promisor-remote: skip move_to_tail when no-op
[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 #include "promisor-remote.h"
20
21 char *odb_pack_name(struct strbuf *buf,
22 const unsigned char *sha1,
23 const char *ext)
24 {
25 strbuf_reset(buf);
26 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27 sha1_to_hex(sha1), ext);
28 return buf->buf;
29 }
30
31 char *sha1_pack_name(const unsigned char *sha1)
32 {
33 static struct strbuf buf = STRBUF_INIT;
34 return odb_pack_name(&buf, sha1, "pack");
35 }
36
37 char *sha1_pack_index_name(const unsigned char *sha1)
38 {
39 static struct strbuf buf = STRBUF_INIT;
40 return odb_pack_name(&buf, sha1, "idx");
41 }
42
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
51
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
54
55 void pack_report(void)
56 {
57 fprintf(stderr,
58 "pack_report: getpagesize() = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size),
63 sz_fmt(packed_git_limit));
64 fprintf(stderr,
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT " / %10" SZ_FMT "\n",
70 pack_used_ctr,
71 pack_mmap_calls,
72 pack_open_windows, peak_pack_open_windows,
73 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
74 }
75
76 /*
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
79 * success.
80 */
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
82 {
83 void *idx_map;
84 size_t idx_size;
85 int fd = git_open(path), ret;
86 struct stat st;
87 const unsigned int hashsz = the_hash_algo->rawsz;
88
89 if (fd < 0)
90 return -1;
91 if (fstat(fd, &st)) {
92 close(fd);
93 return -1;
94 }
95 idx_size = xsize_t(st.st_size);
96 if (idx_size < 4 * 256 + hashsz + hashsz) {
97 close(fd);
98 return error("index file %s is too small", path);
99 }
100 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
101 close(fd);
102
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
104
105 if (ret)
106 munmap(idx_map, idx_size);
107
108 return ret;
109 }
110
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112 size_t idx_size, struct packed_git *p)
113 {
114 struct pack_idx_header *hdr = idx_map;
115 uint32_t version, nr, i, *index;
116
117 if (idx_size < 4 * 256 + hashsz + hashsz)
118 return error("index file %s is too small", path);
119 if (idx_map == NULL)
120 return error("empty data");
121
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123 version = ntohl(hdr->idx_version);
124 if (version < 2 || version > 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
128 path, version);
129 } else
130 version = 1;
131
132 nr = 0;
133 index = idx_map;
134 if (version > 1)
135 index += 2; /* skip index header */
136 for (i = 0; i < 256; i++) {
137 uint32_t n = ntohl(index[i]);
138 if (n < nr)
139 return error("non-monotonic index %s", path);
140 nr = n;
141 }
142
143 if (version == 1) {
144 /*
145 * Total size:
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
149 * - file checksum
150 */
151 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
152 return error("wrong index v1 file size in %s", path);
153 } else if (version == 2) {
154 /*
155 * Minimum size:
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
162 * - file checksum
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
166 */
167 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
168 unsigned long max_size = min_size;
169 if (nr)
170 max_size += (nr - 1)*8;
171 if (idx_size < min_size || idx_size > max_size)
172 return error("wrong index v2 file size in %s", path);
173 if (idx_size != min_size &&
174 /*
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
178 */
179 (sizeof(off_t) <= 4))
180 return error("pack too large for current definition of off_t in %s", path);
181 }
182
183 p->index_version = version;
184 p->index_data = idx_map;
185 p->index_size = idx_size;
186 p->num_objects = nr;
187 return 0;
188 }
189
190 int open_pack_index(struct packed_git *p)
191 {
192 char *idx_name;
193 size_t len;
194 int ret;
195
196 if (p->index_data)
197 return 0;
198
199 if (!strip_suffix(p->pack_name, ".pack", &len))
200 BUG("pack_name does not end in .pack");
201 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
202 ret = check_packed_git_idx(idx_name, p);
203 free(idx_name);
204 return ret;
205 }
206
207 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
208 {
209 const uint32_t *level1_ofs = p->index_data;
210
211 if (!level1_ofs) {
212 if (open_pack_index(p))
213 return 0;
214 level1_ofs = p->index_data;
215 }
216
217 if (p->index_version > 1) {
218 level1_ofs += 2;
219 }
220
221 return ntohl(level1_ofs[value]);
222 }
223
224 static struct packed_git *alloc_packed_git(int extra)
225 {
226 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
227 memset(p, 0, sizeof(*p));
228 p->pack_fd = -1;
229 return p;
230 }
231
232 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
233 {
234 const char *path = sha1_pack_name(sha1);
235 size_t alloc = st_add(strlen(path), 1);
236 struct packed_git *p = alloc_packed_git(alloc);
237
238 memcpy(p->pack_name, path, alloc); /* includes NUL */
239 hashcpy(p->hash, sha1);
240 if (check_packed_git_idx(idx_path, p)) {
241 free(p);
242 return NULL;
243 }
244
245 return p;
246 }
247
248 static void scan_windows(struct packed_git *p,
249 struct packed_git **lru_p,
250 struct pack_window **lru_w,
251 struct pack_window **lru_l)
252 {
253 struct pack_window *w, *w_l;
254
255 for (w_l = NULL, w = p->windows; w; w = w->next) {
256 if (!w->inuse_cnt) {
257 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
258 *lru_p = p;
259 *lru_w = w;
260 *lru_l = w_l;
261 }
262 }
263 w_l = w;
264 }
265 }
266
267 static int unuse_one_window(struct packed_git *current)
268 {
269 struct packed_git *p, *lru_p = NULL;
270 struct pack_window *lru_w = NULL, *lru_l = NULL;
271
272 if (current)
273 scan_windows(current, &lru_p, &lru_w, &lru_l);
274 for (p = the_repository->objects->packed_git; p; p = p->next)
275 scan_windows(p, &lru_p, &lru_w, &lru_l);
276 if (lru_p) {
277 munmap(lru_w->base, lru_w->len);
278 pack_mapped -= lru_w->len;
279 if (lru_l)
280 lru_l->next = lru_w->next;
281 else
282 lru_p->windows = lru_w->next;
283 free(lru_w);
284 pack_open_windows--;
285 return 1;
286 }
287 return 0;
288 }
289
290 void release_pack_memory(size_t need)
291 {
292 size_t cur = pack_mapped;
293 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
294 ; /* nothing */
295 }
296
297 void close_pack_windows(struct packed_git *p)
298 {
299 while (p->windows) {
300 struct pack_window *w = p->windows;
301
302 if (w->inuse_cnt)
303 die("pack '%s' still has open windows to it",
304 p->pack_name);
305 munmap(w->base, w->len);
306 pack_mapped -= w->len;
307 pack_open_windows--;
308 p->windows = w->next;
309 free(w);
310 }
311 }
312
313 int close_pack_fd(struct packed_git *p)
314 {
315 if (p->pack_fd < 0)
316 return 0;
317
318 close(p->pack_fd);
319 pack_open_fds--;
320 p->pack_fd = -1;
321
322 return 1;
323 }
324
325 void close_pack_index(struct packed_git *p)
326 {
327 if (p->index_data) {
328 munmap((void *)p->index_data, p->index_size);
329 p->index_data = NULL;
330 }
331 }
332
333 void close_pack(struct packed_git *p)
334 {
335 close_pack_windows(p);
336 close_pack_fd(p);
337 close_pack_index(p);
338 }
339
340 void close_all_packs(struct raw_object_store *o)
341 {
342 struct packed_git *p;
343
344 for (p = o->packed_git; p; p = p->next)
345 if (p->do_not_close)
346 BUG("want to close pack marked 'do-not-close'");
347 else
348 close_pack(p);
349
350 if (o->multi_pack_index) {
351 close_midx(o->multi_pack_index);
352 o->multi_pack_index = NULL;
353 }
354 }
355
356 /*
357 * The LRU pack is the one with the oldest MRU window, preferring packs
358 * with no used windows, or the oldest mtime if it has no windows allocated.
