
1140 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1141 }
1142
1143 int unpack_object_header(struct packed_git *p,
1144 struct pack_window **w_curs,
1145 off_t *curpos,
1146 unsigned long *sizep)
1147 {
1148 unsigned char *base;
1149 unsigned long left;
1150 unsigned long used;
1151 enum object_type type;
1152
1153 /* use_pack() assures us we have [base, base + 20) available
1154 * as a range that we can look at. (Its actually the hash
1155 * size that is assured.) With our object header encoding
1156 * the maximum deflated object size is 2^137, which is just
1157 * insane, so we know won't exceed what we have been given.
1158 */
1159 base = use_pack(p, w_curs, *curpos, &left);
1160 used = unpack_object_header_buffer(base, left, &type, sizep);
1161 if (!used) {
1162 type = OBJ_BAD;
1163 } else
1164 *curpos += used;
1165
1166 return type;
1167 }
1168
1169 void mark_bad_packed_object(struct packed_git *p, const struct object_id *oid)
1170 {
1171 oidset_insert(&p->bad_objects, oid);
1172 }
1173
1174 const struct packed_git *has_packed_and_bad(struct repository *r,
1175 const struct object_id *oid)
1176 {
1177 struct packed_git *p;
1178
1179 for (p = r->objects->packed_git; p; p = p->next)
1180 if (oidset_contains(&p->bad_objects, oid))
1181 return p;
1182 return NULL;
1183 }
1184
1185 off_t get_delta_base(struct packed_git *p,
1186 struct pack_window **w_curs,
1187 off_t *curpos,
1188 enum object_type type,
1189 off_t delta_obj_offset)
1190 {
1191 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1192 off_t base_offset;
1193
1194 /* use_pack() assured us we have [base_info, base_info + 20)
1195 * as a range that we can look at without walking off the
1196 * end of the mapped window. Its actually the hash size
1197 * that is assured. An OFS_DELTA longer than the hash size
1198 * is stupid, as then a REF_DELTA would be smaller to store.
1199 */
1200 if (type == OBJ_OFS_DELTA) {
1201 unsigned used = 0;
1202 unsigned char c = base_info[used++];
1203 base_offset = c & 127;
1204 while (c & 128) {
1205 base_offset += 1;
1206 if (!base_offset || MSB(base_offset, 7))
1207 return 0; /* overflow */
1208 c = base_info[used++];
1209 base_offset = (base_offset << 7) + (c & 127);
1210 }
1211 base_offset = delta_obj_offset - base_offset;
1212 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1213 return 0; /* out of bound */
1214 *curpos += used;
1215 } else if (type == OBJ_REF_DELTA) {
1216 /* The base entry _must_ be in the same pack */
1217 base_offset = find_pack_entry_one(base_info, p);
1218 *curpos += the_hash_algo->rawsz;
1219 } else
1220 die("I am totally screwed");
1221 return base_offset;
1222 }
1223
1224 /*
1225 * Like get_delta_base above, but we return the sha1 instead of the pack
1226 * offset. This means it is cheaper for REF deltas (we do not have to do
1227 * the final object lookup), but more expensive for OFS deltas (we
1228 * have to load the revidx to convert the offset back into a sha1).
1229 */
1230 static int get_delta_base_oid(struct packed_git *p,
1231 struct pack_window **w_curs,
1232 off_t curpos,
1233 struct object_id *oid,
1234 enum object_type type,
1235 off_t delta_obj_offset)
1236 {
1237 if (type == OBJ_REF_DELTA) {
1238 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1239 oidread(oid, base);
1240 return 0;
1241 } else if (type == OBJ_OFS_DELTA) {
1242 uint32_t base_pos;
1243 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1244 type, delta_obj_offset);
1245
1246 if (!base_offset)
1247 return -1;
1248
1249 if (offset_to_pack_pos(p, base_offset, &base_pos) < 0)
1250 return -1;
1251
1252 return nth_packed_object_id(oid, p,
1253 pack_pos_to_index(p, base_pos));
1254 } else
1255 return -1;
1256 }
1257
1258 static int retry_bad_packed_offset(struct repository *r,
1259 struct packed_git *p,
1260 off_t obj_offset)
1261 {
1262 int type;
1263 uint32_t pos;
1264 struct object_id oid;
1265 if (offset_to_pack_pos(p, obj_offset, &pos) < 0)
1266 return OBJ_BAD;
1267 nth_packed_object_id(&oid, p, pack_pos_to_index(p, pos));
1268 mark_bad_packed_object(p, &oid);
1269 type = oid_object_info(r, &oid, NULL);
1270 if (type <= OBJ_NONE)
1271 return OBJ_BAD;
1272 return type;
1273 }
1274
1275 #define POI_STACK_PREALLOC 64
1276
