midx: fix bug that skips midx with alternates
[git/git.git] / packfile.c
CommitLineData
4f39cd82 1#include "cache.h"
ec2dd32c 2#include "list.h"
0317f455 3#include "pack.h"
031dc927 4#include "repository.h"
0abe14f6
JT
5#include "dir.h"
6#include "mergesort.h"
7#include "packfile.h"
7b3aa75d 8#include "delta.h"
f1d8130b
JT
9#include "list.h"
10#include "streaming.h"
a2551953 11#include "sha1-lookup.h"
498f1f61
JT
12#include "commit.h"
13#include "object.h"
14#include "tag.h"
15#include "tree-walk.h"
16#include "tree.h"
0d4a1321 17#include "object-store.h"
c4d25228 18#include "midx.h"
4f39cd82
JT
19
20char *odb_pack_name(struct strbuf *buf,
21 const unsigned char *sha1,
22 const char *ext)
23{
24 strbuf_reset(buf);
25 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1), ext);
27 return buf->buf;
28}
29
30char *sha1_pack_name(const unsigned char *sha1)
31{
32 static struct strbuf buf = STRBUF_INIT;
33 return odb_pack_name(&buf, sha1, "pack");
34}
35
36char *sha1_pack_index_name(const unsigned char *sha1)
37{
38 static struct strbuf buf = STRBUF_INIT;
39 return odb_pack_name(&buf, sha1, "idx");
40}
6d6a80e0 41
84f80ad5
JT
42static unsigned int pack_used_ctr;
43static unsigned int pack_mmap_calls;
44static unsigned int peak_pack_open_windows;
45static unsigned int pack_open_windows;
e65f1862 46static unsigned int pack_open_fds;
84f80ad5
JT
47static unsigned int pack_max_fds;
48static size_t peak_pack_mapped;
49static size_t pack_mapped;
8e21176c
JT
50
51#define SZ_FMT PRIuMAX
52static inline uintmax_t sz_fmt(size_t s) { return s; }
53
54void pack_report(void)
55{
56 fprintf(stderr,
57 "pack_report: getpagesize() = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size),
62 sz_fmt(packed_git_limit));
63 fprintf(stderr,
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT " / %10" SZ_FMT "\n",
69 pack_used_ctr,
70 pack_mmap_calls,
71 pack_open_windows, peak_pack_open_windows,
72 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
73}
0317f455
JT
74
75/*
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
78 * success.
79 */
80static int check_packed_git_idx(const char *path, struct packed_git *p)
81{
82 void *idx_map;
83 struct pack_idx_header *hdr;
84 size_t idx_size;
85 uint32_t version, nr, i, *index;
86 int fd = git_open(path);
87 struct stat st;
37fec86a 88 const unsigned int hashsz = the_hash_algo->rawsz;
0317f455
JT
89
90 if (fd < 0)
91 return -1;
92 if (fstat(fd, &st)) {
93 close(fd);
94 return -1;
95 }
96 idx_size = xsize_t(st.st_size);
37fec86a 97 if (idx_size < 4 * 256 + hashsz + hashsz) {
0317f455
JT
98 close(fd);
99 return error("index file %s is too small", path);
100 }
101 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
102 close(fd);
103
104 hdr = idx_map;
105 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
106 version = ntohl(hdr->idx_version);
107 if (version < 2 || version > 2) {
108 munmap(idx_map, idx_size);
109 return error("index file %s is version %"PRIu32
110 " and is not supported by this binary"
111 " (try upgrading GIT to a newer version)",
112 path, version);
113 }
114 } else
115 version = 1;
116
117 nr = 0;
118 index = idx_map;
119 if (version > 1)
120 index += 2; /* skip index header */
121 for (i = 0; i < 256; i++) {
122 uint32_t n = ntohl(index[i]);
123 if (n < nr) {
124 munmap(idx_map, idx_size);
125 return error("non-monotonic index %s", path);
126 }
127 nr = n;
128 }
129
130 if (version == 1) {
131 /*
132 * Total size:
133 * - 256 index entries 4 bytes each
37fec86a 134 * - 24-byte entries * nr (object ID + 4-byte offset)
135 * - hash of the packfile
136 * - file checksum
0317f455 137 */
37fec86a 138 if (idx_size != 4*256 + nr * (hashsz + 4) + hashsz + hashsz) {
0317f455
JT
139 munmap(idx_map, idx_size);
140 return error("wrong index v1 file size in %s", path);
141 }
142 } else if (version == 2) {
143 /*
144 * Minimum size:
145 * - 8 bytes of header
146 * - 256 index entries 4 bytes each
37fec86a 147 * - object ID entry * nr
0317f455
JT
148 * - 4-byte crc entry * nr
149 * - 4-byte offset entry * nr
37fec86a 150 * - hash of the packfile
151 * - file checksum
0317f455
JT
152 * And after the 4-byte offset table might be a
153 * variable sized table containing 8-byte entries
154 * for offsets larger than 2^31.
155 */
37fec86a 156 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
0317f455
JT
157 unsigned long max_size = min_size;
158 if (nr)
159 max_size += (nr - 1)*8;
160 if (idx_size < min_size || idx_size > max_size) {
161 munmap(idx_map, idx_size);
162 return error("wrong index v2 file size in %s", path);
163 }
164 if (idx_size != min_size &&
165 /*
166 * make sure we can deal with large pack offsets.
167 * 31-bit signed offset won't be enough, neither
168 * 32-bit unsigned one will be.
169 */
170 (sizeof(off_t) <= 4)) {
171 munmap(idx_map, idx_size);
172 return error("pack too large for current definition of off_t in %s", path);
173 }
174 }
175
176 p->index_version = version;
177 p->index_data = idx_map;
178 p->index_size = idx_size;
179 p->num_objects = nr;
180 return 0;
181}
182
183int open_pack_index(struct packed_git *p)
184{
185 char *idx_name;
186 size_t len;
187 int ret;
188
189 if (p->index_data)
190 return 0;
191
192 if (!strip_suffix(p->pack_name, ".pack", &len))
033abf97 193 BUG("pack_name does not end in .pack");
0317f455
JT
194 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
195 ret = check_packed_git_idx(idx_name, p);
196 free(idx_name);
197 return ret;
198}
199
fe1ed56f
DS
200uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
201{
202 const uint32_t *level1_ofs = p->index_data;
203
204 if (!level1_ofs) {
205 if (open_pack_index(p))
206 return 0;
207 level1_ofs = p->index_data;
208 }
209
210 if (p->index_version > 1) {
211 level1_ofs += 2;
212 }
213
214 return ntohl(level1_ofs[value]);
215}
216
0317f455
JT
217static struct packed_git *alloc_packed_git(int extra)
218{
219 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
220 memset(p, 0, sizeof(*p));
221 p->pack_fd = -1;
222 return p;
223}
224
225struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
226{
227 const char *path = sha1_pack_name(sha1);
228 size_t alloc = st_add(strlen(path), 1);
229 struct packed_git *p = alloc_packed_git(alloc);
230
231 memcpy(p->pack_name, path, alloc); /* includes NUL */
232 hashcpy(p->sha1, sha1);
233 if (check_packed_git_idx(idx_path, p)) {
234 free(p);
235 return NULL;
236 }
237
238 return p;
239}
f0e17e86
JT
240
241static void scan_windows(struct packed_git *p,
242 struct packed_git **lru_p,
243 struct pack_window **lru_w,
244 struct pack_window **lru_l)
245{
246 struct pack_window *w, *w_l;
247
248 for (w_l = NULL, w = p->windows; w; w = w->next) {
249 if (!w->inuse_cnt) {
250 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
251 *lru_p = p;
252 *lru_w = w;
253 *lru_l = w_l;
254 }
255 }
256 w_l = w;
257 }
258}
259
84f80ad5 260static int unuse_one_window(struct packed_git *current)
f0e17e86
JT
261{
262 struct packed_git *p, *lru_p = NULL;
263 struct pack_window *lru_w = NULL, *lru_l = NULL;
264
265 if (current)
266 scan_windows(current, &lru_p, &lru_w, &lru_l);
a80d72db 267 for (p = the_repository->objects->packed_git; p; p = p->next)
f0e17e86
JT
268 scan_windows(p, &lru_p, &lru_w, &lru_l);
269 if (lru_p) {
270 munmap(lru_w->base, lru_w->len);
271 pack_mapped -= lru_w->len;
272 if (lru_l)
273 lru_l->next = lru_w->next;
274 else
275 lru_p->windows = lru_w->next;
276 free(lru_w);
277 pack_open_windows--;
278 return 1;
279 }
280 return 0;
281}
282
283void release_pack_memory(size_t need)
284{
285 size_t cur = pack_mapped;
286 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
287 ; /* nothing */
288}
3836d88a
JT
289
290void close_pack_windows(struct packed_git *p)
291{
292 while (p->windows) {
293 struct pack_window *w = p->windows;
294
295 if (w->inuse_cnt)
296 die("pack '%s' still has open windows to it",
297 p->pack_name);
298 munmap(w->base, w->len);
299 pack_mapped -= w->len;
300 pack_open_windows--;
301 p->windows = w->next;
302 free(w);
303 }
304}
305
84f80ad5 306static int close_pack_fd(struct packed_git *p)
3836d88a
JT
307{
308 if (p->pack_fd < 0)
309 return 0;
310
311 close(p->pack_fd);
312 pack_open_fds--;
313 p->pack_fd = -1;
314
315 return 1;
316}
317
318void close_pack_index(struct packed_git *p)
319{
320 if (p->index_data) {
321 munmap((void *)p->index_data, p->index_size);
322 p->index_data = NULL;
323 }
324}
325
049d51a2 326void close_pack(struct packed_git *p)
3836d88a
JT
327{
328 close_pack_windows(p);
329 close_pack_fd(p);
330 close_pack_index(p);
331}
332
d0b59866 333void close_all_packs(struct raw_object_store *o)
3836d88a
JT
334{
335 struct packed_git *p;
336
d0b59866 337 for (p = o->packed_git; p; p = p->next)
3836d88a 338 if (p->do_not_close)
033abf97 339 BUG("want to close pack marked 'do-not-close'");
3836d88a
JT
340 else
341 close_pack(p);
342}
84f80ad5
JT
343
344/*
345 * The LRU pack is the one with the oldest MRU window, preferring packs
346 * with no used windows, or the oldest mtime if it has no windows allocated.