359 */
360 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
361 {
362 struct pack_window *w, *this_mru_w;
363 int has_windows_inuse = 0;
364
365 /*
366 * Reject this pack if it has windows and the previously selected
367 * one does not. If this pack does not have windows, reject
368 * it if the pack file is newer than the previously selected one.
369 */
370 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
371 return;
372
373 for (w = this_mru_w = p->windows; w; w = w->next) {
374 /*
375 * Reject this pack if any of its windows are in use,
376 * but the previously selected pack did not have any
377 * inuse windows. Otherwise, record that this pack
378 * has windows in use.
379 */
380 if (w->inuse_cnt) {
381 if (*accept_windows_inuse)
382 has_windows_inuse = 1;
383 else
384 return;
385 }
386
387 if (w->last_used > this_mru_w->last_used)
388 this_mru_w = w;
389
390 /*
391 * Reject this pack if it has windows that have been
392 * used more recently than the previously selected pack.
393 * If the previously selected pack had windows inuse and
394 * we have not encountered a window in this pack that is
395 * inuse, skip this check since we prefer a pack with no
396 * inuse windows to one that has inuse windows.
397 */
398 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
399 this_mru_w->last_used > (*mru_w)->last_used)
400 return;
401 }
402
403 /*
404 * Select this pack.
405 */
406 *mru_w = this_mru_w;
407 *lru_p = p;
408 *accept_windows_inuse = has_windows_inuse;
409 }
410
411 static int close_one_pack(void)
412 {
413 struct packed_git *p, *lru_p = NULL;
414 struct pack_window *mru_w = NULL;
415 int accept_windows_inuse = 1;
416
417 for (p = the_repository->objects->packed_git; p; p = p->next) {
418 if (p->pack_fd == -1)
419 continue;
420 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
421 }
422
423 if (lru_p)
424 return close_pack_fd(lru_p);
425
426 return 0;
427 }
428
429 static unsigned int get_max_fd_limit(void)
430 {
431 #ifdef RLIMIT_NOFILE
432 {
433 struct rlimit lim;
434
435 if (!getrlimit(RLIMIT_NOFILE, &lim))
436 return lim.rlim_cur;
437 }
438 #endif
439
440 #ifdef _SC_OPEN_MAX
441 {
442 long open_max = sysconf(_SC_OPEN_MAX);
443 if (0 < open_max)
444 return open_max;
445 /*
446 * Otherwise, we got -1 for one of the two
447 * reasons:
448 *
449 * (1) sysconf() did not understand _SC_OPEN_MAX
450 * and signaled an error with -1; or
451 * (2) sysconf() said there is no limit.
452 *
453 * We _could_ clear errno before calling sysconf() to
454 * tell these two cases apart and return a huge number
455 * in the latter case to let the caller cap it to a
456 * value that is not so selfish, but letting the
457 * fallback OPEN_MAX codepath take care of these cases
458 * is a lot simpler.
459 */
460 }
461 #endif
462
463 #ifdef OPEN_MAX
464 return OPEN_MAX;
465 #else
466 return 1; /* see the caller ;-) */
467 #endif
468 }
469
470 const char *pack_basename(struct packed_git *p)
471 {
472 const char *ret = strrchr(p->pack_name, '/');
473 if (ret)
474 ret = ret + 1; /* skip past slash */
475 else
476 ret = p->pack_name; /* we only have a base */
477 return ret;
478 }
479
480 /*
481 * Do not call this directly as this leaks p->pack_fd on error return;
482 * call open_packed_git() instead.
483 */
484 static int open_packed_git_1(struct packed_git *p)
485 {
486 struct stat st;
487 struct pack_header hdr;
488 unsigned char hash[GIT_MAX_RAWSZ];
489 unsigned char *idx_hash;
490 long fd_flag;
491 ssize_t read_result;
492 const unsigned hashsz = the_hash_algo->rawsz;
493
494 if (!p->index_data) {
495 struct multi_pack_index *m;
496 const char *pack_name = pack_basename(p);
497
498 for (m = the_repository->objects->multi_pack_index;
499 m; m = m->next) {
500 if (midx_contains_pack(m, pack_name))
501 break;
502 }
503
504 if (!m && open_pack_index(p))
505 return error("packfile %s index unavailable", p->pack_name);
506 }
507
508 if (!pack_max_fds) {
509 unsigned int max_fds = get_max_fd_limit();
510
511 /* Save 3 for stdin/stdout/stderr, 22 for work */
512 if (25 < max_fds)
513 pack_max_fds = max_fds - 25;
514 else
515 pack_max_fds = 1;
516 }
517
518 while (pack_max_fds <= pack_open_fds && close_one_pack())
519 ; /* nothing */
520
521 p->pack_fd = git_open(p->pack_name);
522 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
523 return -1;
524 pack_open_fds++;
525
526 /* If we created the struct before we had the pack we lack size. */
527 if (!p->pack_size) {
528 if (!S_ISREG(st.st_mode))
529 return error("packfile %s not a regular file", p->pack_name);
530 p->pack_size = st.st_size;
531 } else if (p->pack_size != st.st_size)
532 return error("packfile %s size changed", p->pack_name);
533
534 /* We leave these file descriptors open with sliding mmap;
535 * there is no point keeping them open across exec(), though.
536 */
537 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
538 if (fd_flag < 0)
539 return error("cannot determine file descriptor flags");
540 fd_flag |= FD_CLOEXEC;
541 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
542 return error("cannot set FD_CLOEXEC");
543
544 /* Verify we recognize this pack file format. */
545 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
546 if (read_result < 0)
547 return error_errno("error reading from %s", p->pack_name);
548 if (read_result != sizeof(hdr))
549 return error("file %s is far too short to be a packfile", p->pack_name);
550 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
551 return error("file %s is not a GIT packfile", p->pack_name);
552 if (!pack_version_ok(hdr.hdr_version))
553 return error("packfile %s is version %"PRIu32" and not"
554 " supported (try upgrading GIT to a newer version)",
555 p->pack_name, ntohl(hdr.hdr_version));
556
557 /* Skip index checking if in multi-pack-index */
558 if (!p->index_data)
559 return 0;
560
561 /* Verify the pack matches its index. */
562 if (p->num_objects != ntohl(hdr.hdr_entries))
563 return error("packfile %s claims to have %"PRIu32" objects"
564 " while index indicates %"PRIu32" objects",
565 p->pack_name, ntohl(hdr.hdr_entries),
566 p->num_objects);
567 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
568 return error("end of packfile %s is unavailable", p->pack_name);
569 read_result = read_in_full(p->pack_fd, hash, hashsz);
570 if (read_result < 0)
571 return error_errno("error reading from %s", p->pack_name);
572 if (read_result != hashsz)
573 return error("packfile %s signature is unavailable", p->pack_name);
574 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
575 if (!hasheq(hash, idx_hash))
576 return error("packfile %s does not match index", p->pack_name);
577 return 0;
578 }
579
580 static int open_packed_git(struct packed_git *p)
581 {
582 if (!open_packed_git_1(p))
583 return 0;
584 close_pack_fd(p);
585 return -1;
586 }
587
588 static int in_window(struct pack_window *win, off_t offset)
589 {
590 /* We must promise at least one full hash after the
591 * offset is available from this window, otherwise the offset
592 * is not actually in this window and a different window (which
593 * has that one hash excess) must be used. This is to support
594 * the object header and delta base parsing routines below.
