stash: make sure to write refreshed cache
[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 "commit-graph.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_object_store(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 close_commit_graph(o);
356 }
357
358 void unlink_pack_path(const char *pack_name, int force_delete)
359 {
360 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
361 int i;
362 struct strbuf buf = STRBUF_INIT;
363 size_t plen;
364
365 strbuf_addstr(&buf, pack_name);
366 strip_suffix_mem(buf.buf, &buf.len, ".pack");
367 plen = buf.len;
368
369 if (!force_delete) {
370 strbuf_addstr(&buf, ".keep");
371 if (!access(buf.buf, F_OK)) {
372 strbuf_release(&buf);
373 return;
374 }
375 }
376
377 for (i = 0; i < ARRAY_SIZE(exts); i++) {
378 strbuf_setlen(&buf, plen);
379 strbuf_addstr(&buf, exts[i]);
380 unlink(buf.buf);
381 }
382
383 strbuf_release(&buf);
384 }
385
386 /*
387 * The LRU pack is the one with the oldest MRU window, preferring packs
388 * with no used windows, or the oldest mtime if it has no windows allocated.
389 */
390 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
391 {
392 struct pack_window *w, *this_mru_w;
393 int has_windows_inuse = 0;
394
395 /*
396 * Reject this pack if it has windows and the previously selected
397 * one does not. If this pack does not have windows, reject
398 * it if the pack file is newer than the previously selected one.
399 */
400 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
401 return;
402
403 for (w = this_mru_w = p->windows; w; w = w->next) {
404 /*
405 * Reject this pack if any of its windows are in use,
406 * but the previously selected pack did not have any
407 * inuse windows. Otherwise, record that this pack
408 * has windows in use.
409 */
410 if (w->inuse_cnt) {
411 if (*accept_windows_inuse)
412 has_windows_inuse = 1;
413 else
414 return;
415 }
416
417 if (w->last_used > this_mru_w->last_used)
418 this_mru_w = w;
419
420 /*
421 * Reject this pack if it has windows that have been
422 * used more recently than the previously selected pack.
423 * If the previously selected pack had windows inuse and
424 * we have not encountered a window in this pack that is
425 * inuse, skip this check since we prefer a pack with no
426 * inuse windows to one that has inuse windows.
427 */
428 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
429 this_mru_w->last_used > (*mru_w)->last_used)
430 return;
431 }
432
433 /*
434 * Select this pack.
435 */
436 *mru_w = this_mru_w;
437 *lru_p = p;
438 *accept_windows_inuse = has_windows_inuse;
439 }
440
441 static int close_one_pack(void)
442 {
443 struct packed_git *p, *lru_p = NULL;
444 struct pack_window *mru_w = NULL;
445 int accept_windows_inuse = 1;
446
447 for (p = the_repository->objects->packed_git; p; p = p->next) {
448 if (p->pack_fd == -1)
449 continue;
450 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
451 }
452
453 if (lru_p)
454 return close_pack_fd(lru_p);
455
456 return 0;
457 }
458
459 static unsigned int get_max_fd_limit(void)
460 {
461 #ifdef RLIMIT_NOFILE
462 {
463 struct rlimit lim;
464
465 if (!getrlimit(RLIMIT_NOFILE, &lim))
466 return lim.rlim_cur;
467 }
468 #endif
469
470 #ifdef _SC_OPEN_MAX
471 {
472 long open_max = sysconf(_SC_OPEN_MAX);
473 if (0 < open_max)
474 return open_max;
475 /*
476 * Otherwise, we got -1 for one of the two
477 * reasons:
478 *
479 * (1) sysconf() did not understand _SC_OPEN_MAX
480 * and signaled an error with -1; or
481 * (2) sysconf() said there is no limit.
482 *
483 * We _could_ clear errno before calling sysconf() to
484 * tell these two cases apart and return a huge number
485 * in the latter case to let the caller cap it to a
486 * value that is not so selfish, but letting the
487 * fallback OPEN_MAX codepath take care of these cases
488 * is a lot simpler.
489 */
490 }
491 #endif
492
493 #ifdef OPEN_MAX
494 return OPEN_MAX;
495 #else
496 return 1; /* see the caller ;-) */
497 #endif
498 }
499
500 const char *pack_basename(struct packed_git *p)
501 {
502 const char *ret = strrchr(p->pack_name, '/');
503 if (ret)
504 ret = ret + 1; /* skip past slash */
505 else
506 ret = p->pack_name; /* we only have a base */
507 return ret;
508 }
509
510 /*
511 * Do not call this directly as this leaks p->pack_fd on error return;
512 * call open_packed_git() instead.