1277 static enum object_type packed_to_object_type(struct repository *r,
1278 struct packed_git *p,
1279 off_t obj_offset,
1280 enum object_type type,
1281 struct pack_window **w_curs,
1282 off_t curpos)
1283 {
1284 off_t small_poi_stack[POI_STACK_PREALLOC];
1285 off_t *poi_stack = small_poi_stack;
1286 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1287
1288 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1289 off_t base_offset;
1290 unsigned long size;
1291 /* Push the object we're going to leave behind */
1292 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1293 poi_stack_alloc = alloc_nr(poi_stack_nr);
1294 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1295 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1296 } else {
1297 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1298 }
1299 poi_stack[poi_stack_nr++] = obj_offset;
1300 /* If parsing the base offset fails, just unwind */
1301 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1302 if (!base_offset)
1303 goto unwind;
1304 curpos = obj_offset = base_offset;
1305 type = unpack_object_header(p, w_curs, &curpos, &size);
1306 if (type <= OBJ_NONE) {
1307 /* If getting the base itself fails, we first
1308 * retry the base, otherwise unwind */
1309 type = retry_bad_packed_offset(r, p, base_offset);
1310 if (type > OBJ_NONE)
1311 goto out;
1312 goto unwind;
1313 }
1314 }
1315
1316 switch (type) {
1317 case OBJ_BAD:
1318 case OBJ_COMMIT:
1319 case OBJ_TREE:
1320 case OBJ_BLOB:
1321 case OBJ_TAG:
1322 break;
1323 default:
1324 error("unknown object type %i at offset %"PRIuMAX" in %s",
1325 type, (uintmax_t)obj_offset, p->pack_name);
1326 type = OBJ_BAD;
1327 }
1328
1329 out:
1330 if (poi_stack != small_poi_stack)
1331 free(poi_stack);
1332 return type;
1333
1334 unwind:
1335 while (poi_stack_nr) {
1336 obj_offset = poi_stack[--poi_stack_nr];
1337 type = retry_bad_packed_offset(r, p, obj_offset);
1338 if (type > OBJ_NONE)
1339 goto out;
1340 }
1341 type = OBJ_BAD;
1342 goto out;
1343 }
1344
1345 static struct hashmap delta_base_cache;
1346 static size_t delta_base_cached;
1347
1348 static LIST_HEAD(delta_base_cache_lru);
1349
1350 struct delta_base_cache_key {
1351 struct packed_git *p;
1352 off_t base_offset;
1353 };
1354
1355 struct delta_base_cache_entry {
1356 struct hashmap_entry ent;
1357 struct delta_base_cache_key key;
1358 struct list_head lru;
1359 void *data;
1360 unsigned long size;
1361 enum object_type type;
1362 };
1363
1364 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1365 {
1366 unsigned int hash;
1367
1368 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1369 hash += (hash >> 8) + (hash >> 16);
1370 return hash;
1371 }
1372
1373 static struct delta_base_cache_entry *
1374 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1375 {
1376 struct hashmap_entry entry, *e;
1377 struct delta_base_cache_key key;
1378
1379 if (!delta_base_cache.cmpfn)
1380 return NULL;
1381
1382 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1383 key.p = p;
1384 key.base_offset = base_offset;
1385 e = hashmap_get(&delta_base_cache, &entry, &key);
1386 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1387 }
1388
1389 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1390 const struct delta_base_cache_key *b)
1391 {
1392 return a->p == b->p && a->base_offset == b->base_offset;
1393 }
1394
1395 static int delta_base_cache_hash_cmp(const void *cmp_data UNUSED,
1396 const struct hashmap_entry *va,
1397 const struct hashmap_entry *vb,
1398 const void *vkey)
1399 {
1400 const struct delta_base_cache_entry *a, *b;
1401 const struct delta_base_cache_key *key = vkey;
1402
1403 a = container_of(va, const struct delta_base_cache_entry, ent);
1404 b = container_of(vb, const struct delta_base_cache_entry, ent);
1405
1406 if (key)
1407 return !delta_base_cache_key_eq(&a->key, key);
1408 else
1409 return !delta_base_cache_key_eq(&a->key, &b->key);
1410 }
1411
1412 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1413 {
1414 return !!get_delta_base_cache_entry(p, base_offset);
1415 }
1416
1417 /*
1418 * Remove the entry from the cache, but do _not_ free the associated
1419 * entry data. The caller takes ownership of the "data" buffer, and
1420 * should copy out any fields it wants before detaching.
