Merge branch 'pw/add-p-recount' into maint
[git/git.git] / packfile.c
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 "list.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
12 #include "commit.h"
13 #include "object.h"
14 #include "tag.h"
15 #include "tree-walk.h"
16 #include "tree.h"
17 #include "object-store.h"
18 #include "midx.h"
19
20 char *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
30 char *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
36 char *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 }
41
42 static unsigned int pack_used_ctr;
43 static unsigned int pack_mmap_calls;
44 static unsigned int peak_pack_open_windows;
45 static unsigned int pack_open_windows;
46 static unsigned int pack_open_fds;
47 static unsigned int pack_max_fds;
48 static size_t peak_pack_mapped;
49 static size_t pack_mapped;
50
51 #define SZ_FMT PRIuMAX
52 static inline uintmax_t sz_fmt(size_t s) { return s; }
53
54 void 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 }
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 */
80 static int check_packed_git_idx(const char *path, struct packed_git *p)
81 {
82 void *idx_map;
83 size_t idx_size;
84 int fd = git_open(path), ret;
85 struct stat st;
86 const unsigned int hashsz = the_hash_algo->rawsz;
87
88 if (fd < 0)
89 return -1;
90 if (fstat(fd, &st)) {
91 close(fd);
92 return -1;
93 }
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + hashsz + hashsz) {
96 close(fd);
97 return error("index file %s is too small", path);
98 }
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
101
102 ret = load_idx(path, hashsz, idx_map, idx_size, p);
103
104 if (ret)
105 munmap(idx_map, idx_size);
106
107 return ret;
108 }
109
110 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
111 size_t idx_size, struct packed_git *p)
112 {
113 struct pack_idx_header *hdr = idx_map;
114 uint32_t version, nr, i, *index;
115
116 if (idx_size < 4 * 256 + hashsz + hashsz)
117 return error("index file %s is too small", path);
118 if (idx_map == NULL)
119 return error("empty data");
120
121 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
122 version = ntohl(hdr->idx_version);
123 if (version < 2 || version > 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
127 path, version);
128 } else
129 version = 1;
130
131 nr = 0;
132 index = idx_map;
133 if (version > 1)
134 index += 2; /* skip index header */
135 for (i = 0; i < 256; i++) {
136 uint32_t n = ntohl(index[i]);
137 if (n < nr)
138 return error("non-monotonic index %s", path);
139 nr = n;
140 }
141
142 if (version == 1) {
143 /*
144 * Total size:
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
148 * - file checksum
149 */
150 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
151 return error("wrong index v1 file size in %s", path);
152 } else if (version == 2) {
153 /*
154 * Minimum size:
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
161 * - file checksum
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
165 */
166 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
167 unsigned long max_size = min_size;
168 if (nr)
169 max_size += (nr - 1)*8;
170 if (idx_size < min_size || idx_size > max_size)
171 return error("wrong index v2 file size in %s", path);
172 if (idx_size != min_size &&
173 /*
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
177 */
178 (sizeof(off_t) <= 4))
179 return error("pack too large for current definition of off_t in %s", path);
180 }
181
182 p->index_version = version;
183 p->index_data = idx_map;
184 p->index_size = idx_size;
185 p->num_objects = nr;
186 return 0;
187 }
188
189 int open_pack_index(struct packed_git *p)
190 {
191 char *idx_name;
192 size_t len;
193 int ret;
194
195 if (p->index_data)
196 return 0;
197
198 if (!strip_suffix(p->pack_name, ".pack", &len))
199 BUG("pack_name does not end in .pack");
200 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
201 ret = check_packed_git_idx(idx_name, p);
202 free(idx_name);
203 return ret;
204 }
205
206 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
207 {
208 const uint32_t *level1_ofs = p->index_data;
209
210 if (!level1_ofs) {
211 if (open_pack_index(p))
212 return 0;
213 level1_ofs = p->index_data;
214 }
215
216 if (p->index_version > 1) {
217 level1_ofs += 2;
218 }
219
220 return ntohl(level1_ofs[value]);
221 }
222
223 static struct packed_git *alloc_packed_git(int extra)
224 {
225 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
226 memset(p, 0, sizeof(*p));
227 p->pack_fd = -1;
228 return p;
229 }
230
231 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
232 {
233 const char *path = sha1_pack_name(sha1);
234 size_t alloc = st_add(strlen(path), 1);
235 struct packed_git *p = alloc_packed_git(alloc);
236
237 memcpy(p->pack_name, path, alloc); /* includes NUL */
238 hashcpy(p->hash, sha1);
239 if (check_packed_git_idx(idx_path, p)) {
240 free(p);
241 return NULL;
242 }
243
244 return p;
245 }
246
247 static void scan_windows(struct packed_git *p,
248 struct packed_git **lru_p,
249 struct pack_window **lru_w,
250 struct pack_window **lru_l)
251 {
252 struct pack_window *w, *w_l;
253
254 for (w_l = NULL, w = p->windows; w; w = w->next) {
255 if (!w->inuse_cnt) {
256 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
257 *lru_p = p;
258 *lru_w = w;
259 *lru_l = w_l;
260 }
261 }
262 w_l = w;
263 }
264 }
265
266 static int unuse_one_window(struct packed_git *current)
267 {
268 struct packed_git *p, *lru_p = NULL;
269 struct pack_window *lru_w = NULL, *lru_l = NULL;
270
271 if (current)
272 scan_windows(current, &lru_p, &lru_w, &lru_l);
273 for (p = the_repository->objects->packed_git; p; p = p->next)
274 scan_windows(p, &lru_p, &lru_w, &lru_l);
275 if (lru_p) {
276 munmap(lru_w->base, lru_w->len);
277 pack_mapped -= lru_w->len;
278 if (lru_l)
279 lru_l->next = lru_w->next;
280 else
281 lru_p->windows = lru_w->next;
282 free(lru_w);
283 pack_open_windows--;
284 return 1;
285 }
286 return 0;
287 }
288
289 void release_pack_memory(size_t need)
290 {
291 size_t cur = pack_mapped;
292 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
293 ; /* nothing */
294 }
295
296 void close_pack_windows(struct packed_git *p)
297 {
298 while (p->windows) {
299 struct pack_window *w = p->windows;
300
301 if (w->inuse_cnt)
302 die("pack '%s' still has open windows to it",
303 p->pack_name);
304 munmap(w->base, w->len);
305 pack_mapped -= w->len;
306 pack_open_windows--;
307 p->windows = w->next;
308 free(w);
309 }
310 }
311
312 int close_pack_fd(struct packed_git *p)
313 {
314 if (p->pack_fd < 0)
315 return 0;
316
317 close(p->pack_fd);
318 pack_open_fds--;
319 p->pack_fd = -1;
320
321 return 1;
322 }
323
324 void close_pack_index(struct packed_git *p)
325 {
326 if (p->index_data) {
327 munmap((void *)p->index_data, p->index_size);
328 p->index_data = NULL;
329 }
330 }
331
332 void close_pack(struct packed_git *p)
333 {
334 close_pack_windows(p);
335 close_pack_fd(p);
336 close_pack_index(p);
337 }
338
339 void close_all_packs(struct raw_object_store *o)
340 {
341 struct packed_git *p;
342
343 for (p = o->packed_git; p; p = p->next)
344 if (p->do_not_close)
345 BUG("want to close pack marked 'do-not-close'");
346 else
347 close_pack(p);
348
349 if (o->multi_pack_index) {
350 close_midx(o->multi_pack_index);
351 o->multi_pack_index = NULL;
352 }
353 }
354
355 /*
356 * The LRU pack is the one with the oldest MRU window, preferring packs
357 * with no used windows, or the oldest mtime if it has no windows allocated.
