packfile: add all_packs list
[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 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;
88 const unsigned int hashsz = the_hash_algo->rawsz;
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);
97 if (idx_size < 4 * 256 + hashsz + hashsz) {
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
134 * - 24-byte entries * nr (object ID + 4-byte offset)
135 * - hash of the packfile
136 * - file checksum
137 */
138 if (idx_size != 4*256 + nr * (hashsz + 4) + hashsz + hashsz) {
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
147 * - object ID entry * nr
148 * - 4-byte crc entry * nr
149 * - 4-byte offset entry * nr
150 * - hash of the packfile
151 * - file checksum
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 */
156 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
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
183 int 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))
193 BUG("pack_name does not end in .pack");
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
200 uint32_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
217 static 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
225 struct 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 }
240
241 static 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
260 static int unuse_one_window(struct packed_git *current)
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);
267 for (p = the_repository->objects->packed_git; p; p = p->next)
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
283 void 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 }
289
290 void 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
306 static int close_pack_fd(struct packed_git *p)
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
318 void 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
326 void close_pack(struct packed_git *p)
327 {
328 close_pack_windows(p);
329 close_pack_fd(p);
330 close_pack_index(p);
331 }
332
333 void close_all_packs(struct raw_object_store *o)
334 {
335 struct packed_git *p;
336
337 for (p = o->packed_git; p; p = p->next)
338 if (p->do_not_close)
339 BUG("want to close pack marked 'do-not-close'");
340 else
341 close_pack(p);
342 }
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 */
348 static 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
399 static 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
405 for (p = the_repository->objects->packed_git; p; p = p->next) {
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
417 static 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 */
462 static int open_packed_git_1(struct packed_git *p)
463 {
464 struct stat st;
465 struct pack_header hdr;
466 unsigned char hash[GIT_MAX_RAWSZ];
467 unsigned char *idx_hash;
468 long fd_flag;
469 ssize_t read_result;
470 const unsigned hashsz = the_hash_algo->rawsz;
471
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 }
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. */
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))
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
535 /* Skip index checking if in multi-pack-index */
536 if (!p->index_data)
537 return 0;
538
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);
545 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
546 return error("end of packfile %s is unavailable", p->pack_name);
547 read_result = read_in_full(p->pack_fd, hash, hashsz);
548 if (read_result < 0)
549 return error_errno("error reading from %s", p->pack_name);
550 if (read_result != hashsz)
551 return error("packfile %s signature is unavailable", p->pack_name);
552 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
553 if (hashcmp(hash, idx_hash))
554 return error("packfile %s does not match index", p->pack_name);
555 return 0;
556 }
557
558 static int open_packed_git(struct packed_git *p)
559 {
560 if (!open_packed_git_1(p))
561 return 0;
562 close_pack_fd(p);
563 return -1;
564 }
565
566 static int in_window(struct pack_window *win, off_t offset)
567 {
568 /* We must promise at least one full hash after the
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
576 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
577 }
578
579 unsigned 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);
593 if (offset > (p->pack_size - the_hash_algo->rawsz))
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 }
651
652 void 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 }
660
661 static void try_to_free_pack_memory(size_t size)
662 {
663 release_pack_memory(size);
664 }
665
666 struct 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 /*
686 * ".promisor" is long enough to hold any suffix we're adding (and
687 * the use xsnprintf double-checks that)
688 */
689 alloc = st_add3(path_len, strlen(".promisor"), 1);
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
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
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;
713 if (path_len < the_hash_algo->hexsz ||
714 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->sha1))
715 hashclr(p->sha1);
716 return p;
717 }
718
719 void install_packed_git(struct repository *r, struct packed_git *pack)
720 {
721 if (pack->pack_fd != -1)
722 pack_open_fds++;
723
724 pack->next = r->objects->packed_git;
725 r->objects->packed_git = pack;
726 }
727
728 void (*report_garbage)(unsigned seen_bits, const char *path);
729
730 static 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
740 static 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
774 void for_each_file_in_pack_dir(const char *objdir,
775 each_file_in_pack_dir_fn fn,
776 void *data)
777 {
778 struct strbuf path = STRBUF_INIT;
779 size_t dirnamelen;
780 DIR *dir;
781 struct dirent *de;