347 */
348static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
349{
350 struct pack_window *w, *this_mru_w;
351 int has_windows_inuse = 0;
352
353 /*
354 * Reject this pack if it has windows and the previously selected
355 * one does not. If this pack does not have windows, reject
356 * it if the pack file is newer than the previously selected one.
357 */
358 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
359 return;
360
361 for (w = this_mru_w = p->windows; w; w = w->next) {
362 /*
363 * Reject this pack if any of its windows are in use,
364 * but the previously selected pack did not have any
365 * inuse windows. Otherwise, record that this pack
366 * has windows in use.
367 */
368 if (w->inuse_cnt) {
369 if (*accept_windows_inuse)
370 has_windows_inuse = 1;
371 else
372 return;
373 }
374
375 if (w->last_used > this_mru_w->last_used)
376 this_mru_w = w;
377
378 /*
379 * Reject this pack if it has windows that have been
380 * used more recently than the previously selected pack.
381 * If the previously selected pack had windows inuse and
382 * we have not encountered a window in this pack that is
383 * inuse, skip this check since we prefer a pack with no
384 * inuse windows to one that has inuse windows.
385 */
386 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
387 this_mru_w->last_used > (*mru_w)->last_used)
388 return;
389 }
390
391 /*
392 * Select this pack.
393 */
394 *mru_w = this_mru_w;
395 *lru_p = p;
396 *accept_windows_inuse = has_windows_inuse;
397}
398
399static int close_one_pack(void)
400{
401 struct packed_git *p, *lru_p = NULL;
402 struct pack_window *mru_w = NULL;
403 int accept_windows_inuse = 1;
404
a80d72db 405 for (p = the_repository->objects->packed_git; p; p = p->next) {
84f80ad5
JT
406 if (p->pack_fd == -1)
407 continue;
408 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
409 }
410
411 if (lru_p)
412 return close_pack_fd(lru_p);
413
414 return 0;
415}
416
417static unsigned int get_max_fd_limit(void)
418{
419#ifdef RLIMIT_NOFILE
420 {
421 struct rlimit lim;
422
423 if (!getrlimit(RLIMIT_NOFILE, &lim))
424 return lim.rlim_cur;
425 }
426#endif
427
428#ifdef _SC_OPEN_MAX
429 {
430 long open_max = sysconf(_SC_OPEN_MAX);
431 if (0 < open_max)
432 return open_max;
433 /*
434 * Otherwise, we got -1 for one of the two
435 * reasons:
436 *
437 * (1) sysconf() did not understand _SC_OPEN_MAX
438 * and signaled an error with -1; or
439 * (2) sysconf() said there is no limit.
440 *
441 * We _could_ clear errno before calling sysconf() to
442 * tell these two cases apart and return a huge number
443 * in the latter case to let the caller cap it to a
444 * value that is not so selfish, but letting the
445 * fallback OPEN_MAX codepath take care of these cases
446 * is a lot simpler.
447 */
448 }
449#endif
450
451#ifdef OPEN_MAX
452 return OPEN_MAX;
453#else
454 return 1; /* see the caller ;-) */
455#endif
456}
457
458/*
459 * Do not call this directly as this leaks p->pack_fd on error return;
460 * call open_packed_git() instead.
461 */
462static int open_packed_git_1(struct packed_git *p)
463{
464 struct stat st;
465 struct pack_header hdr;
37fec86a 466 unsigned char hash[GIT_MAX_RAWSZ];
467 unsigned char *idx_hash;
84f80ad5 468 long fd_flag;
41dcc4dc 469 ssize_t read_result;
37fec86a 470 const unsigned hashsz = the_hash_algo->rawsz;
84f80ad5 471
17c35c89
DS
472 if (!p->index_data) {
473 struct multi_pack_index *m;
474 const char *pack_name = strrchr(p->pack_name, '/');
475
476 for (m = the_repository->objects->multi_pack_index;
477 m; m = m->next) {
478 if (midx_contains_pack(m, pack_name))
479 break;
480 }
481
482 if (!m && open_pack_index(p))
483 return error("packfile %s index unavailable", p->pack_name);
484 }
84f80ad5
JT
485
486 if (!pack_max_fds) {
487 unsigned int max_fds = get_max_fd_limit();
488
489 /* Save 3 for stdin/stdout/stderr, 22 for work */
490 if (25 < max_fds)
491 pack_max_fds = max_fds - 25;
492 else
493 pack_max_fds = 1;
494 }
495
496 while (pack_max_fds <= pack_open_fds && close_one_pack())
497 ; /* nothing */
498
499 p->pack_fd = git_open(p->pack_name);
500 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
501 return -1;
502 pack_open_fds++;
503
504 /* If we created the struct before we had the pack we lack size. */
505 if (!p->pack_size) {
506 if (!S_ISREG(st.st_mode))
507 return error("packfile %s not a regular file", p->pack_name);
508 p->pack_size = st.st_size;
509 } else if (p->pack_size != st.st_size)
510 return error("packfile %s size changed", p->pack_name);
511
512 /* We leave these file descriptors open with sliding mmap;
513 * there is no point keeping them open across exec(), though.
514 */
515 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
516 if (fd_flag < 0)
517 return error("cannot determine file descriptor flags");
518 fd_flag |= FD_CLOEXEC;
519 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
520 return error("cannot set FD_CLOEXEC");
521
522 /* Verify we recognize this pack file format. */
41dcc4dc
JK
523 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
524 if (read_result < 0)
525 return error_errno("error reading from %s", p->pack_name);
526 if (read_result != sizeof(hdr))
84f80ad5
JT
527 return error("file %s is far too short to be a packfile", p->pack_name);
528 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
529 return error("file %s is not a GIT packfile", p->pack_name);
530 if (!pack_version_ok(hdr.hdr_version))
531 return error("packfile %s is version %"PRIu32" and not"
532 " supported (try upgrading GIT to a newer version)",
533 p->pack_name, ntohl(hdr.hdr_version));
534
17c35c89
DS
535 /* Skip index checking if in multi-pack-index */
536 if (!p->index_data)
537 return 0;
538
84f80ad5
JT
539 /* Verify the pack matches its index. */
540 if (p->num_objects != ntohl(hdr.hdr_entries))
541 return error("packfile %s claims to have %"PRIu32" objects"
542 " while index indicates %"PRIu32" objects",
543 p->pack_name, ntohl(hdr.hdr_entries),
544 p->num_objects);
37fec86a 545 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
84f80ad5 546 return error("end of packfile %s is unavailable", p->pack_name);
37fec86a 547 read_result = read_in_full(p->pack_fd, hash, hashsz);
41dcc4dc
JK
548 if (read_result < 0)
549 return error_errno("error reading from %s", p->pack_name);
37fec86a 550 if (read_result != hashsz)
84f80ad5 551 return error("packfile %s signature is unavailable", p->pack_name);
37fec86a 552 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
553 if (hashcmp(hash, idx_hash))
84f80ad5
JT
554 return error("packfile %s does not match index", p->pack_name);
555 return 0;
556}
557
a2551953 558static int open_packed_git(struct packed_git *p)
84f80ad5
JT
559{
560 if (!open_packed_git_1(p))
561 return 0;
562 close_pack_fd(p);
563 return -1;
564}
565
566static int in_window(struct pack_window *win, off_t offset)
567{
37fec86a 568 /* We must promise at least one full hash after the
84f80ad5
JT
569 * offset is available from this window, otherwise the offset
570 * is not actually in this window and a different window (which
571 * has that one hash excess) must be used. This is to support
572 * the object header and delta base parsing routines below.