358 */
359 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
360 {
361 struct pack_window *w, *this_mru_w;
362 int has_windows_inuse = 0;
363
364 /*
365 * Reject this pack if it has windows and the previously selected
366 * one does not. If this pack does not have windows, reject
367 * it if the pack file is newer than the previously selected one.
368 */
369 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
370 return;
371
372 for (w = this_mru_w = p->windows; w; w = w->next) {
373 /*
374 * Reject this pack if any of its windows are in use,
375 * but the previously selected pack did not have any
376 * inuse windows. Otherwise, record that this pack
377 * has windows in use.
378 */
379 if (w->inuse_cnt) {
380 if (*accept_windows_inuse)
381 has_windows_inuse = 1;
382 else
383 return;
384 }
385
386 if (w->last_used > this_mru_w->last_used)
387 this_mru_w = w;
388
389 /*
390 * Reject this pack if it has windows that have been
391 * used more recently than the previously selected pack.
392 * If the previously selected pack had windows inuse and
393 * we have not encountered a window in this pack that is
394 * inuse, skip this check since we prefer a pack with no
395 * inuse windows to one that has inuse windows.
396 */
397 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
398 this_mru_w->last_used > (*mru_w)->last_used)
399 return;
400 }
401
402 /*
403 * Select this pack.
404 */
405 *mru_w = this_mru_w;
406 *lru_p = p;
407 *accept_windows_inuse = has_windows_inuse;
408 }
409
410 static int close_one_pack(void)
411 {
412 struct packed_git *p, *lru_p = NULL;
413 struct pack_window *mru_w = NULL;
414 int accept_windows_inuse = 1;
415
416 for (p = the_repository->objects->packed_git; p; p = p->next) {
417 if (p->pack_fd == -1)
418 continue;
419 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
420 }
421
422 if (lru_p)
423 return close_pack_fd(lru_p);
424
425 return 0;
426 }
427
428 static unsigned int get_max_fd_limit(void)
429 {
430 #ifdef RLIMIT_NOFILE
431 {
432 struct rlimit lim;
433
434 if (!getrlimit(RLIMIT_NOFILE, &lim))
435 return lim.rlim_cur;
436 }
437 #endif
438
439 #ifdef _SC_OPEN_MAX
440 {
441 long open_max = sysconf(_SC_OPEN_MAX);
442 if (0 < open_max)
443 return open_max;
444 /*
445 * Otherwise, we got -1 for one of the two
446 * reasons:
447 *
448 * (1) sysconf() did not understand _SC_OPEN_MAX
449 * and signaled an error with -1; or
450 * (2) sysconf() said there is no limit.
451 *
452 * We _could_ clear errno before calling sysconf() to
453 * tell these two cases apart and return a huge number
454 * in the latter case to let the caller cap it to a
455 * value that is not so selfish, but letting the
456 * fallback OPEN_MAX codepath take care of these cases
457 * is a lot simpler.
458 */
459 }
460 #endif
461
462 #ifdef OPEN_MAX
463 return OPEN_MAX;
464 #else
465 return 1; /* see the caller ;-) */
466 #endif
467 }
468
469 const char *pack_basename(struct packed_git *p)
470 {
471 const char *ret = strrchr(p->pack_name, '/');
472 if (ret)
473 ret = ret + 1; /* skip past slash */
474 else
475 ret = p->pack_name; /* we only have a base */
476 return ret;
477 }
478
479 /*
480 * Do not call this directly as this leaks p->pack_fd on error return;
481 * call open_packed_git() instead.
482 */
483 static int open_packed_git_1(struct packed_git *p)
484 {
485 struct stat st;
486 struct pack_header hdr;
487 unsigned char hash[GIT_MAX_RAWSZ];
488 unsigned char *idx_hash;
489 long fd_flag;
490 ssize_t read_result;
491 const unsigned hashsz = the_hash_algo->rawsz;
492
493 if (!p->index_data) {
494 struct multi_pack_index *m;
495 const char *pack_name = pack_basename(p);
496
497 for (m = the_repository->objects->multi_pack_index;
498 m; m = m->next) {
499 if (midx_contains_pack(m, pack_name))
500 break;
501 }
502
503 if (!m && open_pack_index(p))
504 return error("packfile %s index unavailable", p->pack_name);
505 }
506
507 if (!pack_max_fds) {
508 unsigned int max_fds = get_max_fd_limit();
509
510 /* Save 3 for stdin/stdout/stderr, 22 for work */
511 if (25 < max_fds)
512 pack_max_fds = max_fds - 25;
513 else
514 pack_max_fds = 1;
515 }
516
517 while (pack_max_fds <= pack_open_fds && close_one_pack())
518 ; /* nothing */
519
520 p->pack_fd = git_open(p->pack_name);
521 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
522 return -1;
523 pack_open_fds++;
524
525 /* If we created the struct before we had the pack we lack size. */
526 if (!p->pack_size) {
527 if (!S_ISREG(st.st_mode))
528 return error("packfile %s not a regular file", p->pack_name);
529 p->pack_size = st.st_size;
530 } else if (p->pack_size != st.st_size)
531 return error("packfile %s size changed", p->pack_name);
532
533 /* We leave these file descriptors open with sliding mmap;
534 * there is no point keeping them open across exec(), though.
535 */
536 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
537 if (fd_flag < 0)
538 return error("cannot determine file descriptor flags");
539 fd_flag |= FD_CLOEXEC;
540 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
541 return error("cannot set FD_CLOEXEC");
542
543 /* Verify we recognize this pack file format. */
544 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
545 if (read_result < 0)
546 return error_errno("error reading from %s", p->pack_name);
547 if (read_result != sizeof(hdr))
548 return error("file %s is far too short to be a packfile", p->pack_name);
549 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
550 return error("file %s is not a GIT packfile", p->pack_name);
551 if (!pack_version_ok(hdr.hdr_version))
552 return error("packfile %s is version %"PRIu32" and not"
553 " supported (try upgrading GIT to a newer version)",
554 p->pack_name, ntohl(hdr.hdr_version));
555
556 /* Skip index checking if in multi-pack-index */
557 if (!p->index_data)
558 return 0;
559
560 /* Verify the pack matches its index. */
561 if (p->num_objects != ntohl(hdr.hdr_entries))
562 return error("packfile %s claims to have %"PRIu32" objects"
563 " while index indicates %"PRIu32" objects",
564 p->pack_name, ntohl(hdr.hdr_entries),
565 p->num_objects);
566 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
567 return error("end of packfile %s is unavailable", p->pack_name);
568 read_result = read_in_full(p->pack_fd, hash, hashsz);
569 if (read_result < 0)
570 return error_errno("error reading from %s", p->pack_name);
571 if (read_result != hashsz)
572 return error("packfile %s signature is unavailable", p->pack_name);
573 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
574 if (!hasheq(hash, idx_hash))
575 return error("packfile %s does not match index", p->pack_name);
576 return 0;
577 }
578
579 static int open_packed_git(struct packed_git *p)
580 {
581 if (!open_packed_git_1(p))
582 return 0;
583 close_pack_fd(p);
584 return -1;
585 }
586
587 static int in_window(struct pack_window *win, off_t offset)
588 {
589 /* We must promise at least one full hash after the
590 * offset is available from this window, otherwise the offset
591 * is not actually in this window and a different window (which
592 * has that one hash excess) must be used. This is to support
593 * the object header and delta base parsing routines below.