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) {
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
802 fn(path.buf, path.len, de->d_name, data);
803 }
804
805 closedir(dir);
806 strbuf_release(&path);
807 }
808
809 struct prepare_pack_data {
810 struct repository *r;
811 struct string_list *garbage;
812 int local;
813 struct multi_pack_index *m;
814 };
815
816 static 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
823 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
824 !(data->m && midx_contains_pack(data->m, file_name))) {
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
844 if (!strcmp(file_name, "multi-pack-index"))
845 return;
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
856 static 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
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
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
877 static void prepare_packed_git(struct repository *r);
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 */
885 unsigned long approximate_object_count(void)
886 {
887 if (!the_repository->objects->approximate_object_count_valid) {
888 unsigned long count;
889 struct multi_pack_index *m;
890 struct packed_git *p;
891
892 prepare_packed_git(the_repository);
893 count = 0;
894 for (m = get_multi_pack_index(the_repository); m; m = m->next)
895 count += m->num_objects;
896 for (p = the_repository->objects->packed_git; p; p = p->next) {
897 if (open_pack_index(p))
898 continue;
899 count += p->num_objects;
900 }
901 the_repository->objects->approximate_object_count = count;
902 }
903 return the_repository->objects->approximate_object_count;
904 }
905
906 static void *get_next_packed_git(const void *p)
907 {
908 return ((const struct packed_git *)p)->next;
909 }
910
911 static void set_next_packed_git(void *p, void *next)
912 {
913 ((struct packed_git *)p)->next = next;
914 }
915
916 static 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
944 static void rearrange_packed_git(struct repository *r)
945 {
946 r->objects->packed_git = llist_mergesort(
947 r->objects->packed_git, get_next_packed_git,
948 set_next_packed_git, sort_pack);
949 }
950
951 static void prepare_packed_git_mru(struct repository *r)
952 {
953 struct packed_git *p;
954
955 INIT_LIST_HEAD(&r->objects->packed_git_mru);
956
957 for (p = r->objects->packed_git; p; p = p->next)
958 list_add_tail(&p->mru, &r->objects->packed_git_mru);
959 }
960
961 static void prepare_packed_git(struct repository *r)
962 {
963 struct alternate_object_database *alt;
964
965 if (r->objects->packed_git_initialized)
966 return;
967 prepare_multi_pack_index_one(r, r->objects->objectdir, 1);
968 prepare_packed_git_one(r, r->objects->objectdir, 1);
969 prepare_alt_odb(r);
970 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
971 prepare_multi_pack_index_one(r, alt->path, 0);
972 prepare_packed_git_one(r, alt->path, 0);
973 }
974 rearrange_packed_git(r);
975
976 r->objects->all_packs = NULL;
977
978 prepare_packed_git_mru(r);
979 r->objects->packed_git_initialized = 1;
980 }
981
982 void reprepare_packed_git(struct repository *r)
983 {
984 r->objects->approximate_object_count_valid = 0;
985 r->objects->packed_git_initialized = 0;
986 prepare_packed_git(r);
987 }
988
989 struct packed_git *get_packed_git(struct repository *r)
990 {
991 prepare_packed_git(r);
992 return r->objects->packed_git;
993 }
994
995 struct multi_pack_index *get_multi_pack_index(struct repository *r)
996 {
997 prepare_packed_git(r);
998 return r->objects->multi_pack_index;
999 }
1000
1001 struct packed_git *get_all_packs(struct repository *r)
1002 {
1003 prepare_packed_git(r);
1004
1005 if (!r->objects->all_packs) {
1006 struct packed_git *p = r->objects->packed_git;
1007 struct multi_pack_index *m;
1008
1009 for (m = r->objects->multi_pack_index; m; m = m->next) {
1010 uint32_t i;
1011 for (i = 0; i < m->num_packs; i++) {
1012 if (!prepare_midx_pack(m, i)) {
1013 m->packs[i]->next = p;
1014 p = m->packs[i];
1015 }
1016 }
1017 }
1018
1019 r->objects->all_packs = p;
1020 }
1021
1022 return r->objects->all_packs;
1023 }
1024
1025 struct list_head *get_packed_git_mru(struct repository *r)
1026 {
1027 prepare_packed_git(r);
1028 return &r->objects->packed_git_mru;
1029 }
1030
1031 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1032 unsigned long len, enum object_type *type, unsigned long *sizep)
1033 {
1034 unsigned shift;
1035 unsigned long size, c;
1036 unsigned long used = 0;
1037
1038 c = buf[used++];
1039 *type = (c >> 4) & 7;
1040 size = c & 15;
1041 shift = 4;
1042 while (c & 0x80) {
1043 if (len <= used || bitsizeof(long) <= shift) {
1044 error("bad object header");
1045 size = used = 0;
1046 break;
1047 }
1048 c = buf[used++];
1049 size += (c & 0x7f) << shift;
1050 shift += 7;
1051 }
1052 *sizep = size;
1053 return used;
1054 }
1055
1056 unsigned long get_size_from_delta(struct packed_git *p,
1057 struct pack_window **w_curs,
1058 off_t curpos)
1059 {
1060 const unsigned char *data;
1061 unsigned char delta_head[20], *in;
1062 git_zstream stream;
1063 int st;
1064
1065 memset(&stream, 0, sizeof(stream));
1066 stream.next_out = delta_head;
1067 stream.avail_out = sizeof(delta_head);
1068
1069 git_inflate_init(&stream);
1070 do {
1071 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1072 stream.next_in = in;
1073 st = git_inflate(&stream, Z_FINISH);
1074 curpos += stream.next_in - in;
1075 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1076 stream.total_out < sizeof(delta_head));
1077 git_inflate_end(&stream);
1078 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1079 error("delta data unpack-initial failed");
1080 return 0;
1081 }
1082
1083 /* Examine the initial part of the delta to figure out
1084 * the result size.