595 */
596 off_t win_off = win->offset;
597 return win_off <= offset
598 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
599 }
600
601 unsigned char *use_pack(struct packed_git *p,
602 struct pack_window **w_cursor,
603 off_t offset,
604 unsigned long *left)
605 {
606 struct pack_window *win = *w_cursor;
607
608 /* Since packfiles end in a hash of their content and it's
609 * pointless to ask for an offset into the middle of that
610 * hash, and the in_window function above wouldn't match
611 * don't allow an offset too close to the end of the file.
612 */
613 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
614 die("packfile %s cannot be accessed", p->pack_name);
615 if (offset > (p->pack_size - the_hash_algo->rawsz))
616 die("offset beyond end of packfile (truncated pack?)");
617 if (offset < 0)
618 die(_("offset before end of packfile (broken .idx?)"));
619
620 if (!win || !in_window(win, offset)) {
621 if (win)
622 win->inuse_cnt--;
623 for (win = p->windows; win; win = win->next) {
624 if (in_window(win, offset))
625 break;
626 }
627 if (!win) {
628 size_t window_align = packed_git_window_size / 2;
629 off_t len;
630
631 if (p->pack_fd == -1 && open_packed_git(p))
632 die("packfile %s cannot be accessed", p->pack_name);
633
634 win = xcalloc(1, sizeof(*win));
635 win->offset = (offset / window_align) * window_align;
636 len = p->pack_size - win->offset;
637 if (len > packed_git_window_size)
638 len = packed_git_window_size;
639 win->len = (size_t)len;
640 pack_mapped += win->len;
641 while (packed_git_limit < pack_mapped
642 && unuse_one_window(p))
643 ; /* nothing */
644 win->base = xmmap_gently(NULL, win->len,
645 PROT_READ, MAP_PRIVATE,
646 p->pack_fd, win->offset);
647 if (win->base == MAP_FAILED)
648 die_errno("packfile %s cannot be mapped",
649 p->pack_name);
650 if (!win->offset && win->len == p->pack_size
651 && !p->do_not_close)
652 close_pack_fd(p);
653 pack_mmap_calls++;
654 pack_open_windows++;
655 if (pack_mapped > peak_pack_mapped)
656 peak_pack_mapped = pack_mapped;
657 if (pack_open_windows > peak_pack_open_windows)
658 peak_pack_open_windows = pack_open_windows;
659 win->next = p->windows;
660 p->windows = win;
661 }
662 }
663 if (win != *w_cursor) {
664 win->last_used = pack_used_ctr++;
665 win->inuse_cnt++;
666 *w_cursor = win;
667 }
668 offset -= win->offset;
669 if (left)
670 *left = win->len - xsize_t(offset);
671 return win->base + offset;
672 }
673
674 void unuse_pack(struct pack_window **w_cursor)
675 {
676 struct pack_window *w = *w_cursor;
677 if (w) {
678 w->inuse_cnt--;
679 *w_cursor = NULL;
680 }
681 }
682
683 static void try_to_free_pack_memory(size_t size)
684 {
685 release_pack_memory(size);
686 }
687
688 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
689 {
690 static int have_set_try_to_free_routine;
691 struct stat st;
692 size_t alloc;
693 struct packed_git *p;
694
695 if (!have_set_try_to_free_routine) {
696 have_set_try_to_free_routine = 1;
697 set_try_to_free_routine(try_to_free_pack_memory);
698 }
699
700 /*
701 * Make sure a corresponding .pack file exists and that
702 * the index looks sane.
703 */
704 if (!strip_suffix_mem(path, &path_len, ".idx"))
705 return NULL;
706
707 /*
708 * ".promisor" is long enough to hold any suffix we're adding (and
709 * the use xsnprintf double-checks that)
710 */
711 alloc = st_add3(path_len, strlen(".promisor"), 1);
712 p = alloc_packed_git(alloc);
713 memcpy(p->pack_name, path, path_len);
714
715 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
716 if (!access(p->pack_name, F_OK))
717 p->pack_keep = 1;
718
719 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
720 if (!access(p->pack_name, F_OK))
721 p->pack_promisor = 1;
722
723 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
724 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
725 free(p);
726 return NULL;
727 }
728
729 /* ok, it looks sane as far as we can check without
730 * actually mapping the pack file.
731 */
732 p->pack_size = st.st_size;
733 p->pack_local = local;
734 p->mtime = st.st_mtime;
735 if (path_len < the_hash_algo->hexsz ||
736 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
737 hashclr(p->hash);
738 return p;
739 }
740
741 void install_packed_git(struct repository *r, struct packed_git *pack)
742 {
743 if (pack->pack_fd != -1)
744 pack_open_fds++;
745
746 pack->next = r->objects->packed_git;
747 r->objects->packed_git = pack;
748 }
749
750 void (*report_garbage)(unsigned seen_bits, const char *path);
751
752 static void report_helper(const struct string_list *list,
753 int seen_bits, int first, int last)
754 {
755 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
756 return;
757
758 for (; first < last; first++)
759 report_garbage(seen_bits, list->items[first].string);
760 }
761
762 static void report_pack_garbage(struct string_list *list)
763 {
764 int i, baselen = -1, first = 0, seen_bits = 0;
765
766 if (!report_garbage)
767 return;
768
769 string_list_sort(list);
770
771 for (i = 0; i < list->nr; i++) {
772 const char *path = list->items[i].string;
773 if (baselen != -1 &&
774 strncmp(path, list->items[first].string, baselen)) {
775 report_helper(list, seen_bits, first, i);
776 baselen = -1;
777 seen_bits = 0;
778 }
779 if (baselen == -1) {
780 const char *dot = strrchr(path, '.');
781 if (!dot) {
782 report_garbage(PACKDIR_FILE_GARBAGE, path);
783 continue;
784 }
785 baselen = dot - path + 1;
786 first = i;
787 }
788 if (!strcmp(path + baselen, "pack"))
789 seen_bits |= 1;
790 else if (!strcmp(path + baselen, "idx"))
791 seen_bits |= 2;
792 }
793 report_helper(list, seen_bits, first, list->nr);
794 }
795
796 void for_each_file_in_pack_dir(const char *objdir,
797 each_file_in_pack_dir_fn fn,
798 void *data)
799 {
800 struct strbuf path = STRBUF_INIT;
801 size_t dirnamelen;
802 DIR *dir;
803 struct dirent *de;
804
805 strbuf_addstr(&path, objdir);
806 strbuf_addstr(&path, "/pack");
807 dir = opendir(path.buf);
808 if (!dir) {
809 if (errno != ENOENT)
810 error_errno("unable to open object pack directory: %s",
811 path.buf);
812 strbuf_release(&path);
813 return;
814 }
815 strbuf_addch(&path, '/');
816 dirnamelen = path.len;
817 while ((de = readdir(dir)) != NULL) {
818 if (is_dot_or_dotdot(de->d_name))
819 continue;
820
821 strbuf_setlen(&path, dirnamelen);
822 strbuf_addstr(&path, de->d_name);
823
824 fn(path.buf, path.len, de->d_name, data);
825 }
826
827 closedir(dir);
828 strbuf_release(&path);
829 }
830
831 struct prepare_pack_data {
832 struct repository *r;
833 struct string_list *garbage;
834 int local;
835 struct multi_pack_index *m;
836 };
837
838 static void prepare_pack(const char *full_name, size_t full_name_len,
839 const char *file_name, void *_data)
840 {
841 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
842 struct packed_git *p;
843 size_t base_len = full_name_len;
844
845 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
846 !(data->m && midx_contains_pack(data->m, file_name))) {
847 /* Don't reopen a pack we already have. */
848 for (p = data->r->objects->packed_git; p; p = p->next) {
849 size_t len;
850 if (strip_suffix(p->pack_name, ".pack", &len) &&
851 len == base_len &&
852 !memcmp(p->pack_name, full_name, len))
853 break;
854 }
855
856 if (!p) {
857 p = add_packed_git(full_name, full_name_len, data->local);
858 if (p)
859 install_packed_git(data->r, p);
860 }
861 }
862
863 if (!report_garbage)
864 return;
865
866 if (!strcmp(file_name, "multi-pack-index"))
867 return;
868 if (ends_with(file_name, ".idx") ||
869 ends_with(file_name, ".pack") ||
870 ends_with(file_name, ".bitmap") ||
871 ends_with(file_name, ".keep") ||
872 ends_with(file_name, ".promisor"))
873 string_list_append(data->garbage, full_name);
874 else
875 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
876 }
877
878 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
879 {
880 struct prepare_pack_data data;
881 struct string_list garbage = STRING_LIST_INIT_DUP;
882
883 data.m = r->objects->multi_pack_index;
884
885 /* look for the multi-pack-index for this object directory */
886 while (data.m && strcmp(data.m->object_dir, objdir))
887 data.m = data.m->next;
888
889 data.r = r;
890 data.garbage = &garbage;
891 data.local = local;
892
893 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
894
895 report_pack_garbage(data.garbage);
896 string_list_clear(data.garbage, 0);
897 }
898
899 static void prepare_packed_git(struct repository *r);
900 /*
901 * Give a fast, rough count of the number of objects in the repository. This
902 * ignores loose objects completely. If you have a lot of them, then either
903 * you should repack because your performance will be awful, or they are
904 * all unreachable objects about to be pruned, in which case they're not really
905 * interesting as a measure of repo size in the first place.