1421 */
1422 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1423 {
1424 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1425 list_del(&ent->lru);
1426 delta_base_cached -= ent->size;
1427 free(ent);
1428 }
1429
1430 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1431 off_t base_offset, unsigned long *base_size,
1432 enum object_type *type)
1433 {
1434 struct delta_base_cache_entry *ent;
1435
1436 ent = get_delta_base_cache_entry(p, base_offset);
1437 if (!ent)
1438 return unpack_entry(r, p, base_offset, type, base_size);
1439
1440 if (type)
1441 *type = ent->type;
1442 if (base_size)
1443 *base_size = ent->size;
1444 return xmemdupz(ent->data, ent->size);
1445 }
1446
1447 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1448 {
1449 free(ent->data);
1450 detach_delta_base_cache_entry(ent);
1451 }
1452
1453 void clear_delta_base_cache(void)
1454 {
1455 struct list_head *lru, *tmp;
1456 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1457 struct delta_base_cache_entry *entry =
1458 list_entry(lru, struct delta_base_cache_entry, lru);
1459 release_delta_base_cache(entry);
1460 }
1461 }
1462
1463 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1464 void *base, unsigned long base_size, enum object_type type)
1465 {
1466 struct delta_base_cache_entry *ent;
1467 struct list_head *lru, *tmp;
1468
1469 /*
1470 * Check required to avoid redundant entries when more than one thread
1471 * is unpacking the same object, in unpack_entry() (since its phases I
1472 * and III might run concurrently across multiple threads).
1473 */
1474 if (in_delta_base_cache(p, base_offset)) {
1475 free(base);
1476 return;
1477 }
1478
1479 delta_base_cached += base_size;
1480
1481 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1482 struct delta_base_cache_entry *f =
1483 list_entry(lru, struct delta_base_cache_entry, lru);
1484 if (delta_base_cached <= delta_base_cache_limit)
1485 break;
1486 release_delta_base_cache(f);
1487 }
1488
1489 ent = xmalloc(sizeof(*ent));
1490 ent->key.p = p;
1491 ent->key.base_offset = base_offset;
1492 ent->type = type;
1493 ent->data = base;
1494 ent->size = base_size;
1495 list_add_tail(&ent->lru, &delta_base_cache_lru);
1496
1497 if (!delta_base_cache.cmpfn)
1498 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1499 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1500 hashmap_add(&delta_base_cache, &ent->ent);
1501 }
1502
1503 int packed_object_info(struct repository *r, struct packed_git *p,
1504 off_t obj_offset, struct object_info *oi)
1505 {
1506 struct pack_window *w_curs = NULL;
1507 unsigned long size;
1508 off_t curpos = obj_offset;
1509 enum object_type type;
1510
1511 /*
1512 * We always get the representation type, but only convert it to
1513 * a "real" type later if the caller is interested.
1514 */
1515 if (oi->contentp) {
1516 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1517 &type);
1518 if (!*oi->contentp)
1519 type = OBJ_BAD;
1520 } else {
1521 type = unpack_object_header(p, &w_curs, &curpos, &size);
1522 }
1523
1524 if (!oi->contentp && oi->sizep) {
1525 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1526 off_t tmp_pos = curpos;
1527 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1528 type, obj_offset);
1529 if (!base_offset) {
1530 type = OBJ_BAD;
1531 goto out;
1532 }
1533 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1534 if (*oi->sizep == 0) {
1535 type = OBJ_BAD;
1536 goto out;
1537 }
1538 } else {
1539 *oi->sizep = size;
1540 }
1541 }
1542
1543 if (oi->disk_sizep) {
1544 uint32_t pos;
1545 if (offset_to_pack_pos(p, obj_offset, &pos) < 0) {
1546 error("could not find object at offset %"PRIuMAX" "
1547 "in pack %s", (uintmax_t)obj_offset, p->pack_name);
1548 type = OBJ_BAD;
1549 goto out;
1550 }
1551
1552 *oi->disk_sizep = pack_pos_to_offset(p, pos + 1) - obj_offset;
1553 }
1554
1555 if (oi->typep || oi->type_name) {
1556 enum object_type ptot;
1557 ptot = packed_to_object_type(r, p, obj_offset,
1558 type, &w_curs, curpos);
1559 if (oi->typep)
1560 *oi->typep = ptot;
1561 if (oi->type_name) {
1562 const char *tn = type_name(ptot);
1563 if (tn)
1564 strbuf_addstr(oi->type_name, tn);
1565 }
1566 if (ptot < 0) {
1567 type = OBJ_BAD;
1568 goto out;
1569 }
1570 }
1571
1572 if (oi->delta_base_oid) {
1573 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1574 if (get_delta_base_oid(p, &w_curs, curpos,
1575 oi->delta_base_oid,
1576 type, obj_offset) < 0) {
1577 type = OBJ_BAD;
1578 goto out;
1579 }
1580 } else
1581 oidclr(oi->delta_base_oid);
1582 }
1583
1584 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1585 OI_PACKED;
1586
1587 out:
1588 unuse_pack(&w_curs);
1589 return type;
1590 }
1591
1592 static void *unpack_compressed_entry(struct packed_git *p,
1593 struct pack_window **w_curs,
1594 off_t curpos,
1595 unsigned long size)
1596 {
1597 int st;
1598 git_zstream stream;
1599 unsigned char *buffer, *in;
1600
1601 buffer = xmallocz_gently(size);
1602 if (!buffer)
1603 return NULL;
1604 memset(&stream, 0, sizeof(stream));
1605 stream.next_out = buffer;
1606 stream.avail_out = size + 1;
1607
1608 git_inflate_init(&stream);
1609 do {
1610 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1611 stream.next_in = in;
1612 /*
1613 * Note: we must ensure the window section returned by
1614 * use_pack() will be available throughout git_inflate()'s
1615 * unlocked execution. Please refer to the comment at
1616 * get_size_from_delta() to see how this is done.
