?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'v2.37.2' into debian-sid
[git/debian.git] / packfile.c
blob8e812a84a3c3395de964bae782d9fe74ae169737
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 static void *get_next_packed_git(const void *p)
945 {
946 return ((const struct packed_git *)p)->next;
947 }
948
949 static void set_next_packed_git(void *p, void *next)
950 {
951 ((struct packed_git *)p)->next = next;
952 }
953
954 static int sort_pack(const void *a_, const void *b_)
955 {
956 const struct packed_git *a = a_;
957 const struct packed_git *b = b_;
958 int st;
959
960 /*
961 * Local packs tend to contain objects specific to our
962 * variant of the project than remote ones. In addition,
963 * remote ones could be on a network mounted filesystem.
964 * Favor local ones for these reasons.
965 */
966 st = a->pack_local - b->pack_local;
967 if (st)
968 return -st;
969
970 /*
971 * Younger packs tend to contain more recent objects,
972 * and more recent objects tend to get accessed more
973 * often.
974 */
975 if (a->mtime < b->mtime)
976 return 1;
977 else if (a->mtime == b->mtime)
978 return 0;
979 return -1;
980 }
981
982 static void rearrange_packed_git(struct repository *r)
983 {
984 r->objects->packed_git = llist_mergesort(
985 r->objects->packed_git, get_next_packed_git,
986 set_next_packed_git, sort_pack);
987 }
988
989 static void prepare_packed_git_mru(struct repository *r)
990 {
991 struct packed_git *p;
992
993 INIT_LIST_HEAD(&r->objects->packed_git_mru);
994
995 for (p = r->objects->packed_git; p; p = p->next)
996 list_add_tail(&p->mru, &r->objects->packed_git_mru);
997 }
998
999 static void prepare_packed_git(struct repository *r)
1000 {
1001 struct object_directory *odb;
1002
1003 if (r->objects->packed_git_initialized)
1004 return;
1005
1006 prepare_alt_odb(r);
1007 for (odb = r->objects->odb; odb; odb = odb->next) {
1008 int local = (odb == r->objects->odb);
1009 prepare_multi_pack_index_one(r, odb->path, local);
1010 prepare_packed_git_one(r, odb->path, local);
1011 }
1012 rearrange_packed_git(r);
1013
1014 prepare_packed_git_mru(r);
1015 r->objects->packed_git_initialized = 1;
1016 }
1017
1018 void reprepare_packed_git(struct repository *r)
1019 {
1020 struct object_directory *odb;
1021
1022 obj_read_lock();
1023 for (odb = r->objects->odb; odb; odb = odb->next)
1024 odb_clear_loose_cache(odb);
1025
1026 r->objects->approximate_object_count_valid = 0;
1027 r->objects->packed_git_initialized = 0;
1028 prepare_packed_git(r);
1029 obj_read_unlock();
1030 }
1031
1032 struct packed_git *get_packed_git(struct repository *r)
1033 {
1034 prepare_packed_git(r);
1035 return r->objects->packed_git;
1036 }
1037
1038 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1039 {
1040 prepare_packed_git(r);
1041 return r->objects->multi_pack_index;
1042 }
1043
1044 struct multi_pack_index *get_local_multi_pack_index(struct repository *r)
1045 {
1046 struct multi_pack_index *m = get_multi_pack_index(r);
1047
1048 /* no need to iterate; we always put the local one first (if any) */
1049 if (m && m->local)
1050 return m;
1051
1052 return NULL;
1053 }
1054
1055 struct packed_git *get_all_packs(struct repository *r)
1056 {
1057 struct multi_pack_index *m;
1058
1059 prepare_packed_git(r);
1060 for (m = r->objects->multi_pack_index; m; m = m->next) {
1061 uint32_t i;
1062 for (i = 0; i < m->num_packs; i++)
1063 prepare_midx_pack(r, m, i);
1064 }
1065
1066 return r->objects->packed_git;
1067 }
1068
1069 struct list_head *get_packed_git_mru(struct repository *r)
1070 {
1071 prepare_packed_git(r);
1072 return &r->objects->packed_git_mru;
1073 }
1074
1075 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1076 unsigned long len, enum object_type *type, unsigned long *sizep)
1077 {
1078 unsigned shift;
1079 size_t size, c;
1080 unsigned long used = 0;
1081
1082 c = buf[used++];
1083 *type = (c >> 4) & 7;
1084 size = c & 15;
1085 shift = 4;
1086 while (c & 0x80) {
1087 if (len <= used || (bitsizeof(long) - 7) < shift) {
1088 error("bad object header");
1089 size = used = 0;
1090 break;
1091 }
1092 c = buf[used++];
1093 size = st_add(size, st_left_shift(c & 0x7f, shift));
1094 shift += 7;
1095 }
1096 *sizep = cast_size_t_to_ulong(size);
1097 return used;
1098 }
1099
1100 unsigned long get_size_from_delta(struct packed_git *p,
1101 struct pack_window **w_curs,
1102 off_t curpos)
1103 {
1104 const unsigned char *data;
1105 unsigned char delta_head[20], *in;
1106 git_zstream stream;
1107 int st;
1108
1109 memset(&stream, 0, sizeof(stream));
1110 stream.next_out = delta_head;
1111 stream.avail_out = sizeof(delta_head);
1112
1113 git_inflate_init(&stream);
1114 do {
1115 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1116 stream.next_in = in;
1117 /*
1118 * Note: the window section returned by use_pack() must be
1119 * available throughout git_inflate()'s unlocked execution. To
1120 * ensure no other thread will modify the window in the
1121 * meantime, we rely on the packed_window.inuse_cnt. This
1122 * counter is incremented before window reading and checked
1123 * before window disposal.