573 */
574 off_t win_off = win->offset;
575 return win_off <= offset
37fec86a 576 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
84f80ad5
JT
577}
578
579unsigned char *use_pack(struct packed_git *p,
580 struct pack_window **w_cursor,
581 off_t offset,
582 unsigned long *left)
583{
584 struct pack_window *win = *w_cursor;
585
586 /* Since packfiles end in a hash of their content and it's
587 * pointless to ask for an offset into the middle of that
588 * hash, and the in_window function above wouldn't match
589 * don't allow an offset too close to the end of the file.
590 */
591 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
592 die("packfile %s cannot be accessed", p->pack_name);
37fec86a 593 if (offset > (p->pack_size - the_hash_algo->rawsz))
84f80ad5
JT
594 die("offset beyond end of packfile (truncated pack?)");
595 if (offset < 0)
596 die(_("offset before end of packfile (broken .idx?)"));
597
598 if (!win || !in_window(win, offset)) {
599 if (win)
600 win->inuse_cnt--;
601 for (win = p->windows; win; win = win->next) {
602 if (in_window(win, offset))
603 break;
604 }
605 if (!win) {
606 size_t window_align = packed_git_window_size / 2;
607 off_t len;
608
609 if (p->pack_fd == -1 && open_packed_git(p))
610 die("packfile %s cannot be accessed", p->pack_name);
611
612 win = xcalloc(1, sizeof(*win));
613 win->offset = (offset / window_align) * window_align;
614 len = p->pack_size - win->offset;
615 if (len > packed_git_window_size)
616 len = packed_git_window_size;
617 win->len = (size_t)len;
618 pack_mapped += win->len;
619 while (packed_git_limit < pack_mapped
620 && unuse_one_window(p))
621 ; /* nothing */
622 win->base = xmmap(NULL, win->len,
623 PROT_READ, MAP_PRIVATE,
624 p->pack_fd, win->offset);
625 if (win->base == MAP_FAILED)
626 die_errno("packfile %s cannot be mapped",
627 p->pack_name);
628 if (!win->offset && win->len == p->pack_size
629 && !p->do_not_close)
630 close_pack_fd(p);
631 pack_mmap_calls++;
632 pack_open_windows++;
633 if (pack_mapped > peak_pack_mapped)
634 peak_pack_mapped = pack_mapped;
635 if (pack_open_windows > peak_pack_open_windows)
636 peak_pack_open_windows = pack_open_windows;
637 win->next = p->windows;
638 p->windows = win;
639 }
640 }
641 if (win != *w_cursor) {
642 win->last_used = pack_used_ctr++;
643 win->inuse_cnt++;
644 *w_cursor = win;
645 }
646 offset -= win->offset;
647 if (left)
648 *left = win->len - xsize_t(offset);
649 return win->base + offset;
650}
97de1803
JT
651
652void unuse_pack(struct pack_window **w_cursor)
653{
654 struct pack_window *w = *w_cursor;
655 if (w) {
656 w->inuse_cnt--;
657 *w_cursor = NULL;
658 }
659}
9a428653
JT
660
661static void try_to_free_pack_memory(size_t size)
662{
663 release_pack_memory(size);
664}
665
666struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
667{
668 static int have_set_try_to_free_routine;
669 struct stat st;
670 size_t alloc;
671 struct packed_git *p;
672
673 if (!have_set_try_to_free_routine) {
674 have_set_try_to_free_routine = 1;
675 set_try_to_free_routine(try_to_free_pack_memory);
676 }
677
678 /*
679 * Make sure a corresponding .pack file exists and that
680 * the index looks sane.
681 */
682 if (!strip_suffix_mem(path, &path_len, ".idx"))
683 return NULL;
684
685 /*
498f1f61 686 * ".promisor" is long enough to hold any suffix we're adding (and
9a428653
JT
687 * the use xsnprintf double-checks that)
688 */
498f1f61 689 alloc = st_add3(path_len, strlen(".promisor"), 1);
9a428653
JT
690 p = alloc_packed_git(alloc);
691 memcpy(p->pack_name, path, path_len);
692
693 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
694 if (!access(p->pack_name, F_OK))
695 p->pack_keep = 1;
696
498f1f61
JT
697 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
698 if (!access(p->pack_name, F_OK))
699 p->pack_promisor = 1;
700
9a428653
JT
701 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
702 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
703 free(p);
704 return NULL;
705 }
706
707 /* ok, it looks sane as far as we can check without
708 * actually mapping the pack file.
709 */
710 p->pack_size = st.st_size;
711 p->pack_local = local;
712 p->mtime = st.st_mtime;
37fec86a 713 if (path_len < the_hash_algo->hexsz ||
714 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->sha1))
9a428653
JT
715 hashclr(p->sha1);
716 return p;
717}
e65f1862 718
5babff16 719void install_packed_git(struct repository *r, struct packed_git *pack)
e65f1862
JT
720{
721 if (pack->pack_fd != -1)
722 pack_open_fds++;
723
5babff16
SB
724 pack->next = r->objects->packed_git;
725 r->objects->packed_git = pack;
e65f1862 726}
0abe14f6
JT
727
728void (*report_garbage)(unsigned seen_bits, const char *path);
729
730static void report_helper(const struct string_list *list,
731 int seen_bits, int first, int last)
732{
733 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
734 return;
735
736 for (; first < last; first++)
737 report_garbage(seen_bits, list->items[first].string);
738}
739
740static void report_pack_garbage(struct string_list *list)
741{
742 int i, baselen = -1, first = 0, seen_bits = 0;
743
744 if (!report_garbage)
745 return;
746
747 string_list_sort(list);
748
749 for (i = 0; i < list->nr; i++) {
750 const char *path = list->items[i].string;
751 if (baselen != -1 &&
752 strncmp(path, list->items[first].string, baselen)) {
753 report_helper(list, seen_bits, first, i);
754 baselen = -1;
755 seen_bits = 0;
756 }
757 if (baselen == -1) {
758 const char *dot = strrchr(path, '.');
759 if (!dot) {
760 report_garbage(PACKDIR_FILE_GARBAGE, path);
761 continue;
762 }
763 baselen = dot - path + 1;
764 first = i;
765 }
766 if (!strcmp(path + baselen, "pack"))
767 seen_bits |= 1;
768 else if (!strcmp(path + baselen, "idx"))
769 seen_bits |= 2;
770 }
771 report_helper(list, seen_bits, first, list->nr);
772}
773
9208e318
DS
774void for_each_file_in_pack_dir(const char *objdir,
775 each_file_in_pack_dir_fn fn,
776 void *data)
0abe14f6
JT
777{
778 struct strbuf path = STRBUF_INIT;
779 size_t dirnamelen;
780 DIR *dir;
781 struct dirent *de;
0abe14f6
JT
782
783 strbuf_addstr(&path, objdir);
784 strbuf_addstr(&path, "/pack");
785 dir = opendir(path.buf);
786 if (!dir) {
787 if (errno != ENOENT)
788 error_errno("unable to open object pack directory: %s",
789 path.buf);
790 strbuf_release(&path);
791 return;
792 }
793 strbuf_addch(&path, '/');
794 dirnamelen = path.len;
795 while ((de = readdir(dir)) != NULL) {
0abe14f6
JT
796 if (is_dot_or_dotdot(de->d_name))
797 continue;
798
799 strbuf_setlen(&path, dirnamelen);
800 strbuf_addstr(&path, de->d_name);
801
9208e318 802 fn(path.buf, path.len, de->d_name, data);
0abe14f6 803 }
9208e318 804
0abe14f6 805 closedir(dir);
0abe14f6
JT
806 strbuf_release(&path);
807}
808
9208e318
DS
809struct prepare_pack_data {
810 struct repository *r;
811 struct string_list *garbage;
812 int local;
f3a002bd 813 struct multi_pack_index *m;
9208e318
DS
814};
815
816static void prepare_pack(const char *full_name, size_t full_name_len,
817 const char *file_name, void *_data)
818{
819 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
820 struct packed_git *p;
821 size_t base_len = full_name_len;
822
fe86c3be
DS
823 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
824 !(data->m && midx_contains_pack(data->m, file_name))) {