1085 */
1086 data = delta_head;
1087
1088 /* ignore base size */
1089 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1090
1091 /* Read the result size */
1092 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1093 }
1094
1095 int unpack_object_header(struct packed_git *p,
1096 struct pack_window **w_curs,
1097 off_t *curpos,
1098 unsigned long *sizep)
1099 {
1100 unsigned char *base;
1101 unsigned long left;
1102 unsigned long used;
1103 enum object_type type;
1104
1105 /* use_pack() assures us we have [base, base + 20) available
1106 * as a range that we can look at. (Its actually the hash
1107 * size that is assured.) With our object header encoding
1108 * the maximum deflated object size is 2^137, which is just
1109 * insane, so we know won't exceed what we have been given.
1110 */
1111 base = use_pack(p, w_curs, *curpos, &left);
1112 used = unpack_object_header_buffer(base, left, &type, sizep);
1113 if (!used) {
1114 type = OBJ_BAD;
1115 } else
1116 *curpos += used;
1117
1118 return type;
1119 }
1120
1121 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1122 {
1123 unsigned i;
1124 for (i = 0; i < p->num_bad_objects; i++)
1125 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1126 return;
1127 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1128 st_mult(GIT_MAX_RAWSZ,
1129 st_add(p->num_bad_objects, 1)));
1130 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1131 p->num_bad_objects++;
1132 }
1133
1134 const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1135 {
1136 struct packed_git *p;
1137 unsigned i;
1138
1139 for (p = the_repository->objects->packed_git; p; p = p->next)
1140 for (i = 0; i < p->num_bad_objects; i++)
1141 if (!hashcmp(sha1,
1142 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1143 return p;
1144 return NULL;
1145 }
1146
1147 static off_t get_delta_base(struct packed_git *p,
1148 struct pack_window **w_curs,
1149 off_t *curpos,
1150 enum object_type type,
1151 off_t delta_obj_offset)
1152 {
1153 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1154 off_t base_offset;
1155
1156 /* use_pack() assured us we have [base_info, base_info + 20)
1157 * as a range that we can look at without walking off the
1158 * end of the mapped window. Its actually the hash size
1159 * that is assured. An OFS_DELTA longer than the hash size
1160 * is stupid, as then a REF_DELTA would be smaller to store.
1161 */
1162 if (type == OBJ_OFS_DELTA) {
1163 unsigned used = 0;
1164 unsigned char c = base_info[used++];
1165 base_offset = c & 127;
1166 while (c & 128) {
1167 base_offset += 1;
1168 if (!base_offset || MSB(base_offset, 7))
1169 return 0; /* overflow */
1170 c = base_info[used++];
1171 base_offset = (base_offset << 7) + (c & 127);
1172 }
1173 base_offset = delta_obj_offset - base_offset;
1174 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1175 return 0; /* out of bound */
1176 *curpos += used;
1177 } else if (type == OBJ_REF_DELTA) {
1178 /* The base entry _must_ be in the same pack */
1179 base_offset = find_pack_entry_one(base_info, p);
1180 *curpos += the_hash_algo->rawsz;
1181 } else
1182 die("I am totally screwed");
1183 return base_offset;
1184 }
1185
1186 /*
1187 * Like get_delta_base above, but we return the sha1 instead of the pack
1188 * offset. This means it is cheaper for REF deltas (we do not have to do
1189 * the final object lookup), but more expensive for OFS deltas (we
1190 * have to load the revidx to convert the offset back into a sha1).
1191 */
1192 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1193 struct pack_window **w_curs,
1194 off_t curpos,
1195 enum object_type type,
1196 off_t delta_obj_offset)
1197 {
1198 if (type == OBJ_REF_DELTA) {
1199 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1200 return base;
1201 } else if (type == OBJ_OFS_DELTA) {
1202 struct revindex_entry *revidx;
1203 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1204 type, delta_obj_offset);
1205
1206 if (!base_offset)
1207 return NULL;
1208
1209 revidx = find_pack_revindex(p, base_offset);
1210 if (!revidx)
1211 return NULL;
1212
1213 return nth_packed_object_sha1(p, revidx->nr);
1214 } else
1215 return NULL;
1216 }
1217
1218 static int retry_bad_packed_offset(struct repository *r,
1219 struct packed_git *p,
1220 off_t obj_offset)
1221 {
1222 int type;
1223 struct revindex_entry *revidx;
1224 struct object_id oid;
1225 revidx = find_pack_revindex(p, obj_offset);
1226 if (!revidx)
1227 return OBJ_BAD;
1228 nth_packed_object_oid(&oid, p, revidx->nr);
1229 mark_bad_packed_object(p, oid.hash);
1230 type = oid_object_info(r, &oid, NULL);
1231 if (type <= OBJ_NONE)
1232 return OBJ_BAD;
1233 return type;
1234 }
1235
1236 #define POI_STACK_PREALLOC 64
1237
1238 static enum object_type packed_to_object_type(struct repository *r,
1239 struct packed_git *p,
1240 off_t obj_offset,
1241 enum object_type type,
1242 struct pack_window **w_curs,
1243 off_t curpos)
1244 {
1245 off_t small_poi_stack[POI_STACK_PREALLOC];
1246 off_t *poi_stack = small_poi_stack;
1247 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1248
1249 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1250 off_t base_offset;
1251 unsigned long size;
1252 /* Push the object we're going to leave behind */
1253 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1254 poi_stack_alloc = alloc_nr(poi_stack_nr);
1255 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1256 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1257 } else {
1258 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1259 }
1260 poi_stack[poi_stack_nr++] = obj_offset;