906 */
907 unsigned long repo_approximate_object_count(struct repository *r)
908 {
909 if (!r->objects->approximate_object_count_valid) {
910 unsigned long count;
911 struct multi_pack_index *m;
912 struct packed_git *p;
913
914 prepare_packed_git(r);
915 count = 0;
916 for (m = get_multi_pack_index(r); m; m = m->next)
917 count += m->num_objects;
918 for (p = r->objects->packed_git; p; p = p->next) {
919 if (open_pack_index(p))
920 continue;
921 count += p->num_objects;
922 }
923 r->objects->approximate_object_count = count;
924 }
925 return r->objects->approximate_object_count;
926 }
927
928 static void *get_next_packed_git(const void *p)
929 {
930 return ((const struct packed_git *)p)->next;
931 }
932
933 static void set_next_packed_git(void *p, void *next)
934 {
935 ((struct packed_git *)p)->next = next;
936 }
937
938 static int sort_pack(const void *a_, const void *b_)
939 {
940 const struct packed_git *a = a_;
941 const struct packed_git *b = b_;
942 int st;
943
944 /*
945 * Local packs tend to contain objects specific to our
946 * variant of the project than remote ones. In addition,
947 * remote ones could be on a network mounted filesystem.
948 * Favor local ones for these reasons.
949 */
950 st = a->pack_local - b->pack_local;
951 if (st)
952 return -st;
953
954 /*
955 * Younger packs tend to contain more recent objects,
956 * and more recent objects tend to get accessed more
957 * often.
958 */
959 if (a->mtime < b->mtime)
960 return 1;
961 else if (a->mtime == b->mtime)
962 return 0;
963 return -1;
964 }
965
966 static void rearrange_packed_git(struct repository *r)
967 {
968 r->objects->packed_git = llist_mergesort(
969 r->objects->packed_git, get_next_packed_git,
970 set_next_packed_git, sort_pack);
971 }
972
973 static void prepare_packed_git_mru(struct repository *r)
974 {
975 struct packed_git *p;
976
977 INIT_LIST_HEAD(&r->objects->packed_git_mru);
978
979 for (p = r->objects->packed_git; p; p = p->next)
980 list_add_tail(&p->mru, &r->objects->packed_git_mru);
981 }
982
983 static void prepare_packed_git(struct repository *r)
984 {
985 struct object_directory *odb;
986
987 if (r->objects->packed_git_initialized)
988 return;
989
990 prepare_alt_odb(r);
991 for (odb = r->objects->odb; odb; odb = odb->next) {
992 int local = (odb == r->objects->odb);
993 prepare_multi_pack_index_one(r, odb->path, local);
994 prepare_packed_git_one(r, odb->path, local);
995 }
996 rearrange_packed_git(r);
997
998 prepare_packed_git_mru(r);
999 r->objects->packed_git_initialized = 1;
1000 }
1001
1002 void reprepare_packed_git(struct repository *r)
1003 {
1004 struct object_directory *odb;
1005
1006 for (odb = r->objects->odb; odb; odb = odb->next)
1007 odb_clear_loose_cache(odb);
1008
1009 r->objects->approximate_object_count_valid = 0;
1010 r->objects->packed_git_initialized = 0;
1011 prepare_packed_git(r);
1012 }
1013
1014 struct packed_git *get_packed_git(struct repository *r)
1015 {
1016 prepare_packed_git(r);
1017 return r->objects->packed_git;
1018 }
1019
1020 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1021 {
1022 prepare_packed_git(r);
1023 return r->objects->multi_pack_index;
1024 }
1025
1026 struct packed_git *get_all_packs(struct repository *r)
1027 {
1028 struct multi_pack_index *m;
1029
1030 prepare_packed_git(r);
1031 for (m = r->objects->multi_pack_index; m; m = m->next) {
1032 uint32_t i;
1033 for (i = 0; i < m->num_packs; i++)
1034 prepare_midx_pack(r, m, i);
1035 }
1036
1037 return r->objects->packed_git;
1038 }
1039
1040 struct list_head *get_packed_git_mru(struct repository *r)
1041 {
1042 prepare_packed_git(r);
1043 return &r->objects->packed_git_mru;
1044 }
1045
1046 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1047 unsigned long len, enum object_type *type, unsigned long *sizep)
1048 {
1049 unsigned shift;
1050 unsigned long size, c;
1051 unsigned long used = 0;
1052
1053 c = buf[used++];
1054 *type = (c >> 4) & 7;
1055 size = c & 15;
1056 shift = 4;
1057 while (c & 0x80) {
1058 if (len <= used || bitsizeof(long) <= shift) {
1059 error("bad object header");
1060 size = used = 0;
1061 break;
1062 }
1063 c = buf[used++];
1064 size += (c & 0x7f) << shift;
1065 shift += 7;
1066 }
1067 *sizep = size;
1068 return used;
1069 }
1070
1071 unsigned long get_size_from_delta(struct packed_git *p,
1072 struct pack_window **w_curs,
1073 off_t curpos)
1074 {
1075 const unsigned char *data;
1076 unsigned char delta_head[20], *in;
1077 git_zstream stream;
1078 int st;
1079
1080 memset(&stream, 0, sizeof(stream));
1081 stream.next_out = delta_head;
1082 stream.avail_out = sizeof(delta_head);
1083
1084 git_inflate_init(&stream);
1085 do {
1086 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1087 stream.next_in = in;
1088 st = git_inflate(&stream, Z_FINISH);
1089 curpos += stream.next_in - in;
1090 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1091 stream.total_out < sizeof(delta_head));
1092 git_inflate_end(&stream);
1093 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1094 error("delta data unpack-initial failed");
1095 return 0;
1096 }
1097
1098 /* Examine the initial part of the delta to figure out
1099 * the result size.