9208e318
DS
825 /* Don't reopen a pack we already have. */
826 for (p = data->r->objects->packed_git; p; p = p->next) {
827 size_t len;
828 if (strip_suffix(p->pack_name, ".pack", &len) &&
829 len == base_len &&
830 !memcmp(p->pack_name, full_name, len))
831 break;
832 }
833
834 if (!p) {
835 p = add_packed_git(full_name, full_name_len, data->local);
836 if (p)
837 install_packed_git(data->r, p);
838 }
839 }
840
841 if (!report_garbage)
842 return;
843
fe86c3be
DS
844 if (!strcmp(file_name, "multi-pack-index"))
845 return;
9208e318
DS
846 if (ends_with(file_name, ".idx") ||
847 ends_with(file_name, ".pack") ||
848 ends_with(file_name, ".bitmap") ||
849 ends_with(file_name, ".keep") ||
850 ends_with(file_name, ".promisor"))
851 string_list_append(data->garbage, full_name);
852 else
853 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
854}
855
856static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
857{
858 struct prepare_pack_data data;
859 struct string_list garbage = STRING_LIST_INIT_DUP;
860
f3a002bd
DS
861 data.m = r->objects->multi_pack_index;
862
863 /* look for the multi-pack-index for this object directory */
864 while (data.m && strcmp(data.m->object_dir, objdir))
865 data.m = data.m->next;
866
9208e318
DS
867 data.r = r;
868 data.garbage = &garbage;
869 data.local = local;
870
871 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
872
873 report_pack_garbage(data.garbage);
874 string_list_clear(data.garbage, 0);
875}
876
464416a2 877static void prepare_packed_git(struct repository *r);
0abe14f6
JT
878/*
879 * Give a fast, rough count of the number of objects in the repository. This
880 * ignores loose objects completely. If you have a lot of them, then either
881 * you should repack because your performance will be awful, or they are
882 * all unreachable objects about to be pruned, in which case they're not really
883 * interesting as a measure of repo size in the first place.
884 */
885unsigned long approximate_object_count(void)
886{
9a00580d
SB
887 if (!the_repository->objects->approximate_object_count_valid) {
888 unsigned long count;
b8990fbf 889 struct multi_pack_index *m;
0abe14f6
JT
890 struct packed_git *p;
891
6fdb4e9f 892 prepare_packed_git(the_repository);
0abe14f6 893 count = 0;
b8990fbf
DS
894 for (m = get_multi_pack_index(the_repository); m; m = m->next)
895 count += m->num_objects;
a80d72db 896 for (p = the_repository->objects->packed_git; p; p = p->next) {
0abe14f6
JT
897 if (open_pack_index(p))
898 continue;
899 count += p->num_objects;
900 }
9a00580d 901 the_repository->objects->approximate_object_count = count;
0abe14f6 902 }
9a00580d 903 return the_repository->objects->approximate_object_count;
0abe14f6
JT
904}
905
906static void *get_next_packed_git(const void *p)
907{
908 return ((const struct packed_git *)p)->next;
909}
910
911static void set_next_packed_git(void *p, void *next)
912{
913 ((struct packed_git *)p)->next = next;
914}
915
916static int sort_pack(const void *a_, const void *b_)
917{
918 const struct packed_git *a = a_;
919 const struct packed_git *b = b_;
920 int st;
921
922 /*
923 * Local packs tend to contain objects specific to our
924 * variant of the project than remote ones. In addition,
925 * remote ones could be on a network mounted filesystem.
926 * Favor local ones for these reasons.
927 */
928 st = a->pack_local - b->pack_local;
929 if (st)
930 return -st;
931
932 /*
933 * Younger packs tend to contain more recent objects,
934 * and more recent objects tend to get accessed more
935 * often.
936 */
937 if (a->mtime < b->mtime)
938 return 1;
939 else if (a->mtime == b->mtime)
940 return 0;
941 return -1;
942}
943
c235beac 944static void rearrange_packed_git(struct repository *r)
0abe14f6 945{
c235beac
SB
946 r->objects->packed_git = llist_mergesort(
947 r->objects->packed_git, get_next_packed_git,
a80d72db 948 set_next_packed_git, sort_pack);
0abe14f6
JT
949}
950
804be796 951static void prepare_packed_git_mru(struct repository *r)
0abe14f6
JT
952{
953 struct packed_git *p;
954
804be796 955 INIT_LIST_HEAD(&r->objects->packed_git_mru);
ec2dd32c 956
804be796
SB
957 for (p = r->objects->packed_git; p; p = p->next)
958 list_add_tail(&p->mru, &r->objects->packed_git_mru);
0abe14f6
JT
959}
960
464416a2 961static void prepare_packed_git(struct repository *r)
0abe14f6
JT
962{
963 struct alternate_object_database *alt;
964
0f90a9f2 965 if (r->objects->packed_git_initialized)
0abe14f6 966 return;
2cf489a3 967 prepare_multi_pack_index_one(r, r->objects->objectdir, 1);
0f90a9f2
SB
968 prepare_packed_git_one(r, r->objects->objectdir, 1);
969 prepare_alt_odb(r);
c4d25228 970 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
2cf489a3 971 prepare_multi_pack_index_one(r, alt->path, 0);
0f90a9f2 972 prepare_packed_git_one(r, alt->path, 0);
c4d25228 973 }
0f90a9f2
SB
974 rearrange_packed_git(r);
975 prepare_packed_git_mru(r);
976 r->objects->packed_git_initialized = 1;
0abe14f6
JT
977}
978
4c2a13b4 979void reprepare_packed_git(struct repository *r)
0abe14f6 980{
4c2a13b4
SB
981 r->objects->approximate_object_count_valid = 0;
982 r->objects->packed_git_initialized = 0;
983 prepare_packed_git(r);
0abe14f6 984}
32b42e15 985
a80d72db
SB
986struct packed_git *get_packed_git(struct repository *r)
987{
464416a2 988 prepare_packed_git(r);
a80d72db
SB
989 return r->objects->packed_git;
990}
991
8aac67a1
DS
992struct multi_pack_index *get_multi_pack_index(struct repository *r)
993{
994 prepare_packed_git(r);
995 return r->objects->multi_pack_index;
996}
997
a80d72db
SB
998struct list_head *get_packed_git_mru(struct repository *r)
999{
464416a2 1000 prepare_packed_git(r);
a80d72db
SB
1001 return &r->objects->packed_git_mru;
1002}
1003
32b42e15
JT
1004unsigned long unpack_object_header_buffer(const unsigned char *buf,
1005 unsigned long len, enum object_type *type, unsigned long *sizep)
1006{
1007 unsigned shift;
1008 unsigned long size, c;
1009 unsigned long used = 0;
1010
1011 c = buf[used++];
1012 *type = (c >> 4) & 7;
1013 size = c & 15;
1014 shift = 4;
1015 while (c & 0x80) {
1016 if (len <= used || bitsizeof(long) <= shift) {
1017 error("bad object header");
1018 size = used = 0;
1019 break;
1020 }
1021 c = buf[used++];
1022 size += (c & 0x7f) << shift;
1023 shift += 7;
1024 }
1025 *sizep = size;
1026 return used;
1027}
7b3aa75d
JT
1028
1029unsigned long get_size_from_delta(struct packed_git *p,
1030 struct pack_window **w_curs,
1031 off_t curpos)
1032{
1033 const unsigned char *data;
1034 unsigned char delta_head[20], *in;
1035 git_zstream stream;
1036 int st;
1037
1038 memset(&stream, 0, sizeof(stream));
1039 stream.next_out = delta_head;
1040 stream.avail_out = sizeof(delta_head);
1041
1042 git_inflate_init(&stream);
1043 do {
1044 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1045 stream.next_in = in;
1046 st = git_inflate(&stream, Z_FINISH);
1047 curpos += stream.next_in - in;
1048 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1049 stream.total_out < sizeof(delta_head));
1050 git_inflate_end(&stream);
1051 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1052 error("delta data unpack-initial failed");
1053 return 0;
1054 }
1055
1056 /* Examine the initial part of the delta to figure out
1057 * the result size.