594 */
595 off_t win_off = win->offset;
596 return win_off <= offset
597 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
598 }
599
600 unsigned char *use_pack(struct packed_git *p,
601 struct pack_window **w_cursor,
602 off_t offset,
603 unsigned long *left)
604 {
605 struct pack_window *win = *w_cursor;
606
607 /* Since packfiles end in a hash of their content and it's
608 * pointless to ask for an offset into the middle of that
609 * hash, and the in_window function above wouldn't match
610 * don't allow an offset too close to the end of the file.
611 */
612 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
613 die("packfile %s cannot be accessed", p->pack_name);
614 if (offset > (p->pack_size - the_hash_algo->rawsz))
615 die("offset beyond end of packfile (truncated pack?)");
616 if (offset < 0)
617 die(_("offset before end of packfile (broken .idx?)"));
618
619 if (!win || !in_window(win, offset)) {
620 if (win)
621 win->inuse_cnt--;
622 for (win = p->windows; win; win = win->next) {
623 if (in_window(win, offset))
624 break;
625 }
626 if (!win) {
627 size_t window_align = packed_git_window_size / 2;
628 off_t len;
629
630 if (p->pack_fd == -1 && open_packed_git(p))
631 die("packfile %s cannot be accessed", p->pack_name);
632
633 win = xcalloc(1, sizeof(*win));
634 win->offset = (offset / window_align) * window_align;
635 len = p->pack_size - win->offset;
636 if (len > packed_git_window_size)
637 len = packed_git_window_size;
638 win->len = (size_t)len;
639 pack_mapped += win->len;
640 while (packed_git_limit < pack_mapped
641 && unuse_one_window(p))
642 ; /* nothing */
643 win->base = xmmap_gently(NULL, win->len,
644 PROT_READ, MAP_PRIVATE,
645 p->pack_fd, win->offset);
646 if (win->base == MAP_FAILED)
647 die_errno("packfile %s cannot be mapped",
648 p->pack_name);
649 if (!win->offset && win->len == p->pack_size
650 && !p->do_not_close)
651 close_pack_fd(p);
652 pack_mmap_calls++;
653 pack_open_windows++;
654 if (pack_mapped > peak_pack_mapped)
655 peak_pack_mapped = pack_mapped;
656 if (pack_open_windows > peak_pack_open_windows)
657 peak_pack_open_windows = pack_open_windows;
658 win->next = p->windows;
659 p->windows = win;
660 }
661 }
662 if (win != *w_cursor) {
663 win->last_used = pack_used_ctr++;
664 win->inuse_cnt++;
665 *w_cursor = win;
666 }
667 offset -= win->offset;
668 if (left)
669 *left = win->len - xsize_t(offset);
670 return win->base + offset;
671 }
672
673 void unuse_pack(struct pack_window **w_cursor)
674 {
675 struct pack_window *w = *w_cursor;
676 if (w) {
677 w->inuse_cnt--;
678 *w_cursor = NULL;
679 }
680 }
681
682 static void try_to_free_pack_memory(size_t size)
683 {
684 release_pack_memory(size);
685 }
686
687 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
688 {
689 static int have_set_try_to_free_routine;
690 struct stat st;
691 size_t alloc;
692 struct packed_git *p;
693
694 if (!have_set_try_to_free_routine) {
695 have_set_try_to_free_routine = 1;
696 set_try_to_free_routine(try_to_free_pack_memory);
697 }
698
699 /*
700 * Make sure a corresponding .pack file exists and that
701 * the index looks sane.
702 */
703 if (!strip_suffix_mem(path, &path_len, ".idx"))
704 return NULL;
705
706 /*
707 * ".promisor" is long enough to hold any suffix we're adding (and
708 * the use xsnprintf double-checks that)
709 */
710 alloc = st_add3(path_len, strlen(".promisor"), 1);
711 p = alloc_packed_git(alloc);
712 memcpy(p->pack_name, path, path_len);
713
714 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
715 if (!access(p->pack_name, F_OK))
716 p->pack_keep = 1;
717
718 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
719 if (!access(p->pack_name, F_OK))
720 p->pack_promisor = 1;
721
722 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
723 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
724 free(p);
725 return NULL;
726 }
727
728 /* ok, it looks sane as far as we can check without
729 * actually mapping the pack file.
730 */
731 p->pack_size = st.st_size;
732 p->pack_local = local;
733 p->mtime = st.st_mtime;
734 if (path_len < the_hash_algo->hexsz ||
735 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
736 hashclr(p->hash);
737 return p;
738 }
739
740 void install_packed_git(struct repository *r, struct packed_git *pack)
741 {
742 if (pack->pack_fd != -1)
743 pack_open_fds++;
744
745 pack->next = r->objects->packed_git;
746 r->objects->packed_git = pack;
747 }
748
749 void (*report_garbage)(unsigned seen_bits, const char *path);
750
751 static void report_helper(const struct string_list *list,
752 int seen_bits, int first, int last)
753 {
754 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
755 return;
756
757 for (; first < last; first++)
758 report_garbage(seen_bits, list->items[first].string);
759 }
760
761 static void report_pack_garbage(struct string_list *list)
762 {
763 int i, baselen = -1, first = 0, seen_bits = 0;
764
765 if (!report_garbage)
766 return;
767
768 string_list_sort(list);
769
770 for (i = 0; i < list->nr; i++) {
771 const char *path = list->items[i].string;
772 if (baselen != -1 &&
773 strncmp(path, list->items[first].string, baselen)) {
774 report_helper(list, seen_bits, first, i);
775 baselen = -1;
776 seen_bits = 0;
777 }
778 if (baselen == -1) {
779 const char *dot = strrchr(path, '.');
780 if (!dot) {
781 report_garbage(PACKDIR_FILE_GARBAGE, path);
782 continue;
783 }
784 baselen = dot - path + 1;
785 first = i;
786 }
787 if (!strcmp(path + baselen, "pack"))
788 seen_bits |= 1;
789 else if (!strcmp(path + baselen, "idx"))
790 seen_bits |= 2;
791 }
792 report_helper(list, seen_bits, first, list->nr);
793 }
794
795 void for_each_file_in_pack_dir(const char *objdir,
796 each_file_in_pack_dir_fn fn,
797 void *data)
798 {
799 struct strbuf path = STRBUF_INIT;
800 size_t dirnamelen;
801 DIR *dir;
802 struct dirent *de;
803
804 strbuf_addstr(&path, objdir);
805 strbuf_addstr(&path, "/pack");
806 dir = opendir(path.buf);
807 if (!dir) {
808 if (errno != ENOENT)
809 error_errno("unable to open object pack directory: %s",
810 path.buf);
811 strbuf_release(&path);
812 return;
813 }
814 strbuf_addch(&path, '/');
815 dirnamelen = path.len;
816 while ((de = readdir(dir)) != NULL) {
817 if (is_dot_or_dotdot(de->d_name))
818 continue;
819
820 strbuf_setlen(&path, dirnamelen);
821 strbuf_addstr(&path, de->d_name);
822
823 fn(path.buf, path.len, de->d_name, data);
824 }
825
826 closedir(dir);
827 strbuf_release(&path);
828 }
829
830 struct prepare_pack_data {
831 struct repository *r;
832 struct string_list *garbage;
833 int local;
834 struct multi_pack_index *m;
835 };
836
837 static void prepare_pack(const char *full_name, size_t full_name_len,
838 const char *file_name, void *_data)
839 {
840 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
841 struct packed_git *p;
842 size_t base_len = full_name_len;
843
844 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
845 !(data->m && midx_contains_pack(data->m, file_name))) {
846 /* Don't reopen a pack we already have. */
847 for (p = data->r->objects->packed_git; p; p = p->next) {
848 size_t len;
849 if (strip_suffix(p->pack_name, ".pack", &len) &&
850 len == base_len &&
851 !memcmp(p->pack_name, full_name, len))
852 break;
853 }
854
855 if (!p) {
856 p = add_packed_git(full_name, full_name_len, data->local);
857 if (p)
858 install_packed_git(data->r, p);
859 }
860 }
861
862 if (!report_garbage)
863 return;
864
865 if (!strcmp(file_name, "multi-pack-index"))
866 return;
867 if (ends_with(file_name, ".idx") ||
868 ends_with(file_name, ".pack") ||
869 ends_with(file_name, ".bitmap") ||
870 ends_with(file_name, ".keep") ||
871 ends_with(file_name, ".promisor"))
872 string_list_append(data->garbage, full_name);
873 else
874 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
875 }
876
877 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
878 {
879 struct prepare_pack_data data;
880 struct string_list garbage = STRING_LIST_INIT_DUP;
881
882 data.m = r->objects->multi_pack_index;
883
884 /* look for the multi-pack-index for this object directory */
885 while (data.m && strcmp(data.m->object_dir, objdir))
886 data.m = data.m->next;
887
888 data.r = r;
889 data.garbage = &garbage;
890 data.local = local;
891
892 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
893
894 report_pack_garbage(data.garbage);
895 string_list_clear(data.garbage, 0);
896 }
897
898 static void prepare_packed_git(struct repository *r);
899 /*
900 * Give a fast, rough count of the number of objects in the repository. This
901 * ignores loose objects completely. If you have a lot of them, then either
902 * you should repack because your performance will be awful, or they are
903 * all unreachable objects about to be pruned, in which case they're not really
904 * interesting as a measure of repo size in the first place.