1617 */
1618 obj_read_unlock();
1619 st = git_inflate(&stream, Z_FINISH);
1620 obj_read_lock();
1621 if (!stream.avail_out)
1622 break; /* the payload is larger than it should be */
1623 curpos += stream.next_in - in;
1624 } while (st == Z_OK || st == Z_BUF_ERROR);
1625 git_inflate_end(&stream);
1626 if ((st != Z_STREAM_END) || stream.total_out != size) {
1627 free(buffer);
1628 return NULL;
1629 }
1630
1631 /* versions of zlib can clobber unconsumed portion of outbuf */
1632 buffer[size] = '\0';
1633
1634 return buffer;
1635 }
1636
1637 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1638 {
1639 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1640 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1641 p->pack_name, (uintmax_t)obj_offset);
1642 }
1643
1644 int do_check_packed_object_crc;
1645
1646 #define UNPACK_ENTRY_STACK_PREALLOC 64
1647 struct unpack_entry_stack_ent {
1648 off_t obj_offset;
1649 off_t curpos;
1650 unsigned long size;
1651 };
1652
1653 static void *read_object(struct repository *r,
1654 const struct object_id *oid,
1655 enum object_type *type,
1656 unsigned long *size)
1657 {
1658 struct object_info oi = OBJECT_INFO_INIT;
1659 void *content;
1660 oi.typep = type;
1661 oi.sizep = size;
1662 oi.contentp = &content;
1663
1664 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1665 return NULL;
1666 return content;
1667 }
1668
1669 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1670 enum object_type *final_type, unsigned long *final_size)
1671 {
1672 struct pack_window *w_curs = NULL;
1673 off_t curpos = obj_offset;
1674 void *data = NULL;
1675 unsigned long size;
1676 enum object_type type;
1677 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1678 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1679 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1680 int base_from_cache = 0;
1681
1682 write_pack_access_log(p, obj_offset);
1683
1684 /* PHASE 1: drill down to the innermost base object */
1685 for (;;) {
1686 off_t base_offset;
1687 int i;
1688 struct delta_base_cache_entry *ent;
1689
1690 ent = get_delta_base_cache_entry(p, curpos);
1691 if (ent) {
1692 type = ent->type;
1693 data = ent->data;
1694 size = ent->size;
1695 detach_delta_base_cache_entry(ent);
1696 base_from_cache = 1;
1697 break;
1698 }
1699
1700 if (do_check_packed_object_crc && p->index_version > 1) {
1701 uint32_t pack_pos, index_pos;
1702 off_t len;
1703
1704 if (offset_to_pack_pos(p, obj_offset, &pack_pos) < 0) {
1705 error("could not find object at offset %"PRIuMAX" in pack %s",
1706 (uintmax_t)obj_offset, p->pack_name);
1707 data = NULL;
1708 goto out;
1709 }
1710
1711 len = pack_pos_to_offset(p, pack_pos + 1) - obj_offset;
1712 index_pos = pack_pos_to_index(p, pack_pos);
1713 if (check_pack_crc(p, &w_curs, obj_offset, len, index_pos)) {
1714 struct object_id oid;
1715 nth_packed_object_id(&oid, p, index_pos);
1716 error("bad packed object CRC for %s",
1717 oid_to_hex(&oid));
1718 mark_bad_packed_object(p, &oid);
1719 data = NULL;
1720 goto out;
1721 }
1722 }
1723
1724 type = unpack_object_header(p, &w_curs, &curpos, &size);
1725 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1726 break;
1727
1728 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1729 if (!base_offset) {
1730 error("failed to validate delta base reference "
1731 "at offset %"PRIuMAX" from %s",
1732 (uintmax_t)curpos, p->pack_name);
1733 /* bail to phase 2, in hopes of recovery */
1734 data = NULL;
1735 break;
1736 }
1737
1738 /* push object, proceed to base */