1058 */
1059 data = delta_head;
1060
1061 /* ignore base size */
1062 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1063
1064 /* Read the result size */
1065 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1066}
3588dd6e
JT
1067
1068int unpack_object_header(struct packed_git *p,
1069 struct pack_window **w_curs,
1070 off_t *curpos,
1071 unsigned long *sizep)
1072{
1073 unsigned char *base;
1074 unsigned long left;
1075 unsigned long used;
1076 enum object_type type;
1077
1078 /* use_pack() assures us we have [base, base + 20) available
1079 * as a range that we can look at. (Its actually the hash
1080 * size that is assured.) With our object header encoding
1081 * the maximum deflated object size is 2^137, which is just
1082 * insane, so we know won't exceed what we have been given.
1083 */
1084 base = use_pack(p, w_curs, *curpos, &left);
1085 used = unpack_object_header_buffer(base, left, &type, sizep);
1086 if (!used) {
1087 type = OBJ_BAD;
1088 } else
1089 *curpos += used;
1090
1091 return type;
1092}
f1d8130b
JT
1093
1094void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1095{
1096 unsigned i;
1097 for (i = 0; i < p->num_bad_objects; i++)
1098 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1099 return;
1100 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1101 st_mult(GIT_MAX_RAWSZ,
1102 st_add(p->num_bad_objects, 1)));
1103 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1104 p->num_bad_objects++;
1105}
1106
1107const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1108{
1109 struct packed_git *p;
1110 unsigned i;
1111
a80d72db 1112 for (p = the_repository->objects->packed_git; p; p = p->next)
f1d8130b 1113 for (i = 0; i < p->num_bad_objects; i++)
37fec86a 1114 if (!hashcmp(sha1,
1115 p->bad_object_sha1 + the_hash_algo->rawsz * i))
f1d8130b
JT
1116 return p;
1117 return NULL;
1118}
1119
1120static off_t get_delta_base(struct packed_git *p,
1121 struct pack_window **w_curs,
1122 off_t *curpos,
1123 enum object_type type,
1124 off_t delta_obj_offset)
1125{
1126 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1127 off_t base_offset;
1128
1129 /* use_pack() assured us we have [base_info, base_info + 20)
1130 * as a range that we can look at without walking off the
1131 * end of the mapped window. Its actually the hash size
1132 * that is assured. An OFS_DELTA longer than the hash size
1133 * is stupid, as then a REF_DELTA would be smaller to store.
1134 */
1135 if (type == OBJ_OFS_DELTA) {
1136 unsigned used = 0;
1137 unsigned char c = base_info[used++];
1138 base_offset = c & 127;
1139 while (c & 128) {
1140 base_offset += 1;
1141 if (!base_offset || MSB(base_offset, 7))
1142 return 0; /* overflow */
1143 c = base_info[used++];
1144 base_offset = (base_offset << 7) + (c & 127);
1145 }
1146 base_offset = delta_obj_offset - base_offset;
1147 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1148 return 0; /* out of bound */
1149 *curpos += used;
1150 } else if (type == OBJ_REF_DELTA) {
1151 /* The base entry _must_ be in the same pack */
1152 base_offset = find_pack_entry_one(base_info, p);
37fec86a 1153 *curpos += the_hash_algo->rawsz;
f1d8130b
JT
1154 } else
1155 die("I am totally screwed");
1156 return base_offset;
1157}
1158
1159/*
1160 * Like get_delta_base above, but we return the sha1 instead of the pack
1161 * offset. This means it is cheaper for REF deltas (we do not have to do
1162 * the final object lookup), but more expensive for OFS deltas (we
1163 * have to load the revidx to convert the offset back into a sha1).
1164 */
1165static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1166 struct pack_window **w_curs,
1167 off_t curpos,
1168 enum object_type type,
1169 off_t delta_obj_offset)
1170{
1171 if (type == OBJ_REF_DELTA) {
1172 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1173 return base;
1174 } else if (type == OBJ_OFS_DELTA) {
1175 struct revindex_entry *revidx;
1176 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1177 type, delta_obj_offset);
1178
1179 if (!base_offset)
1180 return NULL;
1181
1182 revidx = find_pack_revindex(p, base_offset);
1183 if (!revidx)
1184 return NULL;
1185
1186 return nth_packed_object_sha1(p, revidx->nr);
1187 } else
1188 return NULL;
1189}
1190
9d98354f
SB
1191static int retry_bad_packed_offset(struct repository *r,
1192 struct packed_git *p,
1193 off_t obj_offset)
f1d8130b
JT
1194{
1195 int type;
1196 struct revindex_entry *revidx;
d169d664 1197 struct object_id oid;
f1d8130b
JT
1198 revidx = find_pack_revindex(p, obj_offset);
1199 if (!revidx)
1200 return OBJ_BAD;
d169d664 1201 nth_packed_object_oid(&oid, p, revidx->nr);
1202 mark_bad_packed_object(p, oid.hash);
9d98354f 1203 type = oid_object_info(r, &oid, NULL);
f1d8130b
JT
1204 if (type <= OBJ_NONE)
1205 return OBJ_BAD;
1206 return type;
1207}
1208
1209#define POI_STACK_PREALLOC 64
1210
9d98354f
SB
1211static enum object_type packed_to_object_type(struct repository *r,
1212 struct packed_git *p,
f1d8130b
JT
1213 off_t obj_offset,
1214 enum object_type type,
1215 struct pack_window **w_curs,
1216 off_t curpos)
1217{
1218 off_t small_poi_stack[POI_STACK_PREALLOC];
1219 off_t *poi_stack = small_poi_stack;
1220 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1221
1222 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1223 off_t base_offset;
1224 unsigned long size;
1225 /* Push the object we're going to leave behind */
1226 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1227 poi_stack_alloc = alloc_nr(poi_stack_nr);
1228 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1229 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1230 } else {
1231 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1232 }
1233 poi_stack[poi_stack_nr++] = obj_offset;
1234 /* If parsing the base offset fails, just unwind */
1235 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1236 if (!base_offset)
1237 goto unwind;
1238 curpos = obj_offset = base_offset;
1239 type = unpack_object_header(p, w_curs, &curpos, &size);
1240 if (type <= OBJ_NONE) {
1241 /* If getting the base itself fails, we first
1242 * retry the base, otherwise unwind */
9d98354f 1243 type = retry_bad_packed_offset(r, p, base_offset);
f1d8130b
JT
1244 if (type > OBJ_NONE)
1245 goto out;
1246 goto unwind;
1247 }
1248 }
1249
1250 switch (type) {
1251 case OBJ_BAD:
1252 case OBJ_COMMIT:
1253 case OBJ_TREE:
1254 case OBJ_BLOB:
1255 case OBJ_TAG:
1256 break;
1257 default:
1258 error("unknown object type %i at offset %"PRIuMAX" in %s",
1259 type, (uintmax_t)obj_offset, p->pack_name);
1260 type = OBJ_BAD;
1261 }
1262
1263out:
1264 if (poi_stack != small_poi_stack)
1265 free(poi_stack);
1266 return type;
1267
1268unwind:
1269 while (poi_stack_nr) {
1270 obj_offset = poi_stack[--poi_stack_nr];
9d98354f 1271 type = retry_bad_packed_offset(r, p, obj_offset);
f1d8130b
JT
1272 if (type > OBJ_NONE)
1273 goto out;
1274 }
1275 type = OBJ_BAD;
1276 goto out;
1277}
1278
1279static struct hashmap delta_base_cache;
1280static size_t delta_base_cached;
1281
1282static LIST_HEAD(delta_base_cache_lru);
1283
1284struct delta_base_cache_key {
1285 struct packed_git *p;
1286 off_t base_offset;
1287};
1288
1289struct delta_base_cache_entry {
1290 struct hashmap hash;
1291 struct delta_base_cache_key key;
1292 struct list_head lru;
1293 void *data;
1294 unsigned long size;
1295 enum object_type type;
1296};
1297
1298static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1299{
1300 unsigned int hash;
1301
1302 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1303 hash += (hash >> 8) + (hash >> 16);
1304 return hash;
1305}
1306
1307static struct delta_base_cache_entry *
1308get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1309{
1310 struct hashmap_entry entry;
1311 struct delta_base_cache_key key;
1312
1313 if (!delta_base_cache.cmpfn)
1314 return NULL;
1315
1316 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1317 key.p = p;
1318 key.base_offset = base_offset;
1319 return hashmap_get(&delta_base_cache, &entry, &key);
1320}
1321
1322static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1323 const struct delta_base_cache_key *b)
1324{
1325 return a->p == b->p && a->base_offset == b->base_offset;
1326}
1327
1328static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1329 const void *va, const void *vb,
1330 const void *vkey)
1331{
1332 const struct delta_base_cache_entry *a = va, *b = vb;
1333 const struct delta_base_cache_key *key = vkey;
1334 if (key)
1335 return !delta_base_cache_key_eq(&a->key, key);
1336 else
1337 return !delta_base_cache_key_eq(&a->key, &b->key);
1338}
1339
1340static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1341{
1342 return !!get_delta_base_cache_entry(p, base_offset);
1343}
1344
1345/*
1346 * Remove the entry from the cache, but do _not_ free the associated
1347 * entry data. The caller takes ownership of the "data" buffer, and
1348 * should copy out any fields it wants before detaching.