905 */
906 unsigned long repo_approximate_object_count(struct repository *r)
907 {
908 if (!r->objects->approximate_object_count_valid) {
909 unsigned long count;
910 struct multi_pack_index *m;
911 struct packed_git *p;
912
913 prepare_packed_git(r);
914 count = 0;
915 for (m = get_multi_pack_index(r); m; m = m->next)
916 count += m->num_objects;
917 for (p = r->objects->packed_git; p; p = p->next) {
918 if (open_pack_index(p))
919 continue;
920 count += p->num_objects;
921 }
922 r->objects->approximate_object_count = count;
923 }
924 return r->objects->approximate_object_count;
925 }
926
927 static void *get_next_packed_git(const void *p)
928 {
929 return ((const struct packed_git *)p)->next;
930 }
931
932 static void set_next_packed_git(void *p, void *next)
933 {
934 ((struct packed_git *)p)->next = next;
935 }
936
937 static int sort_pack(const void *a_, const void *b_)
938 {
939 const struct packed_git *a = a_;
940 const struct packed_git *b = b_;
941 int st;
942
943 /*
944 * Local packs tend to contain objects specific to our
945 * variant of the project than remote ones. In addition,
946 * remote ones could be on a network mounted filesystem.
947 * Favor local ones for these reasons.
948 */
949 st = a->pack_local - b->pack_local;
950 if (st)
951 return -st;
952
953 /*
954 * Younger packs tend to contain more recent objects,
955 * and more recent objects tend to get accessed more
956 * often.
957 */
958 if (a->mtime < b->mtime)
959 return 1;
960 else if (a->mtime == b->mtime)
961 return 0;
962 return -1;
963 }
964
965 static void rearrange_packed_git(struct repository *r)
966 {
967 r->objects->packed_git = llist_mergesort(
968 r->objects->packed_git, get_next_packed_git,
969 set_next_packed_git, sort_pack);
970 }
971
972 static void prepare_packed_git_mru(struct repository *r)
973 {
974 struct packed_git *p;
975
976 INIT_LIST_HEAD(&r->objects->packed_git_mru);
977
978 for (p = r->objects->packed_git; p; p = p->next)
979 list_add_tail(&p->mru, &r->objects->packed_git_mru);
980 }
981
982 static void prepare_packed_git(struct repository *r)
983 {
984 struct object_directory *odb;
985
986 if (r->objects->packed_git_initialized)
987 return;
988
989 prepare_alt_odb(r);
990 for (odb = r->objects->odb; odb; odb = odb->next) {
991 int local = (odb == r->objects->odb);
992 prepare_multi_pack_index_one(r, odb->path, local);
993 prepare_packed_git_one(r, odb->path, local);
994 }
995 rearrange_packed_git(r);
996
997 prepare_packed_git_mru(r);
998 r->objects->packed_git_initialized = 1;
999 }
1000
1001 void reprepare_packed_git(struct repository *r)
1002 {
1003 struct object_directory *odb;
1004
1005 for (odb = r->objects->odb; odb; odb = odb->next)
1006 odb_clear_loose_cache(odb);
1007
1008 r->objects->approximate_object_count_valid = 0;
1009 r->objects->packed_git_initialized = 0;
1010 prepare_packed_git(r);
1011 }
1012
1013 struct packed_git *get_packed_git(struct repository *r)
1014 {
1015 prepare_packed_git(r);
1016 return r->objects->packed_git;
1017 }
1018
1019 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1020 {
1021 prepare_packed_git(r);
1022 return r->objects->multi_pack_index;
1023 }
1024
1025 struct packed_git *get_all_packs(struct repository *r)
1026 {
1027 struct multi_pack_index *m;
1028
1029 prepare_packed_git(r);
1030 for (m = r->objects->multi_pack_index; m; m = m->next) {
1031 uint32_t i;
1032 for (i = 0; i < m->num_packs; i++)
1033 prepare_midx_pack(r, m, i);
1034 }
1035
1036 return r->objects->packed_git;
1037 }
1038
1039 struct list_head *get_packed_git_mru(struct repository *r)
1040 {
1041 prepare_packed_git(r);
1042 return &r->objects->packed_git_mru;
1043 }
1044
1045 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1046 unsigned long len, enum object_type *type, unsigned long *sizep)
1047 {
1048 unsigned shift;
1049 unsigned long size, c;
1050 unsigned long used = 0;
1051
1052 c = buf[used++];
1053 *type = (c >> 4) & 7;
1054 size = c & 15;
1055 shift = 4;
1056 while (c & 0x80) {
1057 if (len <= used || bitsizeof(long) <= shift) {
1058 error("bad object header");
1059 size = used = 0;
1060 break;
1061 }
1062 c = buf[used++];
1063 size += (c & 0x7f) << shift;
1064 shift += 7;
1065 }
1066 *sizep = size;
1067 return used;
1068 }
1069
1070 unsigned long get_size_from_delta(struct packed_git *p,
1071 struct pack_window **w_curs,
1072 off_t curpos)
1073 {
1074 const unsigned char *data;
1075 unsigned char delta_head[20], *in;
1076 git_zstream stream;
1077 int st;
1078
1079 memset(&stream, 0, sizeof(stream));
1080 stream.next_out = delta_head;
1081 stream.avail_out = sizeof(delta_head);
1082
1083 git_inflate_init(&stream);
1084 do {
1085 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1086 stream.next_in = in;
1087 st = git_inflate(&stream, Z_FINISH);
1088 curpos += stream.next_in - in;
1089 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1090 stream.total_out < sizeof(delta_head));
1091 git_inflate_end(&stream);
1092 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1093 error("delta data unpack-initial failed");
1094 return 0;
1095 }
1096
1097 /* Examine the initial part of the delta to figure out
1098 * the result size.
1099 */
1100 data = delta_head;
1101
1102 /* ignore base size */
1103 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1104
1105 /* Read the result size */
1106 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1107 }
1108
1109 int unpack_object_header(struct packed_git *p,
1110 struct pack_window **w_curs,
1111 off_t *curpos,
1112 unsigned long *sizep)
1113 {
1114 unsigned char *base;
1115 unsigned long left;
1116 unsigned long used;
1117 enum object_type type;
1118
1119 /* use_pack() assures us we have [base, base + 20) available
1120 * as a range that we can look at. (Its actually the hash
1121 * size that is assured.) With our object header encoding
1122 * the maximum deflated object size is 2^137, which is just
1123 * insane, so we know won't exceed what we have been given.