1261 /* If parsing the base offset fails, just unwind */
1262 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1263 if (!base_offset)
1264 goto unwind;
1265 curpos = obj_offset = base_offset;
1266 type = unpack_object_header(p, w_curs, &curpos, &size);
1267 if (type <= OBJ_NONE) {
1268 /* If getting the base itself fails, we first
1269 * retry the base, otherwise unwind */
1270 type = retry_bad_packed_offset(r, p, base_offset);
1271 if (type > OBJ_NONE)
1272 goto out;
1273 goto unwind;
1274 }
1275 }
1276
1277 switch (type) {
1278 case OBJ_BAD:
1279 case OBJ_COMMIT:
1280 case OBJ_TREE:
1281 case OBJ_BLOB:
1282 case OBJ_TAG:
1283 break;
1284 default:
1285 error("unknown object type %i at offset %"PRIuMAX" in %s",
1286 type, (uintmax_t)obj_offset, p->pack_name);
1287 type = OBJ_BAD;
1288 }
1289
1290 out:
1291 if (poi_stack != small_poi_stack)
1292 free(poi_stack);
1293 return type;
1294
1295 unwind:
1296 while (poi_stack_nr) {
1297 obj_offset = poi_stack[--poi_stack_nr];
1298 type = retry_bad_packed_offset(r, p, obj_offset);
1299 if (type > OBJ_NONE)
1300 goto out;
1301 }
1302 type = OBJ_BAD;
1303 goto out;
1304 }
1305
1306 static struct hashmap delta_base_cache;
1307 static size_t delta_base_cached;
1308
1309 static LIST_HEAD(delta_base_cache_lru);
1310
1311 struct delta_base_cache_key {
1312 struct packed_git *p;
1313 off_t base_offset;
1314 };
1315
1316 struct delta_base_cache_entry {
1317 struct hashmap hash;
1318 struct delta_base_cache_key key;
1319 struct list_head lru;
1320 void *data;
1321 unsigned long size;
1322 enum object_type type;
1323 };
1324
1325 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1326 {
1327 unsigned int hash;
1328
1329 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1330 hash += (hash >> 8) + (hash >> 16);
1331 return hash;
1332 }
1333
1334 static struct delta_base_cache_entry *
1335 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1336 {
1337 struct hashmap_entry entry;
1338 struct delta_base_cache_key key;
1339
1340 if (!delta_base_cache.cmpfn)
1341 return NULL;
1342
1343 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1344 key.p = p;
1345 key.base_offset = base_offset;
1346 return hashmap_get(&delta_base_cache, &entry, &key);
1347 }
1348
1349 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1350 const struct delta_base_cache_key *b)
1351 {
1352 return a->p == b->p && a->base_offset == b->base_offset;
1353 }
1354
1355 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1356 const void *va, const void *vb,
1357 const void *vkey)
1358 {
1359 const struct delta_base_cache_entry *a = va, *b = vb;
1360 const struct delta_base_cache_key *key = vkey;
1361 if (key)
1362 return !delta_base_cache_key_eq(&a->key, key);
1363 else
1364 return !delta_base_cache_key_eq(&a->key, &b->key);
1365 }
1366
1367 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1368 {
1369 return !!get_delta_base_cache_entry(p, base_offset);
1370 }
1371
1372 /*
1373 * Remove the entry from the cache, but do _not_ free the associated
1374 * entry data. The caller takes ownership of the "data" buffer, and
1375 * should copy out any fields it wants before detaching.
1376 */
1377 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1378 {
1379 hashmap_remove(&delta_base_cache, ent, &ent->key);
1380 list_del(&ent->lru);
1381 delta_base_cached -= ent->size;
1382 free(ent);
1383 }
1384
1385 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1386 off_t base_offset, unsigned long *base_size,
1387 enum object_type *type)
1388 {
1389 struct delta_base_cache_entry *ent;
1390
1391 ent = get_delta_base_cache_entry(p, base_offset);
1392 if (!ent)
1393 return unpack_entry(r, p, base_offset, type, base_size);
1394
1395 if (type)
1396 *type = ent->type;
1397 if (base_size)
1398 *base_size = ent->size;
1399 return xmemdupz(ent->data, ent->size);
1400 }
1401
1402 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1403 {
1404 free(ent->data);
1405 detach_delta_base_cache_entry(ent);
1406 }
1407
1408 void clear_delta_base_cache(void)
1409 {
1410 struct list_head *lru, *tmp;
1411 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1412 struct delta_base_cache_entry *entry =
1413 list_entry(lru, struct delta_base_cache_entry, lru);
1414 release_delta_base_cache(entry);
1415 }
1416 }
1417
1418 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1419 void *base, unsigned long base_size, enum object_type type)
1420 {
1421 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1422 struct list_head *lru, *tmp;
1423
1424 delta_base_cached += base_size;
1425
1426 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1427 struct delta_base_cache_entry *f =
1428 list_entry(lru, struct delta_base_cache_entry, lru);
1429 if (delta_base_cached <= delta_base_cache_limit)
1430 break;
1431 release_delta_base_cache(f);
1432 }
1433
1434 ent->key.p = p;
1435 ent->key.base_offset = base_offset;
1436 ent->type = type;
1437 ent->data = base;
1438 ent->size = base_size;
1439 list_add_tail(&ent->lru, &delta_base_cache_lru);
1440
1441 if (!delta_base_cache.cmpfn)
1442 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1443 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1444 hashmap_add(&delta_base_cache, ent);
1445 }
1446
1447 int packed_object_info(struct repository *r, struct packed_git *p,
1448 off_t obj_offset, struct object_info *oi)
1449 {
1450 struct pack_window *w_curs = NULL;
1451 unsigned long size;
1452 off_t curpos = obj_offset;
1453 enum object_type type;
1454
1455 /*
1456 * We always get the representation type, but only convert it to
1457 * a "real" type later if the caller is interested.