1100 */
1101 data = delta_head;
1102
1103 /* ignore base size */
1104 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1105
1106 /* Read the result size */
1107 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1108 }
1109
1110 int unpack_object_header(struct packed_git *p,
1111 struct pack_window **w_curs,
1112 off_t *curpos,
1113 unsigned long *sizep)
1114 {
1115 unsigned char *base;
1116 unsigned long left;
1117 unsigned long used;
1118 enum object_type type;
1119
1120 /* use_pack() assures us we have [base, base + 20) available
1121 * as a range that we can look at. (Its actually the hash
1122 * size that is assured.) With our object header encoding
1123 * the maximum deflated object size is 2^137, which is just
1124 * insane, so we know won't exceed what we have been given.
1125 */
1126 base = use_pack(p, w_curs, *curpos, &left);
1127 used = unpack_object_header_buffer(base, left, &type, sizep);
1128 if (!used) {
1129 type = OBJ_BAD;
1130 } else
1131 *curpos += used;
1132
1133 return type;
1134 }
1135
1136 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1137 {
1138 unsigned i;
1139 const unsigned hashsz = the_hash_algo->rawsz;
1140 for (i = 0; i < p->num_bad_objects; i++)
1141 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1142 return;
1143 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1144 st_mult(GIT_MAX_RAWSZ,
1145 st_add(p->num_bad_objects, 1)));
1146 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1147 p->num_bad_objects++;
1148 }
1149
1150 const struct packed_git *has_packed_and_bad(struct repository *r,
1151 const unsigned char *sha1)
1152 {
1153 struct packed_git *p;
1154 unsigned i;
1155
1156 for (p = r->objects->packed_git; p; p = p->next)
1157 for (i = 0; i < p->num_bad_objects; i++)
1158 if (hasheq(sha1,
1159 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1160 return p;
1161 return NULL;
1162 }
1163
1164 static off_t get_delta_base(struct packed_git *p,
1165 struct pack_window **w_curs,
1166 off_t *curpos,
1167 enum object_type type,
1168 off_t delta_obj_offset)
1169 {
1170 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1171 off_t base_offset;
1172
1173 /* use_pack() assured us we have [base_info, base_info + 20)
1174 * as a range that we can look at without walking off the
1175 * end of the mapped window. Its actually the hash size
1176 * that is assured. An OFS_DELTA longer than the hash size
1177 * is stupid, as then a REF_DELTA would be smaller to store.
1178 */
1179 if (type == OBJ_OFS_DELTA) {
1180 unsigned used = 0;
1181 unsigned char c = base_info[used++];
1182 base_offset = c & 127;
1183 while (c & 128) {
1184 base_offset += 1;
1185 if (!base_offset || MSB(base_offset, 7))
1186 return 0; /* overflow */
1187 c = base_info[used++];
1188 base_offset = (base_offset << 7) + (c & 127);
1189 }
1190 base_offset = delta_obj_offset - base_offset;
1191 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1192 return 0; /* out of bound */
1193 *curpos += used;
1194 } else if (type == OBJ_REF_DELTA) {
1195 /* The base entry _must_ be in the same pack */
1196 base_offset = find_pack_entry_one(base_info, p);
1197 *curpos += the_hash_algo->rawsz;
1198 } else
1199 die("I am totally screwed");
1200 return base_offset;
1201 }
1202
1203 /*
1204 * Like get_delta_base above, but we return the sha1 instead of the pack
1205 * offset. This means it is cheaper for REF deltas (we do not have to do
1206 * the final object lookup), but more expensive for OFS deltas (we
1207 * have to load the revidx to convert the offset back into a sha1).
1208 */
1209 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1210 struct pack_window **w_curs,
1211 off_t curpos,
1212 enum object_type type,
1213 off_t delta_obj_offset)
1214 {
1215 if (type == OBJ_REF_DELTA) {
1216 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1217 return base;
1218 } else if (type == OBJ_OFS_DELTA) {
1219 struct revindex_entry *revidx;
1220 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1221 type, delta_obj_offset);
1222
1223 if (!base_offset)
1224 return NULL;
1225
1226 revidx = find_pack_revindex(p, base_offset);
1227 if (!revidx)
1228 return NULL;
1229
1230 return nth_packed_object_sha1(p, revidx->nr);
1231 } else
1232 return NULL;
1233 }
1234
1235 static int retry_bad_packed_offset(struct repository *r,
1236 struct packed_git *p,
1237 off_t obj_offset)
1238 {
1239 int type;
1240 struct revindex_entry *revidx;
1241 struct object_id oid;
1242 revidx = find_pack_revindex(p, obj_offset);
1243 if (!revidx)
1244 return OBJ_BAD;
1245 nth_packed_object_oid(&oid, p, revidx->nr);
1246 mark_bad_packed_object(p, oid.hash);
1247 type = oid_object_info(r, &oid, NULL);
1248 if (type <= OBJ_NONE)
1249 return OBJ_BAD;
1250 return type;
1251 }
1252
1253 #define POI_STACK_PREALLOC 64
1254
1255 static enum object_type packed_to_object_type(struct repository *r,
1256 struct packed_git *p,
1257 off_t obj_offset,
1258 enum object_type type,
1259 struct pack_window **w_curs,
1260 off_t curpos)
1261 {
1262 off_t small_poi_stack[POI_STACK_PREALLOC];
1263 off_t *poi_stack = small_poi_stack;
1264 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1265
1266 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1267 off_t base_offset;
1268 unsigned long size;
1269 /* Push the object we're going to leave behind */
1270 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1271 poi_stack_alloc = alloc_nr(poi_stack_nr);
1272 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1273 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1274 } else {
1275 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1276 }
1277 poi_stack[poi_stack_nr++] = obj_offset;
1278 /* If parsing the base offset fails, just unwind */
1279 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1280 if (!base_offset)
1281 goto unwind;
1282 curpos = obj_offset = base_offset;
1283 type = unpack_object_header(p, w_curs, &curpos, &size);
1284 if (type <= OBJ_NONE) {
1285 /* If getting the base itself fails, we first
1286 * retry the base, otherwise unwind */
1287 type = retry_bad_packed_offset(r, p, base_offset);
1288 if (type > OBJ_NONE)
1289 goto out;
1290 goto unwind;
1291 }
1292 }
1293
1294 switch (type) {
1295 case OBJ_BAD:
1296 case OBJ_COMMIT:
1297 case OBJ_TREE:
1298 case OBJ_BLOB:
1299 case OBJ_TAG:
1300 break;
1301 default:
1302 error("unknown object type %i at offset %"PRIuMAX" in %s",
1303 type, (uintmax_t)obj_offset, p->pack_name);
1304 type = OBJ_BAD;
1305 }
1306
1307 out:
1308 if (poi_stack != small_poi_stack)
1309 free(poi_stack);
1310 return type;
1311
1312 unwind:
1313 while (poi_stack_nr) {
1314 obj_offset = poi_stack[--poi_stack_nr];
1315 type = retry_bad_packed_offset(r, p, obj_offset);
1316 if (type > OBJ_NONE)
1317 goto out;
1318 }
1319 type = OBJ_BAD;
1320 goto out;
1321 }
1322
1323 static struct hashmap delta_base_cache;
1324 static size_t delta_base_cached;
1325
1326 static LIST_HEAD(delta_base_cache_lru);
1327
1328 struct delta_base_cache_key {
1329 struct packed_git *p;
1330 off_t base_offset;
1331 };
1332
1333 struct delta_base_cache_entry {
1334 struct hashmap hash;
1335 struct delta_base_cache_key key;
1336 struct list_head lru;
1337 void *data;
1338 unsigned long size;
1339 enum object_type type;
1340 };
1341
1342 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1343 {
1344 unsigned int hash;
1345
1346 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1347 hash += (hash >> 8) + (hash >> 16);
1348 return hash;
1349 }
1350
1351 static struct delta_base_cache_entry *
1352 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1353 {
1354 struct hashmap_entry entry;
1355 struct delta_base_cache_key key;
1356
1357 if (!delta_base_cache.cmpfn)
1358 return NULL;
1359
1360 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1361 key.p = p;
1362 key.base_offset = base_offset;
1363 return hashmap_get(&delta_base_cache, &entry, &key);
1364 }
1365
1366 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1367 const struct delta_base_cache_key *b)
1368 {
1369 return a->p == b->p && a->base_offset == b->base_offset;
1370 }
1371
1372 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1373 const void *va, const void *vb,
1374 const void *vkey)
1375 {
1376 const struct delta_base_cache_entry *a = va, *b = vb;
1377 const struct delta_base_cache_key *key = vkey;
1378 if (key)
1379 return !delta_base_cache_key_eq(&a->key, key);
1380 else
1381 return !delta_base_cache_key_eq(&a->key, &b->key);
1382 }
1383
1384 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1385 {
1386 return !!get_delta_base_cache_entry(p, base_offset);
1387 }
1388
1389 /*
1390 * Remove the entry from the cache, but do _not_ free the associated
1391 * entry data. The caller takes ownership of the "data" buffer, and
1392 * should copy out any fields it wants before detaching.