1124 */
1125 base = use_pack(p, w_curs, *curpos, &left);
1126 used = unpack_object_header_buffer(base, left, &type, sizep);
1127 if (!used) {
1128 type = OBJ_BAD;
1129 } else
1130 *curpos += used;
1131
1132 return type;
1133 }
1134
1135 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1136 {
1137 unsigned i;
1138 const unsigned hashsz = the_hash_algo->rawsz;
1139 for (i = 0; i < p->num_bad_objects; i++)
1140 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1141 return;
1142 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1143 st_mult(GIT_MAX_RAWSZ,
1144 st_add(p->num_bad_objects, 1)));
1145 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1146 p->num_bad_objects++;
1147 }
1148
1149 const struct packed_git *has_packed_and_bad(struct repository *r,
1150 const unsigned char *sha1)
1151 {
1152 struct packed_git *p;
1153 unsigned i;
1154
1155 for (p = r->objects->packed_git; p; p = p->next)
1156 for (i = 0; i < p->num_bad_objects; i++)
1157 if (hasheq(sha1,
1158 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1159 return p;
1160 return NULL;
1161 }
1162
1163 static off_t get_delta_base(struct packed_git *p,
1164 struct pack_window **w_curs,
1165 off_t *curpos,
1166 enum object_type type,
1167 off_t delta_obj_offset)
1168 {
1169 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1170 off_t base_offset;
1171
1172 /* use_pack() assured us we have [base_info, base_info + 20)
1173 * as a range that we can look at without walking off the
1174 * end of the mapped window. Its actually the hash size
1175 * that is assured. An OFS_DELTA longer than the hash size
1176 * is stupid, as then a REF_DELTA would be smaller to store.
1177 */
1178 if (type == OBJ_OFS_DELTA) {
1179 unsigned used = 0;
1180 unsigned char c = base_info[used++];
1181 base_offset = c & 127;
1182 while (c & 128) {
1183 base_offset += 1;
1184 if (!base_offset || MSB(base_offset, 7))
1185 return 0; /* overflow */
1186 c = base_info[used++];
1187 base_offset = (base_offset << 7) + (c & 127);
1188 }
1189 base_offset = delta_obj_offset - base_offset;
1190 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1191 return 0; /* out of bound */
1192 *curpos += used;
1193 } else if (type == OBJ_REF_DELTA) {
1194 /* The base entry _must_ be in the same pack */
1195 base_offset = find_pack_entry_one(base_info, p);
1196 *curpos += the_hash_algo->rawsz;
1197 } else
1198 die("I am totally screwed");
1199 return base_offset;
1200 }
1201
1202 /*
1203 * Like get_delta_base above, but we return the sha1 instead of the pack
1204 * offset. This means it is cheaper for REF deltas (we do not have to do
1205 * the final object lookup), but more expensive for OFS deltas (we
1206 * have to load the revidx to convert the offset back into a sha1).
1207 */
1208 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1209 struct pack_window **w_curs,
1210 off_t curpos,
1211 enum object_type type,
1212 off_t delta_obj_offset)
1213 {
1214 if (type == OBJ_REF_DELTA) {
1215 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1216 return base;
1217 } else if (type == OBJ_OFS_DELTA) {
1218 struct revindex_entry *revidx;
1219 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1220 type, delta_obj_offset);
1221
1222 if (!base_offset)
1223 return NULL;
1224
1225 revidx = find_pack_revindex(p, base_offset);
1226 if (!revidx)
1227 return NULL;
1228
1229 return nth_packed_object_sha1(p, revidx->nr);
1230 } else
1231 return NULL;
1232 }
1233
1234 static int retry_bad_packed_offset(struct repository *r,
1235 struct packed_git *p,
1236 off_t obj_offset)
1237 {
1238 int type;
1239 struct revindex_entry *revidx;
1240 struct object_id oid;
1241 revidx = find_pack_revindex(p, obj_offset);
1242 if (!revidx)
1243 return OBJ_BAD;
1244 nth_packed_object_oid(&oid, p, revidx->nr);
1245 mark_bad_packed_object(p, oid.hash);
1246 type = oid_object_info(r, &oid, NULL);
1247 if (type <= OBJ_NONE)
1248 return OBJ_BAD;
1249 return type;
1250 }
1251
1252 #define POI_STACK_PREALLOC 64
1253
1254 static enum object_type packed_to_object_type(struct repository *r,
1255 struct packed_git *p,
1256 off_t obj_offset,
1257 enum object_type type,
1258 struct pack_window **w_curs,
1259 off_t curpos)
1260 {
1261 off_t small_poi_stack[POI_STACK_PREALLOC];
1262 off_t *poi_stack = small_poi_stack;
1263 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1264
1265 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1266 off_t base_offset;
1267 unsigned long size;
1268 /* Push the object we're going to leave behind */
1269 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1270 poi_stack_alloc = alloc_nr(poi_stack_nr);
1271 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1272 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1273 } else {
1274 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1275 }
1276 poi_stack[poi_stack_nr++] = obj_offset;
1277 /* If parsing the base offset fails, just unwind */
1278 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1279 if (!base_offset)
1280 goto unwind;
1281 curpos = obj_offset = base_offset;
1282 type = unpack_object_header(p, w_curs, &curpos, &size);
1283 if (type <= OBJ_NONE) {
1284 /* If getting the base itself fails, we first
1285 * retry the base, otherwise unwind */
1286 type = retry_bad_packed_offset(r, p, base_offset);
1287 if (type > OBJ_NONE)
1288 goto out;
1289 goto unwind;
1290 }
1291 }
1292
1293 switch (type) {
1294 case OBJ_BAD:
1295 case OBJ_COMMIT:
1296 case OBJ_TREE:
1297 case OBJ_BLOB:
1298 case OBJ_TAG:
1299 break;
1300 default:
1301 error("unknown object type %i at offset %"PRIuMAX" in %s",
1302 type, (uintmax_t)obj_offset, p->pack_name);
1303 type = OBJ_BAD;
1304 }
1305
1306 out:
1307 if (poi_stack != small_poi_stack)
1308 free(poi_stack);
1309 return type;
1310
1311 unwind:
1312 while (poi_stack_nr) {
1313 obj_offset = poi_stack[--poi_stack_nr];
1314 type = retry_bad_packed_offset(r, p, obj_offset);
1315 if (type > OBJ_NONE)
1316 goto out;
1317 }
1318 type = OBJ_BAD;
1319 goto out;
1320 }
1321
1322 static struct hashmap delta_base_cache;
1323 static size_t delta_base_cached;
1324
1325 static LIST_HEAD(delta_base_cache_lru);
1326
1327 struct delta_base_cache_key {
1328 struct packed_git *p;
1329 off_t base_offset;
1330 };
1331
1332 struct delta_base_cache_entry {
1333 struct hashmap hash;
1334 struct delta_base_cache_key key;
1335 struct list_head lru;
1336 void *data;
1337 unsigned long size;
1338 enum object_type type;
1339 };
1340
1341 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1342 {
1343 unsigned int hash;
1344
1345 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1346 hash += (hash >> 8) + (hash >> 16);
1347 return hash;
1348 }
1349
1350 static struct delta_base_cache_entry *
1351 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1352 {
1353 struct hashmap_entry entry;
1354 struct delta_base_cache_key key;
1355
1356 if (!delta_base_cache.cmpfn)
1357 return NULL;
1358
1359 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1360 key.p = p;
1361 key.base_offset = base_offset;
1362 return hashmap_get(&delta_base_cache, &entry, &key);
1363 }
1364
1365 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1366 const struct delta_base_cache_key *b)
1367 {
1368 return a->p == b->p && a->base_offset == b->base_offset;
1369 }
1370
1371 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1372 const void *va, const void *vb,
1373 const void *vkey)
1374 {
1375 const struct delta_base_cache_entry *a = va, *b = vb;
1376 const struct delta_base_cache_key *key = vkey;
1377 if (key)
1378 return !delta_base_cache_key_eq(&a->key, key);
1379 else
1380 return !delta_base_cache_key_eq(&a->key, &b->key);
1381 }
1382
1383 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1384 {
1385 return !!get_delta_base_cache_entry(p, base_offset);
1386 }
1387
1388 /*
1389 * Remove the entry from the cache, but do _not_ free the associated
1390 * entry data. The caller takes ownership of the "data" buffer, and
1391 * should copy out any fields it wants before detaching.