1349 */
1350static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1351{
1352 hashmap_remove(&delta_base_cache, ent, &ent->key);
1353 list_del(&ent->lru);
1354 delta_base_cached -= ent->size;
1355 free(ent);
1356}
1357
9d98354f
SB
1358static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1359 off_t base_offset, unsigned long *base_size,
1360 enum object_type *type)
f1d8130b
JT
1361{
1362 struct delta_base_cache_entry *ent;
1363
1364 ent = get_delta_base_cache_entry(p, base_offset);
1365 if (!ent)
9d98354f 1366 return unpack_entry(r, p, base_offset, type, base_size);
f1d8130b
JT
1367
1368 if (type)
1369 *type = ent->type;
1370 if (base_size)
1371 *base_size = ent->size;
1372 return xmemdupz(ent->data, ent->size);
1373}
1374
1375static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1376{
1377 free(ent->data);
1378 detach_delta_base_cache_entry(ent);
1379}
1380
1381void clear_delta_base_cache(void)
1382{
1383 struct list_head *lru, *tmp;
1384 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1385 struct delta_base_cache_entry *entry =
1386 list_entry(lru, struct delta_base_cache_entry, lru);
1387 release_delta_base_cache(entry);
1388 }
1389}
1390
1391static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1392 void *base, unsigned long base_size, enum object_type type)
1393{
1394 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1395 struct list_head *lru, *tmp;
1396
1397 delta_base_cached += base_size;
1398
1399 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1400 struct delta_base_cache_entry *f =
1401 list_entry(lru, struct delta_base_cache_entry, lru);
1402 if (delta_base_cached <= delta_base_cache_limit)
1403 break;
1404 release_delta_base_cache(f);
1405 }
1406
1407 ent->key.p = p;
1408 ent->key.base_offset = base_offset;
1409 ent->type = type;
1410 ent->data = base;
1411 ent->size = base_size;
1412 list_add_tail(&ent->lru, &delta_base_cache_lru);
1413
1414 if (!delta_base_cache.cmpfn)
1415 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1416 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1417 hashmap_add(&delta_base_cache, ent);
1418}
1419
9d98354f
SB
1420int packed_object_info(struct repository *r, struct packed_git *p,
1421 off_t obj_offset, struct object_info *oi)
f1d8130b
JT
1422{
1423 struct pack_window *w_curs = NULL;
1424 unsigned long size;
1425 off_t curpos = obj_offset;
1426 enum object_type type;
1427
1428 /*
1429 * We always get the representation type, but only convert it to
1430 * a "real" type later if the caller is interested.
1431 */
1432 if (oi->contentp) {
9d98354f 1433 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
f1d8130b
JT
1434 &type);
1435 if (!*oi->contentp)
1436 type = OBJ_BAD;
1437 } else {
1438 type = unpack_object_header(p, &w_curs, &curpos, &size);
1439 }
1440
1441 if (!oi->contentp && oi->sizep) {
1442 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1443 off_t tmp_pos = curpos;
1444 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1445 type, obj_offset);
1446 if (!base_offset) {
1447 type = OBJ_BAD;
1448 goto out;
1449 }
1450 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1451 if (*oi->sizep == 0) {
1452 type = OBJ_BAD;
1453 goto out;
1454 }
1455 } else {
1456 *oi->sizep = size;
1457 }
1458 }
1459
1460 if (oi->disk_sizep) {
1461 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1462 *oi->disk_sizep = revidx[1].offset - obj_offset;
1463 }
1464
6ca32f47 1465 if (oi->typep || oi->type_name) {
f1d8130b 1466 enum object_type ptot;
9d98354f 1467 ptot = packed_to_object_type(r, p, obj_offset,
144f4948 1468 type, &w_curs, curpos);
f1d8130b
JT
1469 if (oi->typep)
1470 *oi->typep = ptot;
6ca32f47 1471 if (oi->type_name) {
debca9d2 1472 const char *tn = type_name(ptot);
f1d8130b 1473 if (tn)
6ca32f47 1474 strbuf_addstr(oi->type_name, tn);
f1d8130b
JT
1475 }
1476 if (ptot < 0) {
1477 type = OBJ_BAD;
1478 goto out;
1479 }
1480 }
1481
1482 if (oi->delta_base_sha1) {
1483 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1484 const unsigned char *base;
1485
1486 base = get_delta_base_sha1(p, &w_curs, curpos,
1487 type, obj_offset);
1488 if (!base) {
1489 type = OBJ_BAD;
1490 goto out;
1491 }
1492
1493 hashcpy(oi->delta_base_sha1, base);
1494 } else
1495 hashclr(oi->delta_base_sha1);
1496 }
1497
1498 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1499 OI_PACKED;
1500
1501out:
1502 unuse_pack(&w_curs);
1503 return type;
1504}
1505
1506static void *unpack_compressed_entry(struct packed_git *p,
1507 struct pack_window **w_curs,
1508 off_t curpos,
1509 unsigned long size)
1510{
1511 int st;
1512 git_zstream stream;
1513 unsigned char *buffer, *in;
1514
1515 buffer = xmallocz_gently(size);
1516 if (!buffer)
1517 return NULL;
1518 memset(&stream, 0, sizeof(stream));
1519 stream.next_out = buffer;
1520 stream.avail_out = size + 1;
1521
1522 git_inflate_init(&stream);
1523 do {
1524 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1525 stream.next_in = in;
1526 st = git_inflate(&stream, Z_FINISH);
1527 if (!stream.avail_out)
1528 break; /* the payload is larger than it should be */
1529 curpos += stream.next_in - in;
1530 } while (st == Z_OK || st == Z_BUF_ERROR);
1531 git_inflate_end(&stream);
1532 if ((st != Z_STREAM_END) || stream.total_out != size) {
1533 free(buffer);
1534 return NULL;
1535 }
1536
b611396e
JL
1537 /* versions of zlib can clobber unconsumed portion of outbuf */
1538 buffer[size] = '\0';
1539
f1d8130b
JT
1540 return buffer;
1541}
1542
1543static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1544{
1545 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1546 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1547 p->pack_name, (uintmax_t)obj_offset);
1548}
1549
1550int do_check_packed_object_crc;
1551
1552#define UNPACK_ENTRY_STACK_PREALLOC 64
1553struct unpack_entry_stack_ent {
1554 off_t obj_offset;
1555 off_t curpos;
1556 unsigned long size;
1557};
1558
9d98354f
SB
1559static void *read_object(struct repository *r,
1560 const struct object_id *oid,
1561 enum object_type *type,
f1d8130b
JT
1562 unsigned long *size)
1563{
1564 struct object_info oi = OBJECT_INFO_INIT;
1565 void *content;
1566 oi.typep = type;
1567 oi.sizep = size;
1568 oi.contentp = &content;
1569
9d98354f 1570 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
f1d8130b
JT
1571 return NULL;
1572 return content;
1573}
1574
9d98354f 1575void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
f1d8130b
JT
1576 enum object_type *final_type, unsigned long *final_size)
1577{
1578 struct pack_window *w_curs = NULL;
1579 off_t curpos = obj_offset;
1580 void *data = NULL;
1581 unsigned long size;
1582 enum object_type type;
1583 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1584 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1585 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1586 int base_from_cache = 0;
1587
1588 write_pack_access_log(p, obj_offset);
1589
1590 /* PHASE 1: drill down to the innermost base object */
1591 for (;;) {
1592 off_t base_offset;
1593 int i;
1594 struct delta_base_cache_entry *ent;
1595
1596 ent = get_delta_base_cache_entry(p, curpos);
1597 if (ent) {
1598 type = ent->type;
1599 data = ent->data;
1600 size = ent->size;
1601 detach_delta_base_cache_entry(ent);
1602 base_from_cache = 1;
1603 break;
1604 }
1605
1606 if (do_check_packed_object_crc && p->index_version > 1) {
1607 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1608 off_t len = revidx[1].offset - obj_offset;
1609 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
4310b0c4 1610 struct object_id oid;
1611 nth_packed_object_oid(&oid, p, revidx->nr);
f1d8130b 1612 error("bad packed object CRC for %s",
4310b0c4 1613 oid_to_hex(&oid));
1614 mark_bad_packed_object(p, oid.hash);
f1d8130b
JT
1615 data = NULL;
1616 goto out;
1617 }
1618 }
1619
1620 type = unpack_object_header(p, &w_curs, &curpos, &size);