1393 */
1394 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1395 {
1396 hashmap_remove(&delta_base_cache, ent, &ent->key);
1397 list_del(&ent->lru);
1398 delta_base_cached -= ent->size;
1399 free(ent);
1400 }
1401
1402 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1403 off_t base_offset, unsigned long *base_size,
1404 enum object_type *type)
1405 {
1406 struct delta_base_cache_entry *ent;
1407
1408 ent = get_delta_base_cache_entry(p, base_offset);
1409 if (!ent)
1410 return unpack_entry(r, p, base_offset, type, base_size);
1411
1412 if (type)
1413 *type = ent->type;
1414 if (base_size)
1415 *base_size = ent->size;
1416 return xmemdupz(ent->data, ent->size);
1417 }
1418
1419 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1420 {
1421 free(ent->data);
1422 detach_delta_base_cache_entry(ent);
1423 }
1424
1425 void clear_delta_base_cache(void)
1426 {
1427 struct list_head *lru, *tmp;
1428 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1429 struct delta_base_cache_entry *entry =
1430 list_entry(lru, struct delta_base_cache_entry, lru);
1431 release_delta_base_cache(entry);
1432 }
1433 }
1434
1435 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1436 void *base, unsigned long base_size, enum object_type type)
1437 {
1438 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1439 struct list_head *lru, *tmp;
1440
1441 delta_base_cached += base_size;
1442
1443 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1444 struct delta_base_cache_entry *f =
1445 list_entry(lru, struct delta_base_cache_entry, lru);
1446 if (delta_base_cached <= delta_base_cache_limit)
1447 break;
1448 release_delta_base_cache(f);
1449 }
1450
1451 ent->key.p = p;
1452 ent->key.base_offset = base_offset;
1453 ent->type = type;
1454 ent->data = base;
1455 ent->size = base_size;
1456 list_add_tail(&ent->lru, &delta_base_cache_lru);
1457
1458 if (!delta_base_cache.cmpfn)
1459 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1460 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1461 hashmap_add(&delta_base_cache, ent);
1462 }
1463
1464 int packed_object_info(struct repository *r, struct packed_git *p,
1465 off_t obj_offset, struct object_info *oi)
1466 {
1467 struct pack_window *w_curs = NULL;
1468 unsigned long size;
1469 off_t curpos = obj_offset;
1470 enum object_type type;
1471
1472 /*
1473 * We always get the representation type, but only convert it to
1474 * a "real" type later if the caller is interested.
1475 */
1476 if (oi->contentp) {
1477 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1478 &type);
1479 if (!*oi->contentp)
1480 type = OBJ_BAD;
1481 } else {
1482 type = unpack_object_header(p, &w_curs, &curpos, &size);
1483 }
1484
1485 if (!oi->contentp && oi->sizep) {
1486 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1487 off_t tmp_pos = curpos;
1488 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1489 type, obj_offset);
1490 if (!base_offset) {
1491 type = OBJ_BAD;
1492 goto out;
1493 }
1494 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1495 if (*oi->sizep == 0) {
1496 type = OBJ_BAD;
1497 goto out;
1498 }
1499 } else {
1500 *oi->sizep = size;
1501 }
1502 }
1503
1504 if (oi->disk_sizep) {
1505 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1506 *oi->disk_sizep = revidx[1].offset - obj_offset;
1507 }
1508
1509 if (oi->typep || oi->type_name) {
1510 enum object_type ptot;
1511 ptot = packed_to_object_type(r, p, obj_offset,
1512 type, &w_curs, curpos);
1513 if (oi->typep)
1514 *oi->typep = ptot;
1515 if (oi->type_name) {
1516 const char *tn = type_name(ptot);
1517 if (tn)
1518 strbuf_addstr(oi->type_name, tn);
1519 }
1520 if (ptot < 0) {
1521 type = OBJ_BAD;
1522 goto out;
1523 }
1524 }
1525
1526 if (oi->delta_base_sha1) {
1527 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1528 const unsigned char *base;
1529
1530 base = get_delta_base_sha1(p, &w_curs, curpos,
1531 type, obj_offset);
1532 if (!base) {
1533 type = OBJ_BAD;
1534 goto out;
1535 }
1536
1537 hashcpy(oi->delta_base_sha1, base);
1538 } else
1539 hashclr(oi->delta_base_sha1);
1540 }
1541
1542 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1543 OI_PACKED;
1544
1545 out:
1546 unuse_pack(&w_curs);
1547 return type;
1548 }
1549
1550 static void *unpack_compressed_entry(struct packed_git *p,
1551 struct pack_window **w_curs,
1552 off_t curpos,
1553 unsigned long size)
1554 {
1555 int st;
1556 git_zstream stream;
1557 unsigned char *buffer, *in;
1558
1559 buffer = xmallocz_gently(size);
1560 if (!buffer)
1561 return NULL;
1562 memset(&stream, 0, sizeof(stream));
1563 stream.next_out = buffer;
1564 stream.avail_out = size + 1;
1565
1566 git_inflate_init(&stream);
1567 do {
1568 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1569 stream.next_in = in;
1570 st = git_inflate(&stream, Z_FINISH);
1571 if (!stream.avail_out)
1572 break; /* the payload is larger than it should be */
1573 curpos += stream.next_in - in;
1574 } while (st == Z_OK || st == Z_BUF_ERROR);
1575 git_inflate_end(&stream);
1576 if ((st != Z_STREAM_END) || stream.total_out != size) {
1577 free(buffer);
1578 return NULL;
1579 }
1580
1581 /* versions of zlib can clobber unconsumed portion of outbuf */
1582 buffer[size] = '\0';
1583
1584 return buffer;
1585 }
1586
1587 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1588 {
1589 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1590 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1591 p->pack_name, (uintmax_t)obj_offset);
1592 }
1593
1594 int do_check_packed_object_crc;
1595
1596 #define UNPACK_ENTRY_STACK_PREALLOC 64
1597 struct unpack_entry_stack_ent {
1598 off_t obj_offset;
1599 off_t curpos;
1600 unsigned long size;
1601 };
1602
1603 static void *read_object(struct repository *r,
1604 const struct object_id *oid,
1605 enum object_type *type,
1606 unsigned long *size)
1607 {
1608 struct object_info oi = OBJECT_INFO_INIT;
1609 void *content;
1610 oi.typep = type;
1611 oi.sizep = size;
1612 oi.contentp = &content;
1613
1614 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1615 return NULL;
1616 return content;
1617 }
1618
1619 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1620 enum object_type *final_type, unsigned long *final_size)
1621 {
1622 struct pack_window *w_curs = NULL;
1623 off_t curpos = obj_offset;
1624 void *data = NULL;
1625 unsigned long size;
1626 enum object_type type;
1627 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1628 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1629 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1630 int base_from_cache = 0;
1631
1632 write_pack_access_log(p, obj_offset);
1633
1634 /* PHASE 1: drill down to the innermost base object */
1635 for (;;) {
1636 off_t base_offset;
1637 int i;
1638 struct delta_base_cache_entry *ent;
1639
1640 ent = get_delta_base_cache_entry(p, curpos);
1641 if (ent) {
1642 type = ent->type;
1643 data = ent->data;
1644 size = ent->size;
1645 detach_delta_base_cache_entry(ent);
1646 base_from_cache = 1;
1647 break;
1648 }
1649
1650 if (do_check_packed_object_crc && p->index_version > 1) {
1651 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1652 off_t len = revidx[1].offset - obj_offset;
1653 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1654 struct object_id oid;
1655 nth_packed_object_oid(&oid, p, revidx->nr);
1656 error("bad packed object CRC for %s",
1657 oid_to_hex(&oid));
1658 mark_bad_packed_object(p, oid.hash);
1659 data = NULL;
1660 goto out;