1392 */
1393 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1394 {
1395 hashmap_remove(&delta_base_cache, ent, &ent->key);
1396 list_del(&ent->lru);
1397 delta_base_cached -= ent->size;
1398 free(ent);
1399 }
1400
1401 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1402 off_t base_offset, unsigned long *base_size,
1403 enum object_type *type)
1404 {
1405 struct delta_base_cache_entry *ent;
1406
1407 ent = get_delta_base_cache_entry(p, base_offset);
1408 if (!ent)
1409 return unpack_entry(r, p, base_offset, type, base_size);
1410
1411 if (type)
1412 *type = ent->type;
1413 if (base_size)
1414 *base_size = ent->size;
1415 return xmemdupz(ent->data, ent->size);
1416 }
1417
1418 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1419 {
1420 free(ent->data);
1421 detach_delta_base_cache_entry(ent);
1422 }
1423
1424 void clear_delta_base_cache(void)
1425 {
1426 struct list_head *lru, *tmp;
1427 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1428 struct delta_base_cache_entry *entry =
1429 list_entry(lru, struct delta_base_cache_entry, lru);
1430 release_delta_base_cache(entry);
1431 }
1432 }
1433
1434 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1435 void *base, unsigned long base_size, enum object_type type)
1436 {
1437 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1438 struct list_head *lru, *tmp;
1439
1440 delta_base_cached += base_size;
1441
1442 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1443 struct delta_base_cache_entry *f =
1444 list_entry(lru, struct delta_base_cache_entry, lru);
1445 if (delta_base_cached <= delta_base_cache_limit)
1446 break;
1447 release_delta_base_cache(f);
1448 }
1449
1450 ent->key.p = p;
1451 ent->key.base_offset = base_offset;
1452 ent->type = type;
1453 ent->data = base;
1454 ent->size = base_size;
1455 list_add_tail(&ent->lru, &delta_base_cache_lru);
1456
1457 if (!delta_base_cache.cmpfn)
1458 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1459 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1460 hashmap_add(&delta_base_cache, ent);
1461 }
1462
1463 int packed_object_info(struct repository *r, struct packed_git *p,
1464 off_t obj_offset, struct object_info *oi)
1465 {
1466 struct pack_window *w_curs = NULL;
1467 unsigned long size;
1468 off_t curpos = obj_offset;
1469 enum object_type type;
1470
1471 /*
1472 * We always get the representation type, but only convert it to
1473 * a "real" type later if the caller is interested.
1474 */
1475 if (oi->contentp) {
1476 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1477 &type);
1478 if (!*oi->contentp)
1479 type = OBJ_BAD;
1480 } else {
1481 type = unpack_object_header(p, &w_curs, &curpos, &size);
1482 }
1483
1484 if (!oi->contentp && oi->sizep) {
1485 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1486 off_t tmp_pos = curpos;
1487 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1488 type, obj_offset);
1489 if (!base_offset) {
1490 type = OBJ_BAD;
1491 goto out;
1492 }
1493 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1494 if (*oi->sizep == 0) {
1495 type = OBJ_BAD;
1496 goto out;
1497 }
1498 } else {
1499 *oi->sizep = size;
1500 }
1501 }
1502
1503 if (oi->disk_sizep) {
1504 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1505 *oi->disk_sizep = revidx[1].offset - obj_offset;
1506 }
1507
1508 if (oi->typep || oi->type_name) {
1509 enum object_type ptot;
1510 ptot = packed_to_object_type(r, p, obj_offset,
1511 type, &w_curs, curpos);
1512 if (oi->typep)
1513 *oi->typep = ptot;
1514 if (oi->type_name) {
1515 const char *tn = type_name(ptot);
1516 if (tn)
1517 strbuf_addstr(oi->type_name, tn);
1518 }
1519 if (ptot < 0) {
1520 type = OBJ_BAD;
1521 goto out;
1522 }
1523 }
1524
1525 if (oi->delta_base_sha1) {
1526 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1527 const unsigned char *base;
1528
1529 base = get_delta_base_sha1(p, &w_curs, curpos,
1530 type, obj_offset);
1531 if (!base) {
1532 type = OBJ_BAD;
1533 goto out;
1534 }
1535
1536 hashcpy(oi->delta_base_sha1, base);
1537 } else
1538 hashclr(oi->delta_base_sha1);
1539 }
1540
1541 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1542 OI_PACKED;
1543
1544 out:
1545 unuse_pack(&w_curs);
1546 return type;
1547 }
1548
1549 static void *unpack_compressed_entry(struct packed_git *p,
1550 struct pack_window **w_curs,
1551 off_t curpos,
1552 unsigned long size)
1553 {
1554 int st;
1555 git_zstream stream;
1556 unsigned char *buffer, *in;
1557
1558 buffer = xmallocz_gently(size);
1559 if (!buffer)
1560 return NULL;
1561 memset(&stream, 0, sizeof(stream));
1562 stream.next_out = buffer;
1563 stream.avail_out = size + 1;
1564
1565 git_inflate_init(&stream);
1566 do {
1567 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1568 stream.next_in = in;
1569 st = git_inflate(&stream, Z_FINISH);
1570 if (!stream.avail_out)
1571 break; /* the payload is larger than it should be */
1572 curpos += stream.next_in - in;
1573 } while (st == Z_OK || st == Z_BUF_ERROR);
1574 git_inflate_end(&stream);
1575 if ((st != Z_STREAM_END) || stream.total_out != size) {
1576 free(buffer);
1577 return NULL;
1578 }
1579
1580 /* versions of zlib can clobber unconsumed portion of outbuf */
1581 buffer[size] = '\0';
1582
1583 return buffer;
1584 }
1585
1586 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1587 {
1588 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1589 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1590 p->pack_name, (uintmax_t)obj_offset);
1591 }
1592
1593 int do_check_packed_object_crc;
1594
1595 #define UNPACK_ENTRY_STACK_PREALLOC 64
1596 struct unpack_entry_stack_ent {
1597 off_t obj_offset;
1598 off_t curpos;
1599 unsigned long size;
1600 };
1601
1602 static void *read_object(struct repository *r,
1603 const struct object_id *oid,
1604 enum object_type *type,
1605 unsigned long *size)
1606 {
1607 struct object_info oi = OBJECT_INFO_INIT;
1608 void *content;
1609 oi.typep = type;
1610 oi.sizep = size;
1611 oi.contentp = &content;
1612
1613 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1614 return NULL;
1615 return content;
1616 }
1617
1618 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1619 enum object_type *final_type, unsigned long *final_size)
1620 {
1621 struct pack_window *w_curs = NULL;
1622 off_t curpos = obj_offset;
1623 void *data = NULL;
1624 unsigned long size;
1625 enum object_type type;
1626 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1627 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1628 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1629 int base_from_cache = 0;
1630
1631 write_pack_access_log(p, obj_offset);
1632
1633 /* PHASE 1: drill down to the innermost base object */
1634 for (;;) {
1635 off_t base_offset;
1636 int i;
1637 struct delta_base_cache_entry *ent;
1638
1639 ent = get_delta_base_cache_entry(p, curpos);
1640 if (ent) {
1641 type = ent->type;
1642 data = ent->data;
1643 size = ent->size;
1644 detach_delta_base_cache_entry(ent);
1645 base_from_cache = 1;
1646 break;
1647 }
1648
1649 if (do_check_packed_object_crc && p->index_version > 1) {
1650 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1651 off_t len = revidx[1].offset - obj_offset;
1652 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1653 struct object_id oid;
1654 nth_packed_object_oid(&oid, p, revidx->nr);
1655 error("bad packed object CRC for %s",
1656 oid_to_hex(&oid));
1657 mark_bad_packed_object(p, oid.hash);
1658 data = NULL;
1659 goto out;