1458 */
1459 if (oi->contentp) {
1460 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1461 &type);
1462 if (!*oi->contentp)
1463 type = OBJ_BAD;
1464 } else {
1465 type = unpack_object_header(p, &w_curs, &curpos, &size);
1466 }
1467
1468 if (!oi->contentp && oi->sizep) {
1469 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1470 off_t tmp_pos = curpos;
1471 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1472 type, obj_offset);
1473 if (!base_offset) {
1474 type = OBJ_BAD;
1475 goto out;
1476 }
1477 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1478 if (*oi->sizep == 0) {
1479 type = OBJ_BAD;
1480 goto out;
1481 }
1482 } else {
1483 *oi->sizep = size;
1484 }
1485 }
1486
1487 if (oi->disk_sizep) {
1488 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1489 *oi->disk_sizep = revidx[1].offset - obj_offset;
1490 }
1491
1492 if (oi->typep || oi->type_name) {
1493 enum object_type ptot;
1494 ptot = packed_to_object_type(r, p, obj_offset,
1495 type, &w_curs, curpos);
1496 if (oi->typep)
1497 *oi->typep = ptot;
1498 if (oi->type_name) {
1499 const char *tn = type_name(ptot);
1500 if (tn)
1501 strbuf_addstr(oi->type_name, tn);
1502 }
1503 if (ptot < 0) {
1504 type = OBJ_BAD;
1505 goto out;
1506 }
1507 }
1508
1509 if (oi->delta_base_sha1) {
1510 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1511 const unsigned char *base;
1512
1513 base = get_delta_base_sha1(p, &w_curs, curpos,
1514 type, obj_offset);
1515 if (!base) {
1516 type = OBJ_BAD;
1517 goto out;
1518 }
1519
1520 hashcpy(oi->delta_base_sha1, base);
1521 } else
1522 hashclr(oi->delta_base_sha1);
1523 }
1524
1525 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1526 OI_PACKED;
1527
1528 out:
1529 unuse_pack(&w_curs);
1530 return type;
1531 }
1532
1533 static void *unpack_compressed_entry(struct packed_git *p,
1534 struct pack_window **w_curs,
1535 off_t curpos,
1536 unsigned long size)
1537 {
1538 int st;
1539 git_zstream stream;
1540 unsigned char *buffer, *in;
1541
1542 buffer = xmallocz_gently(size);
1543 if (!buffer)
1544 return NULL;
1545 memset(&stream, 0, sizeof(stream));
1546 stream.next_out = buffer;
1547 stream.avail_out = size + 1;
1548
1549 git_inflate_init(&stream);
1550 do {
1551 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1552 stream.next_in = in;
1553 st = git_inflate(&stream, Z_FINISH);
1554 if (!stream.avail_out)
1555 break; /* the payload is larger than it should be */
1556 curpos += stream.next_in - in;
1557 } while (st == Z_OK || st == Z_BUF_ERROR);
1558 git_inflate_end(&stream);
1559 if ((st != Z_STREAM_END) || stream.total_out != size) {
1560 free(buffer);
1561 return NULL;
1562 }
1563
1564 /* versions of zlib can clobber unconsumed portion of outbuf */
1565 buffer[size] = '\0';
1566
1567 return buffer;
1568 }
1569
1570 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1571 {
1572 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1573 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1574 p->pack_name, (uintmax_t)obj_offset);
1575 }
1576
1577 int do_check_packed_object_crc;
1578
1579 #define UNPACK_ENTRY_STACK_PREALLOC 64
1580 struct unpack_entry_stack_ent {
1581 off_t obj_offset;
1582 off_t curpos;
1583 unsigned long size;
1584 };
1585
1586 static void *read_object(struct repository *r,
1587 const struct object_id *oid,
1588 enum object_type *type,
1589 unsigned long *size)
1590 {
1591 struct object_info oi = OBJECT_INFO_INIT;
1592 void *content;
1593 oi.typep = type;
1594 oi.sizep = size;
1595 oi.contentp = &content;
1596
1597 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1598 return NULL;
1599 return content;
1600 }
1601
1602 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1603 enum object_type *final_type, unsigned long *final_size)
1604 {
1605 struct pack_window *w_curs = NULL;
1606 off_t curpos = obj_offset;
1607 void *data = NULL;
1608 unsigned long size;
1609 enum object_type type;
1610 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1611 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1612 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1613 int base_from_cache = 0;
1614
1615 write_pack_access_log(p, obj_offset);
1616
1617 /* PHASE 1: drill down to the innermost base object */
1618 for (;;) {
1619 off_t base_offset;
1620 int i;
1621 struct delta_base_cache_entry *ent;
1622
1623 ent = get_delta_base_cache_entry(p, curpos);