1124 *
1125 * Other worrying sections could be the call to close_pack_fd(),
1126 * which can close packs even with in-use windows, and to
1127 * reprepare_packed_git(). Regarding the former, mmap doc says:
1128 * "closing the file descriptor does not unmap the region". And
1129 * for the latter, it won't re-open already available packs.
1130 */
1131 obj_read_unlock();
1132 st = git_inflate(&stream, Z_FINISH);
1133 obj_read_lock();
1134 curpos += stream.next_in - in;
1135 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1136 stream.total_out < sizeof(delta_head));
1137 git_inflate_end(&stream);
1138 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1139 error("delta data unpack-initial failed");
1140 return 0;
1141 }
1142
1143 /* Examine the initial part of the delta to figure out
1144 * the result size.
1145 */
1146 data = delta_head;
1147
1148 /* ignore base size */
1149 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1150
1151 /* Read the result size */
1152 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1153 }
1154
1155 int unpack_object_header(struct packed_git *p,
1156 struct pack_window **w_curs,
1157 off_t *curpos,
1158 unsigned long *sizep)
1159 {
1160 unsigned char *base;
1161 unsigned long left;
1162 unsigned long used;
1163 enum object_type type;
1164
1165 /* use_pack() assures us we have [base, base + 20) available
1166 * as a range that we can look at. (Its actually the hash
1167 * size that is assured.) With our object header encoding
1168 * the maximum deflated object size is 2^137, which is just
1169 * insane, so we know won't exceed what we have been given.
1170 */
1171 base = use_pack(p, w_curs, *curpos, &left);
1172 used = unpack_object_header_buffer(base, left, &type, sizep);
1173 if (!used) {
1174 type = OBJ_BAD;
1175 } else
1176 *curpos += used;
1177
1178 return type;
1179 }
1180
1181 void mark_bad_packed_object(struct packed_git *p, const struct object_id *oid)
1182 {
1183 oidset_insert(&p->bad_objects, oid);
1184 }
1185
1186 const struct packed_git *has_packed_and_bad(struct repository *r,
1187 const struct object_id *oid)
1188 {
1189 struct packed_git *p;
1190
1191 for (p = r->objects->packed_git; p; p = p->next)
1192 if (oidset_contains(&p->bad_objects, oid))
1193 return p;
1194 return NULL;
1195 }
1196
1197 off_t get_delta_base(struct packed_git *p,
1198 struct pack_window **w_curs,
1199 off_t *curpos,
1200 enum object_type type,
1201 off_t delta_obj_offset)
1202 {
1203 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1204 off_t base_offset;
1205
1206 /* use_pack() assured us we have [base_info, base_info + 20)
1207 * as a range that we can look at without walking off the
1208 * end of the mapped window. Its actually the hash size
1209 * that is assured. An OFS_DELTA longer than the hash size
1210 * is stupid, as then a REF_DELTA would be smaller to store.
1211 */
1212 if (type == OBJ_OFS_DELTA) {
1213 unsigned used = 0;
1214 unsigned char c = base_info[used++];
1215 base_offset = c & 127;
1216 while (c & 128) {
1217 base_offset += 1;
1218 if (!base_offset || MSB(base_offset, 7))
1219 return 0; /* overflow */
1220 c = base_info[used++];
1221 base_offset = (base_offset << 7) + (c & 127);
1222 }
1223 base_offset = delta_obj_offset - base_offset;
1224 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1225 return 0; /* out of bound */
1226 *curpos += used;
1227 } else if (type == OBJ_REF_DELTA) {
1228 /* The base entry _must_ be in the same pack */
1229 base_offset = find_pack_entry_one(base_info, p);
1230 *curpos += the_hash_algo->rawsz;
1231 } else
1232 die("I am totally screwed");
1233 return base_offset;
1234 }
1235
1236 /*
1237 * Like get_delta_base above, but we return the sha1 instead of the pack
1238 * offset. This means it is cheaper for REF deltas (we do not have to do
1239 * the final object lookup), but more expensive for OFS deltas (we
1240 * have to load the revidx to convert the offset back into a sha1).