1739 if (delta_stack_nr >= delta_stack_alloc
1740 && delta_stack == small_delta_stack) {
1741 delta_stack_alloc = alloc_nr(delta_stack_nr);
1742 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1743 COPY_ARRAY(delta_stack, small_delta_stack,
1744 delta_stack_nr);
1745 } else {
1746 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1747 }
1748 i = delta_stack_nr++;
1749 delta_stack[i].obj_offset = obj_offset;
1750 delta_stack[i].curpos = curpos;
1751 delta_stack[i].size = size;
1752
1753 curpos = obj_offset = base_offset;
1754 }
1755
1756 /* PHASE 2: handle the base */
1757 switch (type) {
1758 case OBJ_OFS_DELTA:
1759 case OBJ_REF_DELTA:
1760 if (data)
1761 BUG("unpack_entry: left loop at a valid delta");
1762 break;
1763 case OBJ_COMMIT:
1764 case OBJ_TREE:
1765 case OBJ_BLOB:
1766 case OBJ_TAG:
1767 if (!base_from_cache)
1768 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1769 break;
1770 default:
1771 data = NULL;
1772 error("unknown object type %i at offset %"PRIuMAX" in %s",
1773 type, (uintmax_t)obj_offset, p->pack_name);
1774 }
1775
1776 /* PHASE 3: apply deltas in order */
1777
1778 /* invariants:
1779 * 'data' holds the base data, or NULL if there was corruption
1780 */
1781 while (delta_stack_nr) {
1782 void *delta_data;
1783 void *base = data;
1784 void *external_base = NULL;
1785 unsigned long delta_size, base_size = size;
1786 int i;
1787 off_t base_obj_offset = obj_offset;
1788
1789 data = NULL;
1790
1791 if (!base) {
1792 /*
1793 * We're probably in deep shit, but let's try to fetch
1794 * the required base anyway from another pack or loose.
1795 * This is costly but should happen only in the presence
1796 * of a corrupted pack, and is better than failing outright.
1797 */
1798 uint32_t pos;
1799 struct object_id base_oid;
1800 if (!(offset_to_pack_pos(p, obj_offset, &pos))) {
1801 nth_packed_object_id(&base_oid, p,
1802 pack_pos_to_index(p, pos));
1803 error("failed to read delta base object %s"
1804 " at offset %"PRIuMAX" from %s",
1805 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1806 p->pack_name);
1807 mark_bad_packed_object(p, &base_oid);
1808 base = read_object(r, &base_oid, &type, &base_size);
1809 external_base = base;
1810 }
1811 }
1812
1813 i = --delta_stack_nr;
1814 obj_offset = delta_stack[i].obj_offset;
1815 curpos = delta_stack[i].curpos;
1816 delta_size = delta_stack[i].size;
1817
1818 if (!base)
1819 continue;
1820
1821 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1822
1823 if (!delta_data) {
1824 error("failed to unpack compressed delta "
1825 "at offset %"PRIuMAX" from %s",
1826 (uintmax_t)curpos, p->pack_name);
1827 data = NULL;
1828 } else {
1829 data = patch_delta(base, base_size, delta_data,
1830 delta_size, &size);
1831
1832 /*
1833 * We could not apply the delta; warn the user, but
1834 * keep going. Our failure will be noticed either in
1835 * the next iteration of the loop, or if this is the
1836 * final delta, in the caller when we return NULL.
1837 * Those code paths will take care of making a more
1838 * explicit warning and retrying with another copy of
1839 * the object.
1840 */
1841 if (!data)
1842 error("failed to apply delta");
1843 }
1844
1845 /*
1846 * We delay adding `base` to the cache until the end of the loop
1847 * because unpack_compressed_entry() momentarily releases the
1848 * obj_read_mutex, giving another thread the chance to access
1849 * the cache. Therefore, if `base` was already there, this other
1850 * thread could free() it (e.g. to make space for another entry)
1851 * before we are done using it.