1624 if (ent) {
1625 type = ent->type;
1626 data = ent->data;
1627 size = ent->size;
1628 detach_delta_base_cache_entry(ent);
1629 base_from_cache = 1;
1630 break;
1631 }
1632
1633 if (do_check_packed_object_crc && p->index_version > 1) {
1634 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1635 off_t len = revidx[1].offset - obj_offset;
1636 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1637 struct object_id oid;
1638 nth_packed_object_oid(&oid, p, revidx->nr);
1639 error("bad packed object CRC for %s",
1640 oid_to_hex(&oid));
1641 mark_bad_packed_object(p, oid.hash);
1642 data = NULL;
1643 goto out;
1644 }
1645 }
1646
1647 type = unpack_object_header(p, &w_curs, &curpos, &size);
1648 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1649 break;
1650
1651 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1652 if (!base_offset) {
1653 error("failed to validate delta base reference "
1654 "at offset %"PRIuMAX" from %s",
1655 (uintmax_t)curpos, p->pack_name);
1656 /* bail to phase 2, in hopes of recovery */
1657 data = NULL;
1658 break;
1659 }
1660
1661 /* push object, proceed to base */
1662 if (delta_stack_nr >= delta_stack_alloc
1663 && delta_stack == small_delta_stack) {
1664 delta_stack_alloc = alloc_nr(delta_stack_nr);
1665 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1666 memcpy(delta_stack, small_delta_stack,
1667 sizeof(*delta_stack)*delta_stack_nr);
1668 } else {
1669 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1670 }
1671 i = delta_stack_nr++;
1672 delta_stack[i].obj_offset = obj_offset;
1673 delta_stack[i].curpos = curpos;
1674 delta_stack[i].size = size;
1675
1676 curpos = obj_offset = base_offset;
1677 }
1678
1679 /* PHASE 2: handle the base */
1680 switch (type) {
1681 case OBJ_OFS_DELTA:
1682 case OBJ_REF_DELTA:
1683 if (data)
1684 BUG("unpack_entry: left loop at a valid delta");
1685 break;
1686 case OBJ_COMMIT:
1687 case OBJ_TREE:
1688 case OBJ_BLOB:
1689 case OBJ_TAG:
1690 if (!base_from_cache)
1691 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1692 break;
1693 default:
1694 data = NULL;
1695 error("unknown object type %i at offset %"PRIuMAX" in %s",
1696 type, (uintmax_t)obj_offset, p->pack_name);
1697 }
1698
1699 /* PHASE 3: apply deltas in order */
1700
1701 /* invariants:
1702 * 'data' holds the base data, or NULL if there was corruption
1703 */
1704 while (delta_stack_nr) {
1705 void *delta_data;
1706 void *base = data;
1707 void *external_base = NULL;
1708 unsigned long delta_size, base_size = size;
1709 int i;
1710
1711 data = NULL;
1712
1713 if (base)
1714 add_delta_base_cache(p, obj_offset, base, base_size, type);
1715
1716 if (!base) {
1717 /*
1718 * We're probably in deep shit, but let's try to fetch
1719 * the required base anyway from another pack or loose.
1720 * This is costly but should happen only in the presence
1721 * of a corrupted pack, and is better than failing outright.
1722 */
1723 struct revindex_entry *revidx;
1724 struct object_id base_oid;
1725 revidx = find_pack_revindex(p, obj_offset);
1726 if (revidx) {
1727 nth_packed_object_oid(&base_oid, p, revidx->nr);
1728 error("failed to read delta base object %s"
1729 " at offset %"PRIuMAX" from %s",
1730 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1731 p->pack_name);
1732 mark_bad_packed_object(p, base_oid.hash);
1733 base = read_object(r, &base_oid, &type, &base_size);
1734 external_base = base;
1735 }
1736 }
1737
1738 i = --delta_stack_nr;
1739 obj_offset = delta_stack[i].obj_offset;
1740 curpos = delta_stack[i].curpos;
1741 delta_size = delta_stack[i].size;
1742
1743 if (!base)
1744 continue;
1745
1746 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1747
1748 if (!delta_data) {
1749 error("failed to unpack compressed delta "
1750 "at offset %"PRIuMAX" from %s",
1751 (uintmax_t)curpos, p->pack_name);
1752 data = NULL;
1753 free(external_base);
1754 continue;
1755 }
1756
1757 data = patch_delta(base, base_size,
1758 delta_data, delta_size,
1759 &size);
1760
1761 /*
1762 * We could not apply the delta; warn the user, but keep going.
1763 * Our failure will be noticed either in the next iteration of
1764 * the loop, or if this is the final delta, in the caller when
1765 * we return NULL. Those code paths will take care of making
1766 * a more explicit warning and retrying with another copy of
1767 * the object.