1621 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1622 break;
1623
1624 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1625 if (!base_offset) {
1626 error("failed to validate delta base reference "
1627 "at offset %"PRIuMAX" from %s",
1628 (uintmax_t)curpos, p->pack_name);
1629 /* bail to phase 2, in hopes of recovery */
1630 data = NULL;
1631 break;
1632 }
1633
1634 /* push object, proceed to base */
1635 if (delta_stack_nr >= delta_stack_alloc
1636 && delta_stack == small_delta_stack) {
1637 delta_stack_alloc = alloc_nr(delta_stack_nr);
1638 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1639 memcpy(delta_stack, small_delta_stack,
1640 sizeof(*delta_stack)*delta_stack_nr);
1641 } else {
1642 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1643 }
1644 i = delta_stack_nr++;
1645 delta_stack[i].obj_offset = obj_offset;
1646 delta_stack[i].curpos = curpos;
1647 delta_stack[i].size = size;
1648
1649 curpos = obj_offset = base_offset;
1650 }
1651
1652 /* PHASE 2: handle the base */
1653 switch (type) {
1654 case OBJ_OFS_DELTA:
1655 case OBJ_REF_DELTA:
1656 if (data)
033abf97 1657 BUG("unpack_entry: left loop at a valid delta");
f1d8130b
JT
1658 break;
1659 case OBJ_COMMIT:
1660 case OBJ_TREE:
1661 case OBJ_BLOB:
1662 case OBJ_TAG:
1663 if (!base_from_cache)
1664 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1665 break;
1666 default:
1667 data = NULL;
1668 error("unknown object type %i at offset %"PRIuMAX" in %s",
1669 type, (uintmax_t)obj_offset, p->pack_name);
1670 }
1671
1672 /* PHASE 3: apply deltas in order */
1673
1674 /* invariants:
1675 * 'data' holds the base data, or NULL if there was corruption
1676 */
1677 while (delta_stack_nr) {
1678 void *delta_data;
1679 void *base = data;
1680 void *external_base = NULL;
1681 unsigned long delta_size, base_size = size;
1682 int i;
1683
1684 data = NULL;
1685
1686 if (base)
1687 add_delta_base_cache(p, obj_offset, base, base_size, type);
1688
1689 if (!base) {
1690 /*
1691 * We're probably in deep shit, but let's try to fetch
1692 * the required base anyway from another pack or loose.
1693 * This is costly but should happen only in the presence
1694 * of a corrupted pack, and is better than failing outright.
1695 */
1696 struct revindex_entry *revidx;
4310b0c4 1697 struct object_id base_oid;
f1d8130b
JT
1698 revidx = find_pack_revindex(p, obj_offset);
1699 if (revidx) {
4310b0c4 1700 nth_packed_object_oid(&base_oid, p, revidx->nr);
f1d8130b
JT
1701 error("failed to read delta base object %s"
1702 " at offset %"PRIuMAX" from %s",
4310b0c4 1703 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
f1d8130b 1704 p->pack_name);
4310b0c4 1705 mark_bad_packed_object(p, base_oid.hash);
9d98354f 1706 base = read_object(r, &base_oid, &type, &base_size);
f1d8130b
JT
1707 external_base = base;
1708 }
1709 }
1710
1711 i = --delta_stack_nr;
1712 obj_offset = delta_stack[i].obj_offset;
1713 curpos = delta_stack[i].curpos;
1714 delta_size = delta_stack[i].size;
1715
1716 if (!base)
1717 continue;
1718
1719 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1720
1721 if (!delta_data) {
1722 error("failed to unpack compressed delta "
1723 "at offset %"PRIuMAX" from %s",
1724 (uintmax_t)curpos, p->pack_name);
1725 data = NULL;
1726 free(external_base);
1727 continue;
1728 }
1729
1730 data = patch_delta(base, base_size,
1731 delta_data, delta_size,
1732 &size);
1733
1734 /*
1735 * We could not apply the delta; warn the user, but keep going.
1736 * Our failure will be noticed either in the next iteration of
1737 * the loop, or if this is the final delta, in the caller when
1738 * we return NULL. Those code paths will take care of making
1739 * a more explicit warning and retrying with another copy of
1740 * the object.
1741 */
1742 if (!data)
1743 error("failed to apply delta");
1744
1745 free(delta_data);
1746 free(external_base);
1747 }
1748
1749 if (final_type)
1750 *final_type = type;
1751 if (final_size)
1752 *final_size = size;
1753
1754out:
1755 unuse_pack(&w_curs);
1756
1757 if (delta_stack != small_delta_stack)
1758 free(delta_stack);
1759
1760 return data;
1761}
d5a16761 1762
3d475f46
DS
1763int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1764{
1765 const unsigned char *index_fanout = p->index_data;
1766 const unsigned char *index_lookup;
37fec86a 1767 const unsigned int hashsz = the_hash_algo->rawsz;
3d475f46
DS
1768 int index_lookup_width;
1769
1770 if (!index_fanout)
1771 BUG("bsearch_pack called without a valid pack-index");
1772
1773 index_lookup = index_fanout + 4 * 256;
1774 if (p->index_version == 1) {
37fec86a 1775 index_lookup_width = hashsz + 4;
3d475f46
DS
1776 index_lookup += 4;
1777 } else {
37fec86a 1778 index_lookup_width = hashsz;
3d475f46
DS
1779 index_fanout += 8;
1780 index_lookup += 8;
1781 }
1782
1783 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1784 index_lookup, index_lookup_width, result);
1785}
1786
d5a16761
JT
1787const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1788 uint32_t n)
1789{
1790 const unsigned char *index = p->index_data;
37fec86a 1791 const unsigned int hashsz = the_hash_algo->rawsz;
d5a16761
JT
1792 if (!index) {
1793 if (open_pack_index(p))
1794 return NULL;
1795 index = p->index_data;
1796 }
1797 if (n >= p->num_objects)
1798 return NULL;
1799 index += 4 * 256;
1800 if (p->index_version == 1) {
37fec86a 1801 return index + (hashsz + 4) * n + 4;
d5a16761
JT
1802 } else {
1803 index += 8;
37fec86a 1804 return index + hashsz * n;
d5a16761
JT
1805 }
1806}
1807
1808const struct object_id *nth_packed_object_oid(struct object_id *oid,
1809 struct packed_git *p,
1810 uint32_t n)
1811{
1812 const unsigned char *hash = nth_packed_object_sha1(p, n);
1813 if (!hash)
1814 return NULL;
1815 hashcpy(oid->hash, hash);
1816 return oid;
1817}
9e0f45f5
JT
1818
1819void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1820{
1821 const unsigned char *ptr = vptr;
1822 const unsigned char *start = p->index_data;
1823 const unsigned char *end = start + p->index_size;
1824 if (ptr < start)
1825 die(_("offset before start of pack index for %s (corrupt index?)"),
1826 p->pack_name);
1827 /* No need to check for underflow; .idx files must be at least 8 bytes */
1828 if (ptr >= end - 8)
1829 die(_("offset beyond end of pack index for %s (truncated index?)"),
1830 p->pack_name);
1831}
1832
1833off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1834{
1835 const unsigned char *index = p->index_data;
37fec86a 1836 const unsigned int hashsz = the_hash_algo->rawsz;
9e0f45f5
JT
1837 index += 4 * 256;
1838 if (p->index_version == 1) {
37fec86a 1839 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
9e0f45f5
JT
1840 } else {
1841 uint32_t off;
37fec86a 1842 index += 8 + p->num_objects * (hashsz + 4);
9e0f45f5
JT
1843 off = ntohl(*((uint32_t *)(index + 4 * n)));
1844 if (!(off & 0x80000000))
1845 return off;
1846 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1847 check_pack_index_ptr(p, index);
ad622a25 1848 return get_be64(index);
9e0f45f5
JT
1849 }
1850}
a2551953
JT
1851
1852off_t find_pack_entry_one(const unsigned char *sha1,
1853 struct packed_git *p)
1854{
a2551953 1855 const unsigned char *index = p->index_data;
3d475f46 1856 struct object_id oid;
b4e00f73 1857 uint32_t result;
a2551953
JT
1858
1859 if (!index) {
1860 if (open_pack_index(p))
1861 return 0;
a2551953
JT
1862 }
1863
3d475f46
DS
1864 hashcpy(oid.hash, sha1);
1865 if (bsearch_pack(&oid, p, &result))
b4e00f73 1866 return nth_packed_object_offset(p, result);
a2551953
JT
1867 return 0;
1868}
1869
1870int is_pack_valid(struct packed_git *p)
1871{
1872 /* An already open pack is known to be valid. */
1873 if (p->pack_fd != -1)
1874 return 1;
1875
1876 /* If the pack has one window completely covering the
1877 * file size, the pack is known to be valid even if
1878 * the descriptor is not currently open.