513 */
514 static int open_packed_git_1(struct packed_git *p)
515 {
516 struct stat st;
517 struct pack_header hdr;
518 unsigned char hash[GIT_MAX_RAWSZ];
519 unsigned char *idx_hash;
520 long fd_flag;
521 ssize_t read_result;
522 const unsigned hashsz = the_hash_algo->rawsz;
523
524 if (!p->index_data) {
525 struct multi_pack_index *m;
526 const char *pack_name = pack_basename(p);
527
528 for (m = the_repository->objects->multi_pack_index;
529 m; m = m->next) {
530 if (midx_contains_pack(m, pack_name))
531 break;
532 }
533
534 if (!m && open_pack_index(p))
535 return error("packfile %s index unavailable", p->pack_name);
536 }
537
538 if (!pack_max_fds) {
539 unsigned int max_fds = get_max_fd_limit();
540
541 /* Save 3 for stdin/stdout/stderr, 22 for work */
542 if (25 < max_fds)
543 pack_max_fds = max_fds - 25;
544 else
545 pack_max_fds = 1;
546 }
547
548 while (pack_max_fds <= pack_open_fds && close_one_pack())
549 ; /* nothing */
550
551 p->pack_fd = git_open(p->pack_name);
552 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
553 return -1;
554 pack_open_fds++;
555
556 /* If we created the struct before we had the pack we lack size. */
557 if (!p->pack_size) {
558 if (!S_ISREG(st.st_mode))
559 return error("packfile %s not a regular file", p->pack_name);
560 p->pack_size = st.st_size;
561 } else if (p->pack_size != st.st_size)
562 return error("packfile %s size changed", p->pack_name);
563
564 /* We leave these file descriptors open with sliding mmap;
565 * there is no point keeping them open across exec(), though.
566 */
567 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
568 if (fd_flag < 0)
569 return error("cannot determine file descriptor flags");
570 fd_flag |= FD_CLOEXEC;
571 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
572 return error("cannot set FD_CLOEXEC");
573
574 /* Verify we recognize this pack file format. */
575 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
576 if (read_result < 0)
577 return error_errno("error reading from %s", p->pack_name);
578 if (read_result != sizeof(hdr))
579 return error("file %s is far too short to be a packfile", p->pack_name);
580 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
581 return error("file %s is not a GIT packfile", p->pack_name);
582 if (!pack_version_ok(hdr.hdr_version))
583 return error("packfile %s is version %"PRIu32" and not"
584 " supported (try upgrading GIT to a newer version)",
585 p->pack_name, ntohl(hdr.hdr_version));
586
587 /* Skip index checking if in multi-pack-index */
588 if (!p->index_data)
589 return 0;
590
591 /* Verify the pack matches its index. */
592 if (p->num_objects != ntohl(hdr.hdr_entries))
593 return error("packfile %s claims to have %"PRIu32" objects"
594 " while index indicates %"PRIu32" objects",
595 p->pack_name, ntohl(hdr.hdr_entries),
596 p->num_objects);
597 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
598 return error("end of packfile %s is unavailable", p->pack_name);
599 read_result = read_in_full(p->pack_fd, hash, hashsz);
600 if (read_result < 0)
601 return error_errno("error reading from %s", p->pack_name);
602 if (read_result != hashsz)
603 return error("packfile %s signature is unavailable", p->pack_name);
604 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
605 if (!hasheq(hash, idx_hash))
606 return error("packfile %s does not match index", p->pack_name);
607 return 0;
608 }
609
610 static int open_packed_git(struct packed_git *p)
611 {
612 if (!open_packed_git_1(p))
613 return 0;
614 close_pack_fd(p);
615 return -1;
616 }
617
618 static int in_window(struct pack_window *win, off_t offset)
619 {
620 /* We must promise at least one full hash after the
621 * offset is available from this window, otherwise the offset
622 * is not actually in this window and a different window (which
623 * has that one hash excess) must be used. This is to support
624 * the object header and delta base parsing routines below.
625 */
626 off_t win_off = win->offset;
627 return win_off <= offset
628 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
629 }
630
631 unsigned char *use_pack(struct packed_git *p,
632 struct pack_window **w_cursor,
633 off_t offset,
634 unsigned long *left)
635 {
636 struct pack_window *win = *w_cursor;
637
638 /* Since packfiles end in a hash of their content and it's
639 * pointless to ask for an offset into the middle of that
640 * hash, and the in_window function above wouldn't match
641 * don't allow an offset too close to the end of the file.