1852 */
1853 if (!external_base)
1854 add_delta_base_cache(p, base_obj_offset, base, base_size, type);
1855
1856 free(delta_data);
1857 free(external_base);
1858 }
1859
1860 if (final_type)
1861 *final_type = type;
1862 if (final_size)
1863 *final_size = size;
1864
1865 out:
1866 unuse_pack(&w_curs);
1867
1868 if (delta_stack != small_delta_stack)
1869 free(delta_stack);
1870
1871 return data;
1872 }
1873
1874 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1875 {
1876 const unsigned char *index_fanout = p->index_data;
1877 const unsigned char *index_lookup;
1878 const unsigned int hashsz = the_hash_algo->rawsz;
1879 int index_lookup_width;
1880
1881 if (!index_fanout)
1882 BUG("bsearch_pack called without a valid pack-index");
1883
1884 index_lookup = index_fanout + 4 * 256;
1885 if (p->index_version == 1) {
1886 index_lookup_width = hashsz + 4;
1887 index_lookup += 4;
1888 } else {
1889 index_lookup_width = hashsz;
1890 index_fanout += 8;
1891 index_lookup += 8;
1892 }
1893
1894 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1895 index_lookup, index_lookup_width, result);
1896 }
1897
1898 int nth_packed_object_id(struct object_id *oid,
1899 struct packed_git *p,
1900 uint32_t n)
1901 {
1902 const unsigned char *index = p->index_data;
1903 const unsigned int hashsz = the_hash_algo->rawsz;
1904 if (!index) {
1905 if (open_pack_index(p))
1906 return -1;
1907 index = p->index_data;
1908 }
1909 if (n >= p->num_objects)
1910 return -1;
1911 index += 4 * 256;
1912 if (p->index_version == 1) {
1913 oidread(oid, index + (hashsz + 4) * n + 4);
1914 } else {
1915 index += 8;
1916 oidread(oid, index + hashsz * n);
1917 }
1918 return 0;
1919 }
1920
1921 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1922 {
1923 const unsigned char *ptr = vptr;
1924 const unsigned char *start = p->index_data;
1925 const unsigned char *end = start + p->index_size;
1926 if (ptr < start)
1927 die(_("offset before start of pack index for %s (corrupt index?)"),
1928 p->pack_name);
1929 /* No need to check for underflow; .idx files must be at least 8 bytes */
1930 if (ptr >= end - 8)
1931 die(_("offset beyond end of pack index for %s (truncated index?)"),
1932 p->pack_name);
1933 }
1934
1935 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1936 {
1937 const unsigned char *index = p->index_data;
1938 const unsigned int hashsz = the_hash_algo->rawsz;
1939 index += 4 * 256;
1940 if (p->index_version == 1) {
1941 return ntohl(*((uint32_t *)(index + (hashsz + 4) * (size_t)n)));
1942 } else {
1943 uint32_t off;
1944 index += 8 + (size_t)p->num_objects * (hashsz + 4);
1945 off = ntohl(*((uint32_t *)(index + 4 * n)));
1946 if (!(off & 0x80000000))
1947 return off;
1948 index += (size_t)p->num_objects * 4 + (off & 0x7fffffff) * 8;
1949 check_pack_index_ptr(p, index);
1950 return get_be64(index);
1951 }
1952 }
1953
1954 off_t find_pack_entry_one(const unsigned char *sha1,
1955 struct packed_git *p)
1956 {
1957 const unsigned char *index = p->index_data;
1958 struct object_id oid;
1959 uint32_t result;
1960
1961 if (!index) {
1962 if (open_pack_index(p))
1963 return 0;
1964 }
1965
1966 hashcpy(oid.hash, sha1);
1967 if (bsearch_pack(&oid, p, &result))
1968 return nth_packed_object_offset(p, result);
1969 return 0;
1970 }
1971
1972 int is_pack_valid(struct packed_git *p)
1973 {
1974 /* An already open pack is known to be valid. */
1975 if (p->pack_fd != -1)
1976 return 1;
1977
1978 /* If the pack has one window completely covering the
1979 * file size, the pack is known to be valid even if
1980 * the descriptor is not currently open.