1241 */
1242 static int get_delta_base_oid(struct packed_git *p,
1243 struct pack_window **w_curs,
1244 off_t curpos,
1245 struct object_id *oid,
1246 enum object_type type,
1247 off_t delta_obj_offset)
1248 {
1249 if (type == OBJ_REF_DELTA) {
1250 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1251 oidread(oid, base);
1252 return 0;
1253 } else if (type == OBJ_OFS_DELTA) {
1254 uint32_t base_pos;
1255 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1256 type, delta_obj_offset);
1257
1258 if (!base_offset)
1259 return -1;
1260
1261 if (offset_to_pack_pos(p, base_offset, &base_pos) < 0)
1262 return -1;
1263
1264 return nth_packed_object_id(oid, p,
1265 pack_pos_to_index(p, base_pos));
1266 } else
1267 return -1;
1268 }
1269
1270 static int retry_bad_packed_offset(struct repository *r,
1271 struct packed_git *p,
1272 off_t obj_offset)
1273 {
1274 int type;
1275 uint32_t pos;
1276 struct object_id oid;
1277 if (offset_to_pack_pos(p, obj_offset, &pos) < 0)
1278 return OBJ_BAD;
1279 nth_packed_object_id(&oid, p, pack_pos_to_index(p, pos));
1280 mark_bad_packed_object(p, &oid);
1281 type = oid_object_info(r, &oid, NULL);
1282 if (type <= OBJ_NONE)
1283 return OBJ_BAD;
1284 return type;
1285 }
1286
1287 #define POI_STACK_PREALLOC 64
1288
1289 static enum object_type packed_to_object_type(struct repository *r,
1290 struct packed_git *p,
1291 off_t obj_offset,
1292 enum object_type type,
1293 struct pack_window **w_curs,
1294 off_t curpos)
1295 {
1296 off_t small_poi_stack[POI_STACK_PREALLOC];
1297 off_t *poi_stack = small_poi_stack;
1298 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1299
1300 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1301 off_t base_offset;
1302 unsigned long size;
1303 /* Push the object we're going to leave behind */
1304 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1305 poi_stack_alloc = alloc_nr(poi_stack_nr);
1306 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1307 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1308 } else {
1309 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1310 }
1311 poi_stack[poi_stack_nr++] = obj_offset;
1312 /* If parsing the base offset fails, just unwind */
1313 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1314 if (!base_offset)
1315 goto unwind;
1316 curpos = obj_offset = base_offset;
1317 type = unpack_object_header(p, w_curs, &curpos, &size);
1318 if (type <= OBJ_NONE) {
1319 /* If getting the base itself fails, we first
1320 * retry the base, otherwise unwind */
1321 type = retry_bad_packed_offset(r, p, base_offset);
1322 if (type > OBJ_NONE)
1323 goto out;
1324 goto unwind;
1325 }
1326 }
1327
1328 switch (type) {
1329 case OBJ_BAD:
1330 case OBJ_COMMIT:
1331 case OBJ_TREE:
1332 case OBJ_BLOB:
1333 case OBJ_TAG:
1334 break;
1335 default:
1336 error("unknown object type %i at offset %"PRIuMAX" in %s",
1337 type, (uintmax_t)obj_offset, p->pack_name);
1338 type = OBJ_BAD;
1339 }
1340
1341 out:
1342 if (poi_stack != small_poi_stack)
1343 free(poi_stack);
1344 return type;
1345
1346 unwind:
1347 while (poi_stack_nr) {
1348 obj_offset = poi_stack[--poi_stack_nr];
1349 type = retry_bad_packed_offset(r, p, obj_offset);
1350 if (type > OBJ_NONE)
1351 goto out;
1352 }
1353 type = OBJ_BAD;
1354 goto out;
1355 }
1356
1357 static struct hashmap delta_base_cache;
1358 static size_t delta_base_cached;
1359
1360 static LIST_HEAD(delta_base_cache_lru);
1361
1362 struct delta_base_cache_key {
1363 struct packed_git *p;
1364 off_t base_offset;
1365 };
1366
1367 struct delta_base_cache_entry {
1368 struct hashmap_entry ent;
1369 struct delta_base_cache_key key;
1370 struct list_head lru;
1371 void *data;
1372 unsigned long size;
1373 enum object_type type;
1374 };
1375
1376 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1377 {
1378 unsigned int hash;
1379
1380 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1381 hash += (hash >> 8) + (hash >> 16);
1382 return hash;
1383 }
1384
1385 static struct delta_base_cache_entry *
1386 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1387 {
1388 struct hashmap_entry entry, *e;
1389 struct delta_base_cache_key key;
1390
1391 if (!delta_base_cache.cmpfn)
1392 return NULL;
1393
1394 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1395 key.p = p;
1396 key.base_offset = base_offset;
1397 e = hashmap_get(&delta_base_cache, &entry, &key);
1398 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1399 }
1400
1401 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1402 const struct delta_base_cache_key *b)
1403 {
1404 return a->p == b->p && a->base_offset == b->base_offset;
1405 }
1406
1407 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1408 const struct hashmap_entry *va,
1409 const struct hashmap_entry *vb,
1410 const void *vkey)
1411 {
1412 const struct delta_base_cache_entry *a, *b;
1413 const struct delta_base_cache_key *key = vkey;
1414
1415 a = container_of(va, const struct delta_base_cache_entry, ent);
1416 b = container_of(vb, const struct delta_base_cache_entry, ent);
1417
1418 if (key)
1419 return !delta_base_cache_key_eq(&a->key, key);
1420 else
1421 return !delta_base_cache_key_eq(&a->key, &b->key);
1422 }
1423
1424 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1425 {
1426 return !!get_delta_base_cache_entry(p, base_offset);
1427 }
1428
1429 /*
1430 * Remove the entry from the cache, but do _not_ free the associated
1431 * entry data. The caller takes ownership of the "data" buffer, and
1432 * should copy out any fields it wants before detaching.