1768 */
1769 if (!data)
1770 error("failed to apply delta");
1771
1772 free(delta_data);
1773 free(external_base);
1774 }
1775
1776 if (final_type)
1777 *final_type = type;
1778 if (final_size)
1779 *final_size = size;
1780
1781 out:
1782 unuse_pack(&w_curs);
1783
1784 if (delta_stack != small_delta_stack)
1785 free(delta_stack);
1786
1787 return data;
1788 }
1789
1790 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1791 {
1792 const unsigned char *index_fanout = p->index_data;
1793 const unsigned char *index_lookup;
1794 const unsigned int hashsz = the_hash_algo->rawsz;
1795 int index_lookup_width;
1796
1797 if (!index_fanout)
1798 BUG("bsearch_pack called without a valid pack-index");
1799
1800 index_lookup = index_fanout + 4 * 256;
1801 if (p->index_version == 1) {
1802 index_lookup_width = hashsz + 4;
1803 index_lookup += 4;
1804 } else {
1805 index_lookup_width = hashsz;
1806 index_fanout += 8;
1807 index_lookup += 8;
1808 }
1809
1810 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1811 index_lookup, index_lookup_width, result);
1812 }
1813
1814 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1815 uint32_t n)
1816 {
1817 const unsigned char *index = p->index_data;
1818 const unsigned int hashsz = the_hash_algo->rawsz;
1819 if (!index) {
1820 if (open_pack_index(p))
1821 return NULL;
1822 index = p->index_data;
1823 }
1824 if (n >= p->num_objects)
1825 return NULL;
1826 index += 4 * 256;
1827 if (p->index_version == 1) {
1828 return index + (hashsz + 4) * n + 4;
1829 } else {
1830 index += 8;
1831 return index + hashsz * n;
1832 }
1833 }
1834
1835 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1836 struct packed_git *p,
1837 uint32_t n)
1838 {
1839 const unsigned char *hash = nth_packed_object_sha1(p, n);
1840 if (!hash)
1841 return NULL;
1842 hashcpy(oid->hash, hash);
1843 return oid;
1844 }
1845
1846 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1847 {
1848 const unsigned char *ptr = vptr;
1849 const unsigned char *start = p->index_data;
1850 const unsigned char *end = start + p->index_size;
1851 if (ptr < start)
1852 die(_("offset before start of pack index for %s (corrupt index?)"),
1853 p->pack_name);
1854 /* No need to check for underflow; .idx files must be at least 8 bytes */
1855 if (ptr >= end - 8)
1856 die(_("offset beyond end of pack index for %s (truncated index?)"),
1857 p->pack_name);
1858 }
1859
1860 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1861 {
1862 const unsigned char *index = p->index_data;
1863 const unsigned int hashsz = the_hash_algo->rawsz;
1864 index += 4 * 256;
1865 if (p->index_version == 1) {
1866 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1867 } else {
1868 uint32_t off;
1869 index += 8 + p->num_objects * (hashsz + 4);
1870 off = ntohl(*((uint32_t *)(index + 4 * n)));
1871 if (!(off & 0x80000000))
1872 return off;
1873 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1874 check_pack_index_ptr(p, index);
1875 return get_be64(index);
1876 }
1877 }
1878
1879 off_t find_pack_entry_one(const unsigned char *sha1,
1880 struct packed_git *p)
1881 {
1882 const unsigned char *index = p->index_data;
1883 struct object_id oid;
1884 uint32_t result;
1885
1886 if (!index) {
1887 if (open_pack_index(p))
1888 return 0;
1889 }
1890
1891 hashcpy(oid.hash, sha1);
1892 if (bsearch_pack(&oid, p, &result))
1893 return nth_packed_object_offset(p, result);
1894 return 0;
1895 }
1896
1897 int is_pack_valid(struct packed_git *p)
1898 {
1899 /* An already open pack is known to be valid. */
1900 if (p->pack_fd != -1)
1901 return 1;
1902
1903 /* If the pack has one window completely covering the
1904 * file size, the pack is known to be valid even if
1905 * the descriptor is not currently open.
1906 */
1907 if (p->windows) {
1908 struct pack_window *w = p->windows;
1909
1910 if (!w->offset && w->len == p->pack_size)
1911 return 1;
1912 }
1913
1914 /* Force the pack to open to prove its valid. */
1915 return !open_packed_git(p);
1916 }
1917
1918 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1919 struct packed_git *packs)
1920 {
1921 struct packed_git *p;
1922
1923 for (p = packs; p; p = p->next) {
1924 if (find_pack_entry_one(sha1, p))
1925 return p;
1926 }
1927 return NULL;
1928
1929 }
1930
1931 static int fill_pack_entry(const struct object_id *oid,
1932 struct pack_entry *e,
1933 struct packed_git *p)
1934 {
1935 off_t offset;
1936
1937 if (p->num_bad_objects) {
1938 unsigned i;
1939 for (i = 0; i < p->num_bad_objects; i++)
1940 if (!hashcmp(oid->hash,
1941 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1942 return 0;
1943 }
1944
1945 offset = find_pack_entry_one(oid->hash, p);
1946 if (!offset)
1947 return 0;
1948
1949 /*
1950 * We are about to tell the caller where they can locate the
1951 * requested object. We better make sure the packfile is
1952 * still here and can be accessed before supplying that
1953 * answer, as it may have been deleted since the index was
1954 * loaded!