1660 }
1661 }
1662
1663 type = unpack_object_header(p, &w_curs, &curpos, &size);
1664 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1665 break;
1666
1667 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1668 if (!base_offset) {
1669 error("failed to validate delta base reference "
1670 "at offset %"PRIuMAX" from %s",
1671 (uintmax_t)curpos, p->pack_name);
1672 /* bail to phase 2, in hopes of recovery */
1673 data = NULL;
1674 break;
1675 }
1676
1677 /* push object, proceed to base */
1678 if (delta_stack_nr >= delta_stack_alloc
1679 && delta_stack == small_delta_stack) {
1680 delta_stack_alloc = alloc_nr(delta_stack_nr);
1681 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1682 COPY_ARRAY(delta_stack, small_delta_stack,
1683 delta_stack_nr);
1684 } else {
1685 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1686 }
1687 i = delta_stack_nr++;
1688 delta_stack[i].obj_offset = obj_offset;
1689 delta_stack[i].curpos = curpos;
1690 delta_stack[i].size = size;
1691
1692 curpos = obj_offset = base_offset;
1693 }
1694
1695 /* PHASE 2: handle the base */
1696 switch (type) {
1697 case OBJ_OFS_DELTA:
1698 case OBJ_REF_DELTA:
1699 if (data)
1700 BUG("unpack_entry: left loop at a valid delta");
1701 break;
1702 case OBJ_COMMIT:
1703 case OBJ_TREE:
1704 case OBJ_BLOB:
1705 case OBJ_TAG:
1706 if (!base_from_cache)
1707 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1708 break;
1709 default:
1710 data = NULL;
1711 error("unknown object type %i at offset %"PRIuMAX" in %s",
1712 type, (uintmax_t)obj_offset, p->pack_name);
1713 }
1714
1715 /* PHASE 3: apply deltas in order */
1716
1717 /* invariants:
1718 * 'data' holds the base data, or NULL if there was corruption
1719 */
1720 while (delta_stack_nr) {
1721 void *delta_data;
1722 void *base = data;
1723 void *external_base = NULL;
1724 unsigned long delta_size, base_size = size;
1725 int i;
1726
1727 data = NULL;
1728
1729 if (base)
1730 add_delta_base_cache(p, obj_offset, base, base_size, type);
1731
1732 if (!base) {
1733 /*
1734 * We're probably in deep shit, but let's try to fetch
1735 * the required base anyway from another pack or loose.
1736 * This is costly but should happen only in the presence
1737 * of a corrupted pack, and is better than failing outright.
1738 */
1739 struct revindex_entry *revidx;
1740 struct object_id base_oid;
1741 revidx = find_pack_revindex(p, obj_offset);
1742 if (revidx) {
1743 nth_packed_object_oid(&base_oid, p, revidx->nr);
1744 error("failed to read delta base object %s"
1745 " at offset %"PRIuMAX" from %s",
1746 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1747 p->pack_name);
1748 mark_bad_packed_object(p, base_oid.hash);
1749 base = read_object(r, &base_oid, &type, &base_size);
1750 external_base = base;
1751 }
1752 }
1753
1754 i = --delta_stack_nr;
1755 obj_offset = delta_stack[i].obj_offset;
1756 curpos = delta_stack[i].curpos;
1757 delta_size = delta_stack[i].size;
1758
1759 if (!base)
1760 continue;
1761
1762 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1763
1764 if (!delta_data) {
1765 error("failed to unpack compressed delta "
1766 "at offset %"PRIuMAX" from %s",
1767 (uintmax_t)curpos, p->pack_name);
1768 data = NULL;
1769 free(external_base);
1770 continue;
1771 }
1772
1773 data = patch_delta(base, base_size,
1774 delta_data, delta_size,
1775 &size);
1776
1777 /*
1778 * We could not apply the delta; warn the user, but keep going.
1779 * Our failure will be noticed either in the next iteration of
1780 * the loop, or if this is the final delta, in the caller when
1781 * we return NULL. Those code paths will take care of making
1782 * a more explicit warning and retrying with another copy of
1783 * the object.
1784 */
1785 if (!data)
1786 error("failed to apply delta");
1787
1788 free(delta_data);
1789 free(external_base);
1790 }
1791
1792 if (final_type)
1793 *final_type = type;
1794 if (final_size)
1795 *final_size = size;
1796
1797 out:
1798 unuse_pack(&w_curs);
1799
1800 if (delta_stack != small_delta_stack)
1801 free(delta_stack);
1802
1803 return data;
1804 }
1805
1806 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1807 {
1808 const unsigned char *index_fanout = p->index_data;
1809 const unsigned char *index_lookup;
1810 const unsigned int hashsz = the_hash_algo->rawsz;
1811 int index_lookup_width;
1812
1813 if (!index_fanout)
1814 BUG("bsearch_pack called without a valid pack-index");
1815
1816 index_lookup = index_fanout + 4 * 256;
1817 if (p->index_version == 1) {
1818 index_lookup_width = hashsz + 4;
1819 index_lookup += 4;
1820 } else {
1821 index_lookup_width = hashsz;
1822 index_fanout += 8;
1823 index_lookup += 8;
1824 }
1825
1826 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1827 index_lookup, index_lookup_width, result);
1828 }
1829
1830 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1831 uint32_t n)
1832 {
1833 const unsigned char *index = p->index_data;
1834 const unsigned int hashsz = the_hash_algo->rawsz;
1835 if (!index) {
1836 if (open_pack_index(p))
1837 return NULL;
1838 index = p->index_data;
1839 }
1840 if (n >= p->num_objects)
1841 return NULL;
1842 index += 4 * 256;
1843 if (p->index_version == 1) {
1844 return index + (hashsz + 4) * n + 4;
1845 } else {
1846 index += 8;
1847 return index + hashsz * n;
1848 }
1849 }
1850
1851 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1852 struct packed_git *p,
1853 uint32_t n)
1854 {
1855 const unsigned char *hash = nth_packed_object_sha1(p, n);
1856 if (!hash)
1857 return NULL;
1858 hashcpy(oid->hash, hash);
1859 return oid;
1860 }
1861
1862 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1863 {
1864 const unsigned char *ptr = vptr;
1865 const unsigned char *start = p->index_data;
1866 const unsigned char *end = start + p->index_size;
1867 if (ptr < start)
1868 die(_("offset before start of pack index for %s (corrupt index?)"),
1869 p->pack_name);
1870 /* No need to check for underflow; .idx files must be at least 8 bytes */
1871 if (ptr >= end - 8)
1872 die(_("offset beyond end of pack index for %s (truncated index?)"),
1873 p->pack_name);
1874 }
1875
1876 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1877 {
1878 const unsigned char *index = p->index_data;
1879 const unsigned int hashsz = the_hash_algo->rawsz;
1880 index += 4 * 256;
1881 if (p->index_version == 1) {
1882 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1883 } else {
1884 uint32_t off;
1885 index += 8 + p->num_objects * (hashsz + 4);
1886 off = ntohl(*((uint32_t *)(index + 4 * n)));
1887 if (!(off & 0x80000000))
1888 return off;
1889 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1890 check_pack_index_ptr(p, index);
1891 return get_be64(index);
1892 }
1893 }
1894
1895 off_t find_pack_entry_one(const unsigned char *sha1,
1896 struct packed_git *p)
1897 {
1898 const unsigned char *index = p->index_data;
1899 struct object_id oid;
1900 uint32_t result;
1901
1902 if (!index) {
1903 if (open_pack_index(p))
1904 return 0;
1905 }
1906
1907 hashcpy(oid.hash, sha1);
1908 if (bsearch_pack(&oid, p, &result))
1909 return nth_packed_object_offset(p, result);
1910 return 0;
1911 }
1912
1913 int is_pack_valid(struct packed_git *p)
1914 {
1915 /* An already open pack is known to be valid. */
1916 if (p->pack_fd != -1)
1917 return 1;
1918
1919 /* If the pack has one window completely covering the
1920 * file size, the pack is known to be valid even if
1921 * the descriptor is not currently open.