1433 */
1434 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1435 {
1436 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1437 list_del(&ent->lru);
1438 delta_base_cached -= ent->size;
1439 free(ent);
1440 }
1441
1442 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1443 off_t base_offset, unsigned long *base_size,
1444 enum object_type *type)
1445 {
1446 struct delta_base_cache_entry *ent;
1447
1448 ent = get_delta_base_cache_entry(p, base_offset);
1449 if (!ent)
1450 return unpack_entry(r, p, base_offset, type, base_size);
1451
1452 if (type)
1453 *type = ent->type;
1454 if (base_size)
1455 *base_size = ent->size;
1456 return xmemdupz(ent->data, ent->size);
1457 }
1458
1459 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1460 {
1461 free(ent->data);
1462 detach_delta_base_cache_entry(ent);
1463 }
1464
1465 void clear_delta_base_cache(void)
1466 {
1467 struct list_head *lru, *tmp;
1468 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1469 struct delta_base_cache_entry *entry =
1470 list_entry(lru, struct delta_base_cache_entry, lru);
1471 release_delta_base_cache(entry);
1472 }
1473 }
1474
1475 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1476 void *base, unsigned long base_size, enum object_type type)
1477 {
1478 struct delta_base_cache_entry *ent;
1479 struct list_head *lru, *tmp;
1480
1481 /*
1482 * Check required to avoid redundant entries when more than one thread
1483 * is unpacking the same object, in unpack_entry() (since its phases I
1484 * and III might run concurrently across multiple threads).
1485 */
1486 if (in_delta_base_cache(p, base_offset)) {
1487 free(base);
1488 return;
1489 }
1490
1491 delta_base_cached += base_size;
1492
1493 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1494 struct delta_base_cache_entry *f =
1495 list_entry(lru, struct delta_base_cache_entry, lru);
1496 if (delta_base_cached <= delta_base_cache_limit)
1497 break;
1498 release_delta_base_cache(f);
1499 }
1500
1501 ent = xmalloc(sizeof(*ent));
1502 ent->key.p = p;
1503 ent->key.base_offset = base_offset;
1504 ent->type = type;
1505 ent->data = base;
1506 ent->size = base_size;
1507 list_add_tail(&ent->lru, &delta_base_cache_lru);
1508
1509 if (!delta_base_cache.cmpfn)
1510 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1511 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1512 hashmap_add(&delta_base_cache, &ent->ent);
1513 }
1514
1515 int packed_object_info(struct repository *r, struct packed_git *p,
1516 off_t obj_offset, struct object_info *oi)
1517 {
1518 struct pack_window *w_curs = NULL;
1519 unsigned long size;
1520 off_t curpos = obj_offset;
1521 enum object_type type;
1522
1523 /*
1524 * We always get the representation type, but only convert it to
1525 * a "real" type later if the caller is interested.
1526 */
1527 if (oi->contentp) {
1528 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1529 &type);
1530 if (!*oi->contentp)
1531 type = OBJ_BAD;
1532 } else {
1533 type = unpack_object_header(p, &w_curs, &curpos, &size);
1534 }
1535
1536 if (!oi->contentp && oi->sizep) {
1537 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1538 off_t tmp_pos = curpos;
1539 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1540 type, obj_offset);
1541 if (!base_offset) {
1542 type = OBJ_BAD;
1543 goto out;
1544 }
1545 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1546 if (*oi->sizep == 0) {
1547 type = OBJ_BAD;
1548 goto out;
1549 }
1550 } else {
1551 *oi->sizep = size;
1552 }
1553 }
1554
1555 if (oi->disk_sizep) {
1556 uint32_t pos;
1557 if (offset_to_pack_pos(p, obj_offset, &pos) < 0) {
1558 error("could not find object at offset %"PRIuMAX" "
1559 "in pack %s", (uintmax_t)obj_offset, p->pack_name);
1560 type = OBJ_BAD;
1561 goto out;
1562 }
1563
1564 *oi->disk_sizep = pack_pos_to_offset(p, pos + 1) - obj_offset;
1565 }
1566
1567 if (oi->typep || oi->type_name) {
1568 enum object_type ptot;
1569 ptot = packed_to_object_type(r, p, obj_offset,
1570 type, &w_curs, curpos);
1571 if (oi->typep)
1572 *oi->typep = ptot;
1573 if (oi->type_name) {
1574 const char *tn = type_name(ptot);
1575 if (tn)
1576 strbuf_addstr(oi->type_name, tn);
1577 }
1578 if (ptot < 0) {
1579 type = OBJ_BAD;
1580 goto out;
1581 }
1582 }
1583
1584 if (oi->delta_base_oid) {
1585 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1586 if (get_delta_base_oid(p, &w_curs, curpos,
1587 oi->delta_base_oid,
1588 type, obj_offset) < 0) {
1589 type = OBJ_BAD;
1590 goto out;
1591 }
1592 } else
1593 oidclr(oi->delta_base_oid);
1594 }
1595
1596 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1597 OI_PACKED;
1598
1599 out:
1600 unuse_pack(&w_curs);
1601 return type;
1602 }
1603
1604 static void *unpack_compressed_entry(struct packed_git *p,
1605 struct pack_window **w_curs,
1606 off_t curpos,
1607 unsigned long size)
1608 {
1609 int st;
1610 git_zstream stream;
1611 unsigned char *buffer, *in;
1612
1613 buffer = xmallocz_gently(size);
1614 if (!buffer)
1615 return NULL;
1616 memset(&stream, 0, sizeof(stream));
1617 stream.next_out = buffer;
1618 stream.avail_out = size + 1;
1619
1620 git_inflate_init(&stream);
1621 do {
1622 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1623 stream.next_in = in;
1624 /*
1625 * Note: we must ensure the window section returned by
1626 * use_pack() will be available throughout git_inflate()'s
1627 * unlocked execution. Please refer to the comment at
1628 * get_size_from_delta() to see how this is done.