1955 */
1956 if (!is_pack_valid(p))
1957 return 0;
1958 e->offset = offset;
1959 e->p = p;
1960 return 1;
1961 }
1962
1963 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1964 {
1965 struct list_head *pos;
1966 struct multi_pack_index *m;
1967
1968 prepare_packed_git(r);
1969 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1970 return 0;
1971
1972 for (m = r->objects->multi_pack_index; m; m = m->next) {
1973 if (fill_midx_entry(oid, e, m))
1974 return 1;
1975 }
1976
1977 list_for_each(pos, &r->objects->packed_git_mru) {
1978 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1979 if (fill_pack_entry(oid, e, p)) {
1980 list_move(&p->mru, &r->objects->packed_git_mru);
1981 return 1;
1982 }
1983 }
1984 return 0;
1985 }
1986
1987 int has_object_pack(const struct object_id *oid)
1988 {
1989 struct pack_entry e;
1990 return find_pack_entry(the_repository, oid, &e);
1991 }
1992
1993 int has_pack_index(const unsigned char *sha1)
1994 {
1995 struct stat st;
1996 if (stat(sha1_pack_index_name(sha1), &st))
1997 return 0;
1998 return 1;
1999 }
2000
2001 int for_each_object_in_pack(struct packed_git *p,
2002 each_packed_object_fn cb, void *data,
2003 enum for_each_object_flags flags)
2004 {
2005 uint32_t i;
2006 int r = 0;
2007
2008 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2009 load_pack_revindex(p);
2010
2011 for (i = 0; i < p->num_objects; i++) {
2012 uint32_t pos;
2013 struct object_id oid;
2014
2015 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2016 pos = p->revindex[i].nr;
2017 else
2018 pos = i;
2019
2020 if (!nth_packed_object_oid(&oid, p, pos))
2021 return error("unable to get sha1 of object %u in %s",
2022 pos, p->pack_name);
2023
2024 r = cb(&oid, p, pos, data);
2025 if (r)
2026 break;
2027 }
2028 return r;
2029 }
2030
2031 int for_each_packed_object(each_packed_object_fn cb, void *data,
2032 enum for_each_object_flags flags)
2033 {
2034 struct packed_git *p;
2035 int r = 0;
2036 int pack_errors = 0;
2037
2038 prepare_packed_git(the_repository);
2039 for (p = the_repository->objects->packed_git; p; p = p->next) {
2040 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2041 continue;
2042 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2043 !p->pack_promisor)
2044 continue;
2045 if (open_pack_index(p)) {
2046 pack_errors = 1;
2047 continue;
2048 }
2049 r = for_each_object_in_pack(p, cb, data, flags);
2050 if (r)
2051 break;
2052 }
2053 return r ? r : pack_errors;
2054 }
2055
2056 static int add_promisor_object(const struct object_id *oid,
2057 struct packed_git *pack,
2058 uint32_t pos,
2059 void *set_)
2060 {
2061 struct oidset *set = set_;
2062 struct object *obj = parse_object(the_repository, oid);
2063 if (!obj)
2064 return 1;
2065
2066 oidset_insert(set, oid);
2067
2068 /*
2069 * If this is a tree, commit, or tag, the objects it refers
2070 * to are also promisor objects. (Blobs refer to no objects->)
2071 */
2072 if (obj->type == OBJ_TREE) {
2073 struct tree *tree = (struct tree *)obj;
2074 struct tree_desc desc;
2075 struct name_entry entry;
2076 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2077 /*
2078 * Error messages are given when packs are
2079 * verified, so do not print any here.
2080 */
2081 return 0;
2082 while (tree_entry_gently(&desc, &entry))
2083 oidset_insert(set, entry.oid);
2084 } else if (obj->type == OBJ_COMMIT) {
2085 struct commit *commit = (struct commit *) obj;
2086 struct commit_list *parents = commit->parents;
2087
2088 oidset_insert(set, get_commit_tree_oid(commit));
2089 for (; parents; parents = parents->next)
2090 oidset_insert(set, &parents->item->object.oid);
2091 } else if (obj->type == OBJ_TAG) {
2092 struct tag *tag = (struct tag *) obj;
2093 oidset_insert(set, &tag->tagged->oid);
2094 }
2095 return 0;
2096 }
2097
2098 int is_promisor_object(const struct object_id *oid)
2099 {
2100 static struct oidset promisor_objects;
2101 static int promisor_objects_prepared;
2102
2103 if (!promisor_objects_prepared) {
2104 if (repository_format_partial_clone) {
2105 for_each_packed_object(add_promisor_object,
2106 &promisor_objects,
2107 FOR_EACH_OBJECT_PROMISOR_ONLY);
2108 }
2109 promisor_objects_prepared = 1;
2110 }
2111 return oidset_contains(&promisor_objects, oid);
2112 }