1922 */
1923 if (p->windows) {
1924 struct pack_window *w = p->windows;
1925
1926 if (!w->offset && w->len == p->pack_size)
1927 return 1;
1928 }
1929
1930 /* Force the pack to open to prove its valid. */
1931 return !open_packed_git(p);
1932 }
1933
1934 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1935 struct packed_git *packs)
1936 {
1937 struct packed_git *p;
1938
1939 for (p = packs; p; p = p->next) {
1940 if (find_pack_entry_one(sha1, p))
1941 return p;
1942 }
1943 return NULL;
1944
1945 }
1946
1947 static int fill_pack_entry(const struct object_id *oid,
1948 struct pack_entry *e,
1949 struct packed_git *p)
1950 {
1951 off_t offset;
1952
1953 if (p->num_bad_objects) {
1954 unsigned i;
1955 for (i = 0; i < p->num_bad_objects; i++)
1956 if (hasheq(oid->hash,
1957 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1958 return 0;
1959 }
1960
1961 offset = find_pack_entry_one(oid->hash, p);
1962 if (!offset)
1963 return 0;
1964
1965 /*
1966 * We are about to tell the caller where they can locate the
1967 * requested object. We better make sure the packfile is
1968 * still here and can be accessed before supplying that
1969 * answer, as it may have been deleted since the index was
1970 * loaded!
1971 */
1972 if (!is_pack_valid(p))
1973 return 0;
1974 e->offset = offset;
1975 e->p = p;
1976 return 1;
1977 }
1978
1979 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1980 {
1981 struct list_head *pos;
1982 struct multi_pack_index *m;
1983
1984 prepare_packed_git(r);
1985 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1986 return 0;
1987
1988 for (m = r->objects->multi_pack_index; m; m = m->next) {
1989 if (fill_midx_entry(r, oid, e, m))
1990 return 1;
1991 }
1992
1993 list_for_each(pos, &r->objects->packed_git_mru) {
1994 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1995 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
1996 list_move(&p->mru, &r->objects->packed_git_mru);
1997 return 1;
1998 }
1999 }
2000 return 0;
2001 }
2002
2003 int has_object_pack(const struct object_id *oid)
2004 {
2005 struct pack_entry e;
2006 return find_pack_entry(the_repository, oid, &e);
2007 }
2008
2009 int has_pack_index(const unsigned char *sha1)
2010 {
2011 struct stat st;
2012 if (stat(sha1_pack_index_name(sha1), &st))
2013 return 0;
2014 return 1;
2015 }
2016
2017 int for_each_object_in_pack(struct packed_git *p,
2018 each_packed_object_fn cb, void *data,
2019 enum for_each_object_flags flags)
2020 {
2021 uint32_t i;
2022 int r = 0;
2023
2024 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2025 if (load_pack_revindex(p))
2026 return -1;
2027 }
2028
2029 for (i = 0; i < p->num_objects; i++) {
2030 uint32_t pos;
2031 struct object_id oid;
2032
2033 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2034 pos = p->revindex[i].nr;
2035 else
2036 pos = i;
2037
2038 if (!nth_packed_object_oid(&oid, p, pos))
2039 return error("unable to get sha1 of object %u in %s",
2040 pos, p->pack_name);
2041
2042 r = cb(&oid, p, pos, data);
2043 if (r)
2044 break;
2045 }
2046 return r;
2047 }
2048
2049 int for_each_packed_object(each_packed_object_fn cb, void *data,
2050 enum for_each_object_flags flags)
2051 {
2052 struct packed_git *p;
2053 int r = 0;
2054 int pack_errors = 0;
2055
2056 prepare_packed_git(the_repository);
2057 for (p = get_all_packs(the_repository); p; p = p->next) {
2058 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2059 continue;
2060 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2061 !p->pack_promisor)
2062 continue;
2063 if (open_pack_index(p)) {
2064 pack_errors = 1;
2065 continue;
2066 }
2067 r = for_each_object_in_pack(p, cb, data, flags);
2068 if (r)
2069 break;
2070 }
2071 return r ? r : pack_errors;
2072 }
2073
2074 static int add_promisor_object(const struct object_id *oid,
2075 struct packed_git *pack,
2076 uint32_t pos,
2077 void *set_)
2078 {
2079 struct oidset *set = set_;
2080 struct object *obj = parse_object(the_repository, oid);
2081 if (!obj)
2082 return 1;
2083
2084 oidset_insert(set, oid);
2085
2086 /*
2087 * If this is a tree, commit, or tag, the objects it refers
2088 * to are also promisor objects. (Blobs refer to no objects->)
2089 */
2090 if (obj->type == OBJ_TREE) {
2091 struct tree *tree = (struct tree *)obj;
2092 struct tree_desc desc;
2093 struct name_entry entry;
2094 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2095 /*
2096 * Error messages are given when packs are
2097 * verified, so do not print any here.
2098 */
2099 return 0;
2100 while (tree_entry_gently(&desc, &entry))
2101 oidset_insert(set, &entry.oid);
2102 } else if (obj->type == OBJ_COMMIT) {
2103 struct commit *commit = (struct commit *) obj;
2104 struct commit_list *parents = commit->parents;
2105
2106 oidset_insert(set, get_commit_tree_oid(commit));
2107 for (; parents; parents = parents->next)
2108 oidset_insert(set, &parents->item->object.oid);
2109 } else if (obj->type == OBJ_TAG) {
2110 struct tag *tag = (struct tag *) obj;
2111 oidset_insert(set, &tag->tagged->oid);
2112 }
2113 return 0;
2114 }
2115
2116 int is_promisor_object(const struct object_id *oid)
2117 {
2118 static struct oidset promisor_objects;
2119 static int promisor_objects_prepared;
2120
2121 if (!promisor_objects_prepared) {
2122 if (repository_format_partial_clone) {
2123 for_each_packed_object(add_promisor_object,
2124 &promisor_objects,
2125 FOR_EACH_OBJECT_PROMISOR_ONLY);
2126 }
2127 promisor_objects_prepared = 1;
2128 }
2129 return oidset_contains(&promisor_objects, oid);
2130 }