1629 */
1630 obj_read_unlock();
1631 st = git_inflate(&stream, Z_FINISH);
1632 obj_read_lock();
1633 if (!stream.avail_out)
1634 break; /* the payload is larger than it should be */
1635 curpos += stream.next_in - in;
1636 } while (st == Z_OK || st == Z_BUF_ERROR);
1637 git_inflate_end(&stream);
1638 if ((st != Z_STREAM_END) || stream.total_out != size) {
1639 free(buffer);
1640 return NULL;
1641 }
1642
1643 /* versions of zlib can clobber unconsumed portion of outbuf */
1644 buffer[size] = '\0';
1645
1646 return buffer;
1647 }
1648
1649 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1650 {
1651 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1652 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1653 p->pack_name, (uintmax_t)obj_offset);
1654 }
1655
1656 int do_check_packed_object_crc;
1657
1658 #define UNPACK_ENTRY_STACK_PREALLOC 64
1659 struct unpack_entry_stack_ent {
1660 off_t obj_offset;
1661 off_t curpos;
1662 unsigned long size;
1663 };
1664
1665 static void *read_object(struct repository *r,
1666 const struct object_id *oid,
1667 enum object_type *type,
1668 unsigned long *size)
1669 {
1670 struct object_info oi = OBJECT_INFO_INIT;
1671 void *content;
1672 oi.typep = type;
1673 oi.sizep = size;
1674 oi.contentp = &content;
1675
1676 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1677 return NULL;
1678 return content;
1679 }
1680
1681 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1682 enum object_type *final_type, unsigned long *final_size)
1683 {
1684 struct pack_window *w_curs = NULL;
1685 off_t curpos = obj_offset;
1686 void *data = NULL;
1687 unsigned long size;
1688 enum object_type type;
1689 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1690 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1691 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1692 int base_from_cache = 0;
1693
1694 write_pack_access_log(p, obj_offset);
1695
1696 /* PHASE 1: drill down to the innermost base object */
1697 for (;;) {
1698 off_t base_offset;
1699 int i;
1700 struct delta_base_cache_entry *ent;
1701
1702 ent = get_delta_base_cache_entry(p, curpos);
1703 if (ent) {
1704 type = ent->type;
1705 data = ent->data;
1706 size = ent->size;
1707 detach_delta_base_cache_entry(ent);
1708 base_from_cache = 1;
1709 break;
1710 }
1711
1712 if (do_check_packed_object_crc && p->index_version > 1) {
1713 uint32_t pack_pos, index_pos;
1714 off_t len;
1715
1716 if (offset_to_pack_pos(p, obj_offset, &pack_pos) < 0) {
1717 error("could not find object at offset %"PRIuMAX" in pack %s",
1718 (uintmax_t)obj_offset, p->pack_name);
1719 data = NULL;
1720 goto out;
1721 }
1722
1723 len = pack_pos_to_offset(p, pack_pos + 1) - obj_offset;
1724 index_pos = pack_pos_to_index(p, pack_pos);
1725 if (check_pack_crc(p, &w_curs, obj_offset, len, index_pos)) {
1726 struct object_id oid;
1727 nth_packed_object_id(&oid, p, index_pos);
1728 error("bad packed object CRC for %s",
1729 oid_to_hex(&oid));
1730 mark_bad_packed_object(p, &oid);
1731 data = NULL;
1732 goto out;
1733 }
1734 }
1735
1736 type = unpack_object_header(p, &w_curs, &curpos, &size);
1737 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1738 break;
1739
1740 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1741 if (!base_offset) {
1742 error("failed to validate delta base reference "
1743 "at offset %"PRIuMAX" from %s",
1744 (uintmax_t)curpos, p->pack_name);
1745 /* bail to phase 2, in hopes of recovery */
1746 data = NULL;
1747 break;
1748 }
1749
1750 /* push object, proceed to base */
1751 if (delta_stack_nr >= delta_stack_alloc
1752 && delta_stack == small_delta_stack) {
1753 delta_stack_alloc = alloc_nr(delta_stack_nr);
1754 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1755 COPY_ARRAY(delta_stack, small_delta_stack,
1756 delta_stack_nr);
1757 } else {
1758 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1759 }
1760 i = delta_stack_nr++;
1761 delta_stack[i].obj_offset = obj_offset;
1762 delta_stack[i].curpos = curpos;
1763 delta_stack[i].size = size;
1764
1765 curpos = obj_offset = base_offset;
1766 }
1767
1768 /* PHASE 2: handle the base */
1769 switch (type) {
1770 case OBJ_OFS_DELTA:
1771 case OBJ_REF_DELTA:
1772 if (data)
1773 BUG("unpack_entry: left loop at a valid delta");
1774 break;
1775 case OBJ_COMMIT:
1776 case OBJ_TREE:
1777 case OBJ_BLOB:
1778 case OBJ_TAG:
1779 if (!base_from_cache)
1780 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1781 break;
1782 default:
1783 data = NULL;
1784 error("unknown object type %i at offset %"PRIuMAX" in %s",
1785 type, (uintmax_t)obj_offset, p->pack_name);
1786 }
1787
1788 /* PHASE 3: apply deltas in order */
1789
1790 /* invariants:
1791 * 'data' holds the base data, or NULL if there was corruption
1792 */
1793 while (delta_stack_nr) {
1794 void *delta_data;
1795 void *base = data;
1796 void *external_base = NULL;
1797 unsigned long delta_size, base_size = size;
1798 int i;
1799 off_t base_obj_offset = obj_offset;
1800
1801 data = NULL;
1802
1803 if (!base) {
1804 /*
1805 * We're probably in deep shit, but let's try to fetch
1806 * the required base anyway from another pack or loose.