1981 */
1982 if (p->windows) {
1983 struct pack_window *w = p->windows;
1984
1985 if (!w->offset && w->len == p->pack_size)
1986 return 1;
1987 }
1988
1989 /* Force the pack to open to prove its valid. */
1990 return !open_packed_git(p);
1991 }
1992
1993 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1994 struct packed_git *packs)
1995 {
1996 struct packed_git *p;
1997
1998 for (p = packs; p; p = p->next) {
1999 if (find_pack_entry_one(sha1, p))
2000 return p;
2001 }
2002 return NULL;
2003
2004 }
2005
2006 static int fill_pack_entry(const struct object_id *oid,
2007 struct pack_entry *e,
2008 struct packed_git *p)
2009 {
2010 off_t offset;
2011
2012 if (oidset_size(&p->bad_objects) &&
2013 oidset_contains(&p->bad_objects, oid))
2014 return 0;
2015
2016 offset = find_pack_entry_one(oid->hash, p);
2017 if (!offset)
2018 return 0;
2019
2020 /*
2021 * We are about to tell the caller where they can locate the
2022 * requested object. We better make sure the packfile is
2023 * still here and can be accessed before supplying that
2024 * answer, as it may have been deleted since the index was
2025 * loaded!
2026 */
2027 if (!is_pack_valid(p))
2028 return 0;
2029 e->offset = offset;
2030 e->p = p;
2031 return 1;
2032 }
2033
2034 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2035 {
2036 struct list_head *pos;
2037 struct multi_pack_index *m;
2038
2039 prepare_packed_git(r);
2040 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2041 return 0;
2042
2043 for (m = r->objects->multi_pack_index; m; m = m->next) {
2044 if (fill_midx_entry(r, oid, e, m))
2045 return 1;
2046 }
2047
2048 list_for_each(pos, &r->objects->packed_git_mru) {
2049 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2050 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2051 list_move(&p->mru, &r->objects->packed_git_mru);
2052 return 1;
2053 }
2054 }
2055 return 0;
2056 }
2057
2058 static void maybe_invalidate_kept_pack_cache(struct repository *r,
2059 unsigned flags)
2060 {
2061 if (!r->objects->kept_pack_cache.packs)
2062 return;
2063 if (r->objects->kept_pack_cache.flags == flags)
2064 return;
2065 FREE_AND_NULL(r->objects->kept_pack_cache.packs);
2066 r->objects->kept_pack_cache.flags = 0;
2067 }
2068
2069 static struct packed_git **kept_pack_cache(struct repository *r, unsigned flags)
2070 {
2071 maybe_invalidate_kept_pack_cache(r, flags);
2072
2073 if (!r->objects->kept_pack_cache.packs) {
2074 struct packed_git **packs = NULL;
2075 size_t nr = 0, alloc = 0;
2076 struct packed_git *p;
2077
2078 /*
2079 * We want "all" packs here, because we need to cover ones that
2080 * are used by a midx, as well. We need to look in every one of
2081 * them (instead of the midx itself) to cover duplicates. It's
2082 * possible that an object is found in two packs that the midx
2083 * covers, one kept and one not kept, but the midx returns only
2084 * the non-kept version.
2085 */
2086 for (p = get_all_packs(r); p; p = p->next) {
2087 if ((p->pack_keep && (flags & ON_DISK_KEEP_PACKS)) ||
2088 (p->pack_keep_in_core && (flags & IN_CORE_KEEP_PACKS))) {
2089 ALLOC_GROW(packs, nr + 1, alloc);
2090 packs[nr++] = p;
2091 }
2092 }
2093 ALLOC_GROW(packs, nr + 1, alloc);
2094 packs[nr] = NULL;
2095
2096 r->objects->kept_pack_cache.packs = packs;
2097 r->objects->kept_pack_cache.flags = flags;
2098 }
2099
2100 return r->objects->kept_pack_cache.packs;
2101 }
2102
2103 int find_kept_pack_entry(struct repository *r,
2104 const struct object_id *oid,
2105 unsigned flags,
2106 struct pack_entry *e)
2107 {
2108 struct packed_git **cache;
2109
2110 for (cache = kept_pack_cache(r, flags); *cache; cache++) {
2111 struct packed_git *p = *cache;
2112 if (fill_pack_entry(oid, e, p))
2113 return 1;
2114 }
2115
2116 return 0;
2117 }
2118
2119 int has_object_pack(const struct object_id *oid)
2120 {
2121 struct pack_entry e;
2122 return find_pack_entry(the_repository, oid, &e);
2123 }
2124
2125 int has_object_kept_pack(const struct object_id *oid, unsigned flags)
2126 {
2127 struct pack_entry e;
2128 return find_kept_pack_entry(the_repository, oid, flags, &e);
2129 }
2130
2131 int has_pack_index(const unsigned char *sha1)
2132 {
2133 struct stat st;
2134 if (stat(sha1_pack_index_name(sha1), &st))
2135 return 0;
2136 return 1;
2137 }
2138
2139 int for_each_object_in_pack(struct packed_git *p,
2140 each_packed_object_fn cb, void *data,
2141 enum for_each_object_flags flags)
2142 {
2143 uint32_t i;
2144 int r = 0;
2145
2146 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2147 if (load_pack_revindex(p))
2148 return -1;
2149 }
2150
2151 for (i = 0; i < p->num_objects; i++) {
2152 uint32_t index_pos;
2153 struct object_id oid;
2154
2155 /*
2156 * We are iterating "i" from 0 up to num_objects, but its
2157 * meaning may be different, depending on the requested output
2158 * order:
2159 *
2160 * - in object-name order, it is the same as the index order
2161 * used by nth_packed_object_id(), so we can pass it
2162 * directly
2163 *
2164 * - in pack-order, it is pack position, which we must
2165 * convert to an index position in order to get the oid.