1661 }
1662 }
1663
1664 type = unpack_object_header(p, &w_curs, &curpos, &size);
1665 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1666 break;
1667
1668 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1669 if (!base_offset) {
1670 error("failed to validate delta base reference "
1671 "at offset %"PRIuMAX" from %s",
1672 (uintmax_t)curpos, p->pack_name);
1673 /* bail to phase 2, in hopes of recovery */
1674 data = NULL;
1675 break;
1676 }
1677
1678 /* push object, proceed to base */
1679 if (delta_stack_nr >= delta_stack_alloc
1680 && delta_stack == small_delta_stack) {
1681 delta_stack_alloc = alloc_nr(delta_stack_nr);
1682 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1683 memcpy(delta_stack, small_delta_stack,
1684 sizeof(*delta_stack)*delta_stack_nr);
1685 } else {
1686 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1687 }
1688 i = delta_stack_nr++;
1689 delta_stack[i].obj_offset = obj_offset;
1690 delta_stack[i].curpos = curpos;
1691 delta_stack[i].size = size;
1692
1693 curpos = obj_offset = base_offset;
1694 }
1695
1696 /* PHASE 2: handle the base */
1697 switch (type) {
1698 case OBJ_OFS_DELTA:
1699 case OBJ_REF_DELTA:
1700 if (data)
1701 BUG("unpack_entry: left loop at a valid delta");
1702 break;
1703 case OBJ_COMMIT:
1704 case OBJ_TREE:
1705 case OBJ_BLOB:
1706 case OBJ_TAG:
1707 if (!base_from_cache)
1708 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1709 break;
1710 default:
1711 data = NULL;
1712 error("unknown object type %i at offset %"PRIuMAX" in %s",
1713 type, (uintmax_t)obj_offset, p->pack_name);
1714 }
1715
1716 /* PHASE 3: apply deltas in order */
1717
1718 /* invariants:
1719 * 'data' holds the base data, or NULL if there was corruption
1720 */
1721 while (delta_stack_nr) {
1722 void *delta_data;
1723 void *base = data;
1724 void *external_base = NULL;
1725 unsigned long delta_size, base_size = size;
1726 int i;
1727
1728 data = NULL;
1729
1730 if (base)
1731 add_delta_base_cache(p, obj_offset, base, base_size, type);
1732
1733 if (!base) {
1734 /*
1735 * We're probably in deep shit, but let's try to fetch
1736 * the required base anyway from another pack or loose.
1737 * This is costly but should happen only in the presence
1738 * of a corrupted pack, and is better than failing outright.
1739 */
1740 struct revindex_entry *revidx;
1741 struct object_id base_oid;
1742 revidx = find_pack_revindex(p, obj_offset);
1743 if (revidx) {
1744 nth_packed_object_oid(&base_oid, p, revidx->nr);
1745 error("failed to read delta base object %s"
1746 " at offset %"PRIuMAX" from %s",
1747 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1748 p->pack_name);
1749 mark_bad_packed_object(p, base_oid.hash);
1750 base = read_object(r, &base_oid, &type, &base_size);
1751 external_base = base;
1752 }
1753 }
1754
1755 i = --delta_stack_nr;
1756 obj_offset = delta_stack[i].obj_offset;
1757 curpos = delta_stack[i].curpos;
1758 delta_size = delta_stack[i].size;
1759
1760 if (!base)
1761 continue;
1762
1763 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1764
1765 if (!delta_data) {
1766 error("failed to unpack compressed delta "
1767 "at offset %"PRIuMAX" from %s",
1768 (uintmax_t)curpos, p->pack_name);
1769 data = NULL;
1770 free(external_base);
1771 continue;
1772 }
1773
1774 data = patch_delta(base, base_size,
1775 delta_data, delta_size,
1776 &size);
1777
1778 /*
1779 * We could not apply the delta; warn the user, but keep going.
1780 * Our failure will be noticed either in the next iteration of
1781 * the loop, or if this is the final delta, in the caller when
1782 * we return NULL. Those code paths will take care of making
1783 * a more explicit warning and retrying with another copy of
1784 * the object.
1785 */
1786 if (!data)
1787 error("failed to apply delta");
1788
1789 free(delta_data);
1790 free(external_base);
1791 }
1792
1793 if (final_type)
1794 *final_type = type;
1795 if (final_size)
1796 *final_size = size;
1797
1798 out:
1799 unuse_pack(&w_curs);
1800
1801 if (delta_stack != small_delta_stack)
1802 free(delta_stack);
1803
1804 return data;
1805 }
1806
1807 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1808 {
1809 const unsigned char *index_fanout = p->index_data;
1810 const unsigned char *index_lookup;
1811 const unsigned int hashsz = the_hash_algo->rawsz;
1812 int index_lookup_width;
1813
1814 if (!index_fanout)
1815 BUG("bsearch_pack called without a valid pack-index");
1816
1817 index_lookup = index_fanout + 4 * 256;
1818 if (p->index_version == 1) {
1819 index_lookup_width = hashsz + 4;
1820 index_lookup += 4;
1821 } else {
1822 index_lookup_width = hashsz;
1823 index_fanout += 8;
1824 index_lookup += 8;
1825 }
1826
1827 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1828 index_lookup, index_lookup_width, result);
1829 }
1830
1831 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1832 uint32_t n)
1833 {
1834 const unsigned char *index = p->index_data;
1835 const unsigned int hashsz = the_hash_algo->rawsz;
1836 if (!index) {
1837 if (open_pack_index(p))
1838 return NULL;
1839 index = p->index_data;
1840 }
1841 if (n >= p->num_objects)
1842 return NULL;
1843 index += 4 * 256;
1844 if (p->index_version == 1) {
1845 return index + (hashsz + 4) * n + 4;
1846 } else {
1847 index += 8;
1848 return index + hashsz * n;
1849 }
1850 }
1851
1852 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1853 struct packed_git *p,
1854 uint32_t n)
1855 {
1856 const unsigned char *hash = nth_packed_object_sha1(p, n);
1857 if (!hash)
1858 return NULL;
1859 hashcpy(oid->hash, hash);
1860 return oid;
1861 }
1862
1863 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1864 {
1865 const unsigned char *ptr = vptr;
1866 const unsigned char *start = p->index_data;
1867 const unsigned char *end = start + p->index_size;
1868 if (ptr < start)
1869 die(_("offset before start of pack index for %s (corrupt index?)"),
1870 p->pack_name);
1871 /* No need to check for underflow; .idx files must be at least 8 bytes */
1872 if (ptr >= end - 8)
1873 die(_("offset beyond end of pack index for %s (truncated index?)"),
1874 p->pack_name);
1875 }
1876
1877 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1878 {
1879 const unsigned char *index = p->index_data;
1880 const unsigned int hashsz = the_hash_algo->rawsz;
1881 index += 4 * 256;
1882 if (p->index_version == 1) {
1883 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1884 } else {
1885 uint32_t off;
1886 index += 8 + p->num_objects * (hashsz + 4);
1887 off = ntohl(*((uint32_t *)(index + 4 * n)));
1888 if (!(off & 0x80000000))
1889 return off;
1890 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1891 check_pack_index_ptr(p, index);
1892 return get_be64(index);
1893 }
1894 }
1895
1896 off_t find_pack_entry_one(const unsigned char *sha1,
1897 struct packed_git *p)
1898 {
1899 const unsigned char *index = p->index_data;
1900 struct object_id oid;
1901 uint32_t result;
1902
1903 if (!index) {
1904 if (open_pack_index(p))
1905 return 0;
1906 }
1907
1908 hashcpy(oid.hash, sha1);
1909 if (bsearch_pack(&oid, p, &result))
1910 return nth_packed_object_offset(p, result);
1911 return 0;
1912 }
1913
1914 int is_pack_valid(struct packed_git *p)
1915 {
1916 /* An already open pack is known to be valid. */
1917 if (p->pack_fd != -1)
1918 return 1;
1919
1920 /* If the pack has one window completely covering the
1921 * file size, the pack is known to be valid even if
1922 * the descriptor is not currently open.