1807 * This is costly but should happen only in the presence
1808 * of a corrupted pack, and is better than failing outright.
1809 */
1810 uint32_t pos;
1811 struct object_id base_oid;
1812 if (!(offset_to_pack_pos(p, obj_offset, &pos))) {
1813 nth_packed_object_id(&base_oid, p,
1814 pack_pos_to_index(p, pos));
1815 error("failed to read delta base object %s"
1816 " at offset %"PRIuMAX" from %s",
1817 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1818 p->pack_name);
1819 mark_bad_packed_object(p, &base_oid);
1820 base = read_object(r, &base_oid, &type, &base_size);
1821 external_base = base;
1822 }
1823 }
1824
1825 i = --delta_stack_nr;
1826 obj_offset = delta_stack[i].obj_offset;
1827 curpos = delta_stack[i].curpos;
1828 delta_size = delta_stack[i].size;
1829
1830 if (!base)
1831 continue;
1832
1833 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1834
1835 if (!delta_data) {
1836 error("failed to unpack compressed delta "
1837 "at offset %"PRIuMAX" from %s",
1838 (uintmax_t)curpos, p->pack_name);
1839 data = NULL;
1840 } else {
1841 data = patch_delta(base, base_size, delta_data,
1842 delta_size, &size);
1843
1844 /*
1845 * We could not apply the delta; warn the user, but
1846 * keep going. Our failure will be noticed either in
1847 * the next iteration of the loop, or if this is the
1848 * final delta, in the caller when we return NULL.
1849 * Those code paths will take care of making a more
1850 * explicit warning and retrying with another copy of
1851 * the object.
1852 */
1853 if (!data)
1854 error("failed to apply delta");
1855 }
1856
1857 /*
1858 * We delay adding `base` to the cache until the end of the loop
1859 * because unpack_compressed_entry() momentarily releases the
1860 * obj_read_mutex, giving another thread the chance to access
1861 * the cache. Therefore, if `base` was already there, this other
1862 * thread could free() it (e.g. to make space for another entry)
1863 * before we are done using it.
1864 */
1865 if (!external_base)
1866 add_delta_base_cache(p, base_obj_offset, base, base_size, type);
1867
1868 free(delta_data);
1869 free(external_base);
1870 }
1871
1872 if (final_type)
1873 *final_type = type;
1874 if (final_size)
1875 *final_size = size;
1876
1877 out:
1878 unuse_pack(&w_curs);
1879
1880 if (delta_stack != small_delta_stack)
1881 free(delta_stack);
1882
1883 return data;
1884 }
1885
1886 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1887 {
1888 const unsigned char *index_fanout = p->index_data;
1889 const unsigned char *index_lookup;
1890 const unsigned int hashsz = the_hash_algo->rawsz;
1891 int index_lookup_width;
1892
1893 if (!index_fanout)
1894 BUG("bsearch_pack called without a valid pack-index");
1895
1896 index_lookup = index_fanout + 4 * 256;
1897 if (p->index_version == 1) {
1898 index_lookup_width = hashsz + 4;
1899 index_lookup += 4;
1900 } else {
1901 index_lookup_width = hashsz;
1902 index_fanout += 8;
1903 index_lookup += 8;
1904 }
1905
1906 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1907 index_lookup, index_lookup_width, result);
1908 }
1909
1910 int nth_packed_object_id(struct object_id *oid,
1911 struct packed_git *p,
1912 uint32_t n)
1913 {
1914 const unsigned char *index = p->index_data;
1915 const unsigned int hashsz = the_hash_algo->rawsz;
1916 if (!index) {
1917 if (open_pack_index(p))
1918 return -1;
1919 index = p->index_data;
1920 }
1921 if (n >= p->num_objects)
1922 return -1;
1923 index += 4 * 256;
1924 if (p->index_version == 1) {
1925 oidread(oid, index + (hashsz + 4) * n + 4);
1926 } else {
1927 index += 8;
1928 oidread(oid, index + hashsz * n);
1929 }
1930 return 0;
1931 }
1932
1933 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1934 {
1935 const unsigned char *ptr = vptr;
1936 const unsigned char *start = p->index_data;
1937 const unsigned char *end = start + p->index_size;
1938 if (ptr < start)
1939 die(_("offset before start of pack index for %s (corrupt index?)"),
1940 p->pack_name);
1941 /* No need to check for underflow; .idx files must be at least 8 bytes */
1942 if (ptr >= end - 8)
1943 die(_("offset beyond end of pack index for %s (truncated index?)"),
1944 p->pack_name);
1945 }
1946
1947 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1948 {
1949 const unsigned char *index = p->index_data;
1950 const unsigned int hashsz = the_hash_algo->rawsz;
1951 index += 4 * 256;
1952 if (p->index_version == 1) {
1953 return ntohl(*((uint32_t *)(index + (hashsz + 4) * (size_t)n)));
1954 } else {
1955 uint32_t off;
1956 index += 8 + (size_t)p->num_objects * (hashsz + 4);
1957 off = ntohl(*((uint32_t *)(index + 4 * n)));
1958 if (!(off & 0x80000000))
1959 return off;
1960 index += (size_t)p->num_objects * 4 + (off & 0x7fffffff) * 8;
1961 check_pack_index_ptr(p, index);
1962 return get_be64(index);
1963 }
1964 }
1965
1966 off_t find_pack_entry_one(const unsigned char *sha1,
1967 struct packed_git *p)
1968 {
1969 const unsigned char *index = p->index_data;
1970 struct object_id oid;
1971 uint32_t result;
1972
1973 if (!index) {
1974 if (open_pack_index(p))
1975 return 0;
1976 }
1977
1978 hashcpy(oid.hash, sha1);
1979 if (bsearch_pack(&oid, p, &result))
1980 return nth_packed_object_offset(p, result);
1981 return 0;
1982 }
1983
1984 int is_pack_valid(struct packed_git *p)
1985 {
1986 /* An already open pack is known to be valid. */
1987 if (p->pack_fd != -1)
1988 return 1;
1989
1990 /* If the pack has one window completely covering the
1991 * file size, the pack is known to be valid even if
1992 * the descriptor is not currently open.