1879 */
1880 if (p->windows) {
1881 struct pack_window *w = p->windows;
1882
1883 if (!w->offset && w->len == p->pack_size)
1884 return 1;
1885 }
1886
1887 /* Force the pack to open to prove its valid. */
1888 return !open_packed_git(p);
1889}
d6fe0036
JT
1890
1891struct packed_git *find_sha1_pack(const unsigned char *sha1,
1892 struct packed_git *packs)
1893{
1894 struct packed_git *p;
1895
1896 for (p = packs; p; p = p->next) {
1897 if (find_pack_entry_one(sha1, p))
1898 return p;
1899 }
1900 return NULL;
1901
1902}
1a1e5d4f 1903
544443cb 1904static int fill_pack_entry(const struct object_id *oid,
1a1e5d4f
JT
1905 struct pack_entry *e,
1906 struct packed_git *p)
1907{
1908 off_t offset;
1909
1910 if (p->num_bad_objects) {
1911 unsigned i;
1912 for (i = 0; i < p->num_bad_objects; i++)
37fec86a 1913 if (!hashcmp(oid->hash,
1914 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1a1e5d4f
JT
1915 return 0;
1916 }
1917
544443cb 1918 offset = find_pack_entry_one(oid->hash, p);
1a1e5d4f
JT
1919 if (!offset)
1920 return 0;
1921
1922 /*
1923 * We are about to tell the caller where they can locate the
1924 * requested object. We better make sure the packfile is
1925 * still here and can be accessed before supplying that
1926 * answer, as it may have been deleted since the index was
1927 * loaded!
1928 */
1929 if (!is_pack_valid(p))
1930 return 0;
1931 e->offset = offset;
1932 e->p = p;
1a1e5d4f
JT
1933 return 1;
1934}
1935
544443cb 1936int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1a1e5d4f 1937{
8865859d 1938 struct list_head *pos;
3715a633 1939 struct multi_pack_index *m;
1a1e5d4f 1940
0a0dd632 1941 prepare_packed_git(r);
3715a633 1942 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1a1e5d4f
JT
1943 return 0;
1944
3715a633
DS
1945 for (m = r->objects->multi_pack_index; m; m = m->next) {
1946 if (fill_midx_entry(oid, e, m))
1947 return 1;
1948 }
1949
0a0dd632 1950 list_for_each(pos, &r->objects->packed_git_mru) {
ec2dd32c 1951 struct packed_git *p = list_entry(pos, struct packed_git, mru);
544443cb 1952 if (fill_pack_entry(oid, e, p)) {
0a0dd632 1953 list_move(&p->mru, &r->objects->packed_git_mru);
1a1e5d4f
JT
1954 return 1;
1955 }
1956 }
1957 return 0;
1958}
150e3001 1959
14c3c80c 1960int has_object_pack(const struct object_id *oid)
150e3001
JT
1961{
1962 struct pack_entry e;
544443cb 1963 return find_pack_entry(the_repository, oid, &e);
150e3001 1964}
f9a8672a
JT
1965
1966int has_pack_index(const unsigned char *sha1)
1967{
1968 struct stat st;
1969 if (stat(sha1_pack_index_name(sha1), &st))
1970 return 0;
1971 return 1;
1972}
7709f468 1973
736eb88f
JK
1974int for_each_object_in_pack(struct packed_git *p,
1975 each_packed_object_fn cb, void *data,
1976 enum for_each_object_flags flags)
7709f468
JT
1977{
1978 uint32_t i;
1979 int r = 0;
1980
736eb88f
JK
1981 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
1982 load_pack_revindex(p);
1983
7709f468 1984 for (i = 0; i < p->num_objects; i++) {
736eb88f 1985 uint32_t pos;
7709f468
JT
1986 struct object_id oid;
1987
736eb88f
JK
1988 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
1989 pos = p->revindex[i].nr;
1990 else
1991 pos = i;
1992
1993 if (!nth_packed_object_oid(&oid, p, pos))
7709f468 1994 return error("unable to get sha1 of object %u in %s",
736eb88f 1995 pos, p->pack_name);
7709f468 1996
736eb88f 1997 r = cb(&oid, p, pos, data);
7709f468
JT
1998 if (r)
1999 break;
2000 }
2001 return r;
2002}
2003
a7ff6f5a
JK
2004int for_each_packed_object(each_packed_object_fn cb, void *data,
2005 enum for_each_object_flags flags)
7709f468
JT
2006{
2007 struct packed_git *p;
2008 int r = 0;
2009 int pack_errors = 0;
2010
6fdb4e9f 2011 prepare_packed_git(the_repository);
a80d72db 2012 for (p = the_repository->objects->packed_git; p; p = p->next) {
7709f468
JT
2013 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2014 continue;
498f1f61
JT
2015 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2016 !p->pack_promisor)
2017 continue;
7709f468
JT
2018 if (open_pack_index(p)) {
2019 pack_errors = 1;
2020 continue;
2021 }
736eb88f 2022 r = for_each_object_in_pack(p, cb, data, flags);
7709f468
JT
2023 if (r)
2024 break;
2025 }
2026 return r ? r : pack_errors;
2027}
498f1f61
JT
2028
2029static int add_promisor_object(const struct object_id *oid,
2030 struct packed_git *pack,
2031 uint32_t pos,
2032 void *set_)
2033{
2034 struct oidset *set = set_;
109cd76d 2035 struct object *obj = parse_object(the_repository, oid);
498f1f61
JT
2036 if (!obj)
2037 return 1;
2038
2039 oidset_insert(set, oid);
2040
2041 /*
2042 * If this is a tree, commit, or tag, the objects it refers
a80d72db 2043 * to are also promisor objects. (Blobs refer to no objects->)
498f1f61
JT
2044 */
2045 if (obj->type == OBJ_TREE) {
2046 struct tree *tree = (struct tree *)obj;
2047 struct tree_desc desc;
2048 struct name_entry entry;
2049 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2050 /*
2051 * Error messages are given when packs are
2052 * verified, so do not print any here.
2053 */
2054 return 0;
2055 while (tree_entry_gently(&desc, &entry))
2056 oidset_insert(set, entry.oid);
2057 } else if (obj->type == OBJ_COMMIT) {
2058 struct commit *commit = (struct commit *) obj;
2059 struct commit_list *parents = commit->parents;
2060
2e27bd77 2061 oidset_insert(set, get_commit_tree_oid(commit));
498f1f61
JT
2062 for (; parents; parents = parents->next)
2063 oidset_insert(set, &parents->item->object.oid);
2064 } else if (obj->type == OBJ_TAG) {
2065 struct tag *tag = (struct tag *) obj;
2066 oidset_insert(set, &tag->tagged->oid);
2067 }
2068 return 0;
2069}
2070
2071int is_promisor_object(const struct object_id *oid)
2072{
2073 static struct oidset promisor_objects;
2074 static int promisor_objects_prepared;
2075
2076 if (!promisor_objects_prepared) {
2077 if (repository_format_partial_clone) {
2078 for_each_packed_object(add_promisor_object,
2079 &promisor_objects,
2080 FOR_EACH_OBJECT_PROMISOR_ONLY);
2081 }
2082 promisor_objects_prepared = 1;
2083 }
2084 return oidset_contains(&promisor_objects, oid);
2085}