1923 */
1924 if (p->windows) {
1925 struct pack_window *w = p->windows;
1926
1927 if (!w->offset && w->len == p->pack_size)
1928 return 1;
1929 }
1930
1931 /* Force the pack to open to prove its valid. */
1932 return !open_packed_git(p);
1933 }
1934
1935 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1936 struct packed_git *packs)
1937 {
1938 struct packed_git *p;
1939
1940 for (p = packs; p; p = p->next) {
1941 if (find_pack_entry_one(sha1, p))
1942 return p;
1943 }
1944 return NULL;
1945
1946 }
1947
1948 static int fill_pack_entry(const struct object_id *oid,
1949 struct pack_entry *e,
1950 struct packed_git *p)
1951 {
1952 off_t offset;
1953
1954 if (p->num_bad_objects) {
1955 unsigned i;
1956 for (i = 0; i < p->num_bad_objects; i++)
1957 if (hasheq(oid->hash,
1958 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1959 return 0;
1960 }
1961
1962 offset = find_pack_entry_one(oid->hash, p);
1963 if (!offset)
1964 return 0;
1965
1966 /*
1967 * We are about to tell the caller where they can locate the
1968 * requested object. We better make sure the packfile is
1969 * still here and can be accessed before supplying that
1970 * answer, as it may have been deleted since the index was
1971 * loaded!
1972 */
1973 if (!is_pack_valid(p))
1974 return 0;
1975 e->offset = offset;
1976 e->p = p;
1977 return 1;
1978 }
1979
1980 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1981 {
1982 struct list_head *pos;
1983 struct multi_pack_index *m;
1984
1985 prepare_packed_git(r);
1986 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1987 return 0;
1988
1989 for (m = r->objects->multi_pack_index; m; m = m->next) {
1990 if (fill_midx_entry(r, oid, e, m))
1991 return 1;
1992 }
1993
1994 list_for_each(pos, &r->objects->packed_git_mru) {
1995 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1996 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
1997 list_move(&p->mru, &r->objects->packed_git_mru);
1998 return 1;
1999 }
2000 }
2001 return 0;
2002 }
2003
2004 int has_object_pack(const struct object_id *oid)
2005 {
2006 struct pack_entry e;
2007 return find_pack_entry(the_repository, oid, &e);
2008 }
2009
2010 int has_pack_index(const unsigned char *sha1)
2011 {
2012 struct stat st;
2013 if (stat(sha1_pack_index_name(sha1), &st))
2014 return 0;
2015 return 1;
2016 }
2017
2018 int for_each_object_in_pack(struct packed_git *p,
2019 each_packed_object_fn cb, void *data,
2020 enum for_each_object_flags flags)
2021 {
2022 uint32_t i;
2023 int r = 0;
2024
2025 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2026 if (load_pack_revindex(p))
2027 return -1;
2028 }
2029
2030 for (i = 0; i < p->num_objects; i++) {
2031 uint32_t pos;
2032 struct object_id oid;
2033
2034 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2035 pos = p->revindex[i].nr;
2036 else
2037 pos = i;
2038
2039 if (!nth_packed_object_oid(&oid, p, pos))
2040 return error("unable to get sha1 of object %u in %s",
2041 pos, p->pack_name);
2042
2043 r = cb(&oid, p, pos, data);
2044 if (r)
2045 break;
2046 }
2047 return r;
2048 }
2049
2050 int for_each_packed_object(each_packed_object_fn cb, void *data,
2051 enum for_each_object_flags flags)
2052 {
2053 struct packed_git *p;
2054 int r = 0;
2055 int pack_errors = 0;
2056
2057 prepare_packed_git(the_repository);
2058 for (p = get_all_packs(the_repository); p; p = p->next) {
2059 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2060 continue;
2061 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2062 !p->pack_promisor)
2063 continue;
2064 if (open_pack_index(p)) {
2065 pack_errors = 1;
2066 continue;
2067 }
2068 r = for_each_object_in_pack(p, cb, data, flags);
2069 if (r)
2070 break;
2071 }
2072 return r ? r : pack_errors;
2073 }
2074
2075 static int add_promisor_object(const struct object_id *oid,
2076 struct packed_git *pack,
2077 uint32_t pos,
2078 void *set_)
2079 {
2080 struct oidset *set = set_;
2081 struct object *obj = parse_object(the_repository, oid);
2082 if (!obj)
2083 return 1;
2084
2085 oidset_insert(set, oid);
2086
2087 /*
2088 * If this is a tree, commit, or tag, the objects it refers
2089 * to are also promisor objects. (Blobs refer to no objects->)
2090 */
2091 if (obj->type == OBJ_TREE) {
2092 struct tree *tree = (struct tree *)obj;
2093 struct tree_desc desc;
2094 struct name_entry entry;
2095 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2096 /*
2097 * Error messages are given when packs are
2098 * verified, so do not print any here.
2099 */
2100 return 0;
2101 while (tree_entry_gently(&desc, &entry))
2102 oidset_insert(set, &entry.oid);
2103 } else if (obj->type == OBJ_COMMIT) {
2104 struct commit *commit = (struct commit *) obj;
2105 struct commit_list *parents = commit->parents;
2106
2107 oidset_insert(set, get_commit_tree_oid(commit));
2108 for (; parents; parents = parents->next)
2109 oidset_insert(set, &parents->item->object.oid);
2110 } else if (obj->type == OBJ_TAG) {
2111 struct tag *tag = (struct tag *) obj;
2112 oidset_insert(set, &tag->tagged->oid);
2113 }
2114 return 0;
2115 }
2116
2117 int is_promisor_object(const struct object_id *oid)
2118 {
2119 static struct oidset promisor_objects;
2120 static int promisor_objects_prepared;
2121
2122 if (!promisor_objects_prepared) {
2123 if (has_promisor_remote()) {
2124 for_each_packed_object(add_promisor_object,
2125 &promisor_objects,
2126 FOR_EACH_OBJECT_PROMISOR_ONLY);
2127 }
2128 promisor_objects_prepared = 1;
2129 }
2130 return oidset_contains(&promisor_objects, oid);
2131 }