1993 */
1994 if (p->windows) {
1995 struct pack_window *w = p->windows;
1996
1997 if (!w->offset && w->len == p->pack_size)
1998 return 1;
1999 }
2000
2001 /* Force the pack to open to prove its valid. */
2002 return !open_packed_git(p);
2003 }
2004
2005 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2006 struct packed_git *packs)
2007 {
2008 struct packed_git *p;
2009
2010 for (p = packs; p; p = p->next) {
2011 if (find_pack_entry_one(sha1, p))
2012 return p;
2013 }
2014 return NULL;
2015
2016 }
2017
2018 static int fill_pack_entry(const struct object_id *oid,
2019 struct pack_entry *e,
2020 struct packed_git *p)
2021 {
2022 off_t offset;
2023
2024 if (oidset_size(&p->bad_objects) &&
2025 oidset_contains(&p->bad_objects, oid))
2026 return 0;
2027
2028 offset = find_pack_entry_one(oid->hash, p);
2029 if (!offset)
2030 return 0;
2031
2032 /*
2033 * We are about to tell the caller where they can locate the
2034 * requested object. We better make sure the packfile is
2035 * still here and can be accessed before supplying that
2036 * answer, as it may have been deleted since the index was
2037 * loaded!
2038 */
2039 if (!is_pack_valid(p))
2040 return 0;
2041 e->offset = offset;
2042 e->p = p;
2043 return 1;
2044 }
2045
2046 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2047 {
2048 struct list_head *pos;
2049 struct multi_pack_index *m;
2050
2051 prepare_packed_git(r);
2052 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2053 return 0;
2054
2055 for (m = r->objects->multi_pack_index; m; m = m->next) {
2056 if (fill_midx_entry(r, oid, e, m))
2057 return 1;
2058 }
2059
2060 list_for_each(pos, &r->objects->packed_git_mru) {
2061 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2062 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2063 list_move(&p->mru, &r->objects->packed_git_mru);
2064 return 1;
2065 }
2066 }
2067 return 0;
2068 }
2069
2070 static void maybe_invalidate_kept_pack_cache(struct repository *r,
2071 unsigned flags)
2072 {
2073 if (!r->objects->kept_pack_cache.packs)
2074 return;
2075 if (r->objects->kept_pack_cache.flags == flags)
2076 return;
2077 FREE_AND_NULL(r->objects->kept_pack_cache.packs);
2078 r->objects->kept_pack_cache.flags = 0;
2079 }
2080
2081 static struct packed_git **kept_pack_cache(struct repository *r, unsigned flags)
2082 {
2083 maybe_invalidate_kept_pack_cache(r, flags);
2084
2085 if (!r->objects->kept_pack_cache.packs) {
2086 struct packed_git **packs = NULL;
2087 size_t nr = 0, alloc = 0;
2088 struct packed_git *p;
2089
2090 /*
2091 * We want "all" packs here, because we need to cover ones that
2092 * are used by a midx, as well. We need to look in every one of
2093 * them (instead of the midx itself) to cover duplicates. It's
2094 * possible that an object is found in two packs that the midx
2095 * covers, one kept and one not kept, but the midx returns only
2096 * the non-kept version.