2166 */
2167 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2168 index_pos = pack_pos_to_index(p, i);
2169 else
2170 index_pos = i;
2171
2172 if (nth_packed_object_id(&oid, p, index_pos) < 0)
2173 return error("unable to get sha1 of object %u in %s",
2174 index_pos, p->pack_name);
2175
2176 r = cb(&oid, p, index_pos, data);
2177 if (r)
2178 break;
2179 }
2180 return r;
2181 }
2182
2183 int for_each_packed_object(each_packed_object_fn cb, void *data,
2184 enum for_each_object_flags flags)
2185 {
2186 struct packed_git *p;
2187 int r = 0;
2188 int pack_errors = 0;
2189
2190 prepare_packed_git(the_repository);
2191 for (p = get_all_packs(the_repository); p; p = p->next) {
2192 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2193 continue;
2194 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2195 !p->pack_promisor)
2196 continue;
2197 if ((flags & FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS) &&
2198 p->pack_keep_in_core)
2199 continue;
2200 if ((flags & FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS) &&
2201 p->pack_keep)
2202 continue;
2203 if (open_pack_index(p)) {
2204 pack_errors = 1;
2205 continue;
2206 }
2207 r = for_each_object_in_pack(p, cb, data, flags);
2208 if (r)
2209 break;
2210 }
2211 return r ? r : pack_errors;
2212 }
2213
2214 static int add_promisor_object(const struct object_id *oid,
2215 struct packed_git *pack,
2216 uint32_t pos,
2217 void *set_)
2218 {
2219 struct oidset *set = set_;
2220 struct object *obj;
2221 int we_parsed_object;
2222
2223 obj = lookup_object(the_repository, oid);
2224 if (obj && obj->parsed) {
2225 we_parsed_object = 0;
2226 } else {
2227 we_parsed_object = 1;
2228 obj = parse_object(the_repository, oid);
2229 }
2230
2231 if (!obj)
2232 return 1;
2233
2234 oidset_insert(set, oid);
2235
2236 /*
2237 * If this is a tree, commit, or tag, the objects it refers
2238 * to are also promisor objects. (Blobs refer to no objects->)
2239 */
2240 if (obj->type == OBJ_TREE) {
2241 struct tree *tree = (struct tree *)obj;
2242 struct tree_desc desc;
2243 struct name_entry entry;
2244 if (init_tree_desc_gently(&desc, tree->buffer, tree->size, 0))
2245 /*
2246 * Error messages are given when packs are
2247 * verified, so do not print any here.
2248 */
2249 return 0;
2250 while (tree_entry_gently(&desc, &entry))
2251 oidset_insert(set, &entry.oid);
2252 if (we_parsed_object)
2253 free_tree_buffer(tree);
2254 } else if (obj->type == OBJ_COMMIT) {
2255 struct commit *commit = (struct commit *) obj;
2256 struct commit_list *parents = commit->parents;
2257
2258 oidset_insert(set, get_commit_tree_oid(commit));
2259 for (; parents; parents = parents->next)
2260 oidset_insert(set, &parents->item->object.oid);
2261 } else if (obj->type == OBJ_TAG) {
2262 struct tag *tag = (struct tag *) obj;
2263 oidset_insert(set, get_tagged_oid(tag));
2264 }
2265 return 0;
2266 }
2267
2268 int is_promisor_object(const struct object_id *oid)
2269 {
2270 static struct oidset promisor_objects;
2271 static int promisor_objects_prepared;
2272
2273 if (!promisor_objects_prepared) {
2274 if (has_promisor_remote()) {
2275 for_each_packed_object(add_promisor_object,
2276 &promisor_objects,
2277 FOR_EACH_OBJECT_PROMISOR_ONLY |
2278 FOR_EACH_OBJECT_PACK_ORDER);
2279 }
2280 promisor_objects_prepared = 1;
2281 }
2282 return oidset_contains(&promisor_objects, oid);
2283 }
git for Debian