2097 */
2098 for (p = get_all_packs(r); p; p = p->next) {
2099 if ((p->pack_keep && (flags & ON_DISK_KEEP_PACKS)) ||
2100 (p->pack_keep_in_core && (flags & IN_CORE_KEEP_PACKS))) {
2101 ALLOC_GROW(packs, nr + 1, alloc);
2102 packs[nr++] = p;
2103 }
2104 }
2105 ALLOC_GROW(packs, nr + 1, alloc);
2106 packs[nr] = NULL;
2107
2108 r->objects->kept_pack_cache.packs = packs;
2109 r->objects->kept_pack_cache.flags = flags;
2110 }
2111
2112 return r->objects->kept_pack_cache.packs;
2113 }
2114
2115 int find_kept_pack_entry(struct repository *r,
2116 const struct object_id *oid,
2117 unsigned flags,
2118 struct pack_entry *e)
2119 {
2120 struct packed_git **cache;
2121
2122 for (cache = kept_pack_cache(r, flags); *cache; cache++) {
2123 struct packed_git *p = *cache;
2124 if (fill_pack_entry(oid, e, p))
2125 return 1;
2126 }
2127
2128 return 0;
2129 }
2130
2131 int has_object_pack(const struct object_id *oid)
2132 {
2133 struct pack_entry e;
2134 return find_pack_entry(the_repository, oid, &e);
2135 }
2136
2137 int has_object_kept_pack(const struct object_id *oid, unsigned flags)
2138 {
2139 struct pack_entry e;
2140 return find_kept_pack_entry(the_repository, oid, flags, &e);
2141 }
2142
2143 int has_pack_index(const unsigned char *sha1)
2144 {
2145 struct stat st;
2146 if (stat(sha1_pack_index_name(sha1), &st))
2147 return 0;
2148 return 1;
2149 }
2150
2151 int for_each_object_in_pack(struct packed_git *p,
2152 each_packed_object_fn cb, void *data,
2153 enum for_each_object_flags flags)
2154 {
2155 uint32_t i;
2156 int r = 0;
2157
2158 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2159 if (load_pack_revindex(p))
2160 return -1;
2161 }
2162
2163 for (i = 0; i < p->num_objects; i++) {
2164 uint32_t index_pos;
2165 struct object_id oid;
2166
2167 /*
2168 * We are iterating "i" from 0 up to num_objects, but its
2169 * meaning may be different, depending on the requested output
2170 * order:
2171 *
2172 * - in object-name order, it is the same as the index order
2173 * used by nth_packed_object_id(), so we can pass it
2174 * directly
2175 *
2176 * - in pack-order, it is pack position, which we must
2177 * convert to an index position in order to get the oid.
2178 */
2179 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2180 index_pos = pack_pos_to_index(p, i);
2181 else
2182 index_pos = i;
2183
2184 if (nth_packed_object_id(&oid, p, index_pos) < 0)
2185 return error("unable to get sha1 of object %u in %s",
2186 index_pos, p->pack_name);
2187
2188 r = cb(&oid, p, index_pos, data);
2189 if (r)
2190 break;
2191 }
2192 return r;
2193 }
2194
2195 int for_each_packed_object(each_packed_object_fn cb, void *data,
2196 enum for_each_object_flags flags)
2197 {
2198 struct packed_git *p;
2199 int r = 0;
2200 int pack_errors = 0;
2201
2202 prepare_packed_git(the_repository);
2203 for (p = get_all_packs(the_repository); p; p = p->next) {
2204 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2205 continue;
2206 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2207 !p->pack_promisor)
2208 continue;
2209 if ((flags & FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS) &&
2210 p->pack_keep_in_core)
2211 continue;
2212 if ((flags & FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS) &&
2213 p->pack_keep)
2214 continue;
2215 if (open_pack_index(p)) {
2216 pack_errors = 1;
2217 continue;
2218 }
2219 r = for_each_object_in_pack(p, cb, data, flags);
2220 if (r)
2221 break;
2222 }
2223 return r ? r : pack_errors;
2224 }
2225
2226 static int add_promisor_object(const struct object_id *oid,
2227 struct packed_git *pack,
2228 uint32_t pos,
2229 void *set_)
2230 {
2231 struct oidset *set = set_;
2232 struct object *obj = parse_object(the_repository, oid);
2233 if (!obj)
2234 return 1;
2235
2236 oidset_insert(set, oid);
2237
2238 /*
2239 * If this is a tree, commit, or tag, the objects it refers
2240 * to are also promisor objects. (Blobs refer to no objects->)
2241 */
2242 if (obj->type == OBJ_TREE) {
2243 struct tree *tree = (struct tree *)obj;
2244 struct tree_desc desc;
2245 struct name_entry entry;
2246 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2247 /*
2248 * Error messages are given when packs are
2249 * verified, so do not print any here.
2250 */
2251 return 0;
2252 while (tree_entry_gently(&desc, &entry))
2253 oidset_insert(set, &entry.oid);
2254 free_tree_buffer(tree);
2255 } else if (obj->type == OBJ_COMMIT) {
2256 struct commit *commit = (struct commit *) obj;
2257 struct commit_list *parents = commit->parents;
2258
2259 oidset_insert(set, get_commit_tree_oid(commit));
2260 for (; parents; parents = parents->next)
2261 oidset_insert(set, &parents->item->object.oid);
2262 } else if (obj->type == OBJ_TAG) {
2263 struct tag *tag = (struct tag *) obj;
2264 oidset_insert(set, get_tagged_oid(tag));
2265 }
2266 return 0;
2267 }
2268
2269 int is_promisor_object(const struct object_id *oid)
2270 {
2271 static struct oidset promisor_objects;
2272 static int promisor_objects_prepared;
2273
2274 if (!promisor_objects_prepared) {
2275 if (has_promisor_remote()) {
2276 for_each_packed_object(add_promisor_object,
2277 &promisor_objects,
2278 FOR_EACH_OBJECT_PROMISOR_ONLY);
2279 }
2280 promisor_objects_prepared = 1;
2281 }
2282 return oidset_contains(&promisor_objects, oid);
2283 }
git for Debian