642 */
643 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
644 die("packfile %s cannot be accessed", p->pack_name);
645 if (offset > (p->pack_size - the_hash_algo->rawsz))
646 die("offset beyond end of packfile (truncated pack?)");
647 if (offset < 0)
648 die(_("offset before end of packfile (broken .idx?)"));
649
650 if (!win || !in_window(win, offset)) {
651 if (win)
652 win->inuse_cnt--;
653 for (win = p->windows; win; win = win->next) {
654 if (in_window(win, offset))
655 break;
656 }
657 if (!win) {
658 size_t window_align = packed_git_window_size / 2;
659 off_t len;
660
661 if (p->pack_fd == -1 && open_packed_git(p))
662 die("packfile %s cannot be accessed", p->pack_name);
663
664 win = xcalloc(1, sizeof(*win));
665 win->offset = (offset / window_align) * window_align;
666 len = p->pack_size - win->offset;
667 if (len > packed_git_window_size)
668 len = packed_git_window_size;
669 win->len = (size_t)len;
670 pack_mapped += win->len;
671 while (packed_git_limit < pack_mapped
672 && unuse_one_window(p))
673 ; /* nothing */
674 win->base = xmmap_gently(NULL, win->len,
675 PROT_READ, MAP_PRIVATE,
676 p->pack_fd, win->offset);
677 if (win->base == MAP_FAILED)
678 die_errno("packfile %s cannot be mapped",
679 p->pack_name);
680 if (!win->offset && win->len == p->pack_size
681 && !p->do_not_close)
682 close_pack_fd(p);
683 pack_mmap_calls++;
684 pack_open_windows++;
685 if (pack_mapped > peak_pack_mapped)
686 peak_pack_mapped = pack_mapped;
687 if (pack_open_windows > peak_pack_open_windows)
688 peak_pack_open_windows = pack_open_windows;
689 win->next = p->windows;
690 p->windows = win;
691 }
692 }
693 if (win != *w_cursor) {
694 win->last_used = pack_used_ctr++;
695 win->inuse_cnt++;
696 *w_cursor = win;
697 }
698 offset -= win->offset;
699 if (left)
700 *left = win->len - xsize_t(offset);
701 return win->base + offset;
702 }
703
704 void unuse_pack(struct pack_window **w_cursor)
705 {
706 struct pack_window *w = *w_cursor;
707 if (w) {
708 w->inuse_cnt--;
709 *w_cursor = NULL;
710 }
711 }
712
713 static void try_to_free_pack_memory(size_t size)
714 {
715 release_pack_memory(size);
716 }
717
718 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
719 {
720 static int have_set_try_to_free_routine;
721 struct stat st;
722 size_t alloc;
723 struct packed_git *p;
724
725 if (!have_set_try_to_free_routine) {
726 have_set_try_to_free_routine = 1;
727 set_try_to_free_routine(try_to_free_pack_memory);
728 }
729
730 /*
731 * Make sure a corresponding .pack file exists and that
732 * the index looks sane.
733 */
734 if (!strip_suffix_mem(path, &path_len, ".idx"))
735 return NULL;
736
737 /*
738 * ".promisor" is long enough to hold any suffix we're adding (and
739 * the use xsnprintf double-checks that)
740 */
741 alloc = st_add3(path_len, strlen(".promisor"), 1);
742 p = alloc_packed_git(alloc);
743 memcpy(p->pack_name, path, path_len);
744
745 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
746 if (!access(p->pack_name, F_OK))
747 p->pack_keep = 1;
748
749 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
750 if (!access(p->pack_name, F_OK))
751 p->pack_promisor = 1;
752
753 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
754 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
755 free(p);
756 return NULL;
757 }
758
759 /* ok, it looks sane as far as we can check without
760 * actually mapping the pack file.
761 */
762 p->pack_size = st.st_size;
763 p->pack_local = local;
764 p->mtime = st.st_mtime;
765 if (path_len < the_hash_algo->hexsz ||
766 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
767 hashclr(p->hash);
768 return p;
769 }
770
771 void install_packed_git(struct repository *r, struct packed_git *pack)
772 {
773 if (pack->pack_fd != -1)
774 pack_open_fds++;
775
776 pack->next = r->objects->packed_git;
777 r->objects->packed_git = pack;
778 }
779
780 void (*report_garbage)(unsigned seen_bits, const char *path);
781
782 static void report_helper(const struct string_list *list,
783 int seen_bits, int first, int last)
784 {
785 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
786 return;
787
788 for (; first < last; first++)
789 report_garbage(seen_bits, list->items[first].string);
790 }
791
792 static void report_pack_garbage(struct string_list *list)
793 {
794 int i, baselen = -1, first = 0, seen_bits = 0;
795
796 if (!report_garbage)
797 return;
798
799 string_list_sort(list);
800
801 for (i = 0; i < list->nr; i++) {
802 const char *path = list->items[i].string;
803 if (baselen != -1 &&
804 strncmp(path, list->items[first].string, baselen)) {
805 report_helper(list, seen_bits, first, i);
806 baselen = -1;
807 seen_bits = 0;
808 }
809 if (baselen == -1) {
810 const char *dot = strrchr(path, '.');
811 if (!dot) {
812 report_garbage(PACKDIR_FILE_GARBAGE, path);
813 continue;
814 }
815 baselen = dot - path + 1;
816 first = i;
817 }
818 if (!strcmp(path + baselen, "pack"))
819 seen_bits |= 1;
820 else if (!strcmp(path + baselen, "idx"))
821 seen_bits |= 2;
822 }
823 report_helper(list, seen_bits, first, list->nr);
824 }
825
826 void for_each_file_in_pack_dir(const char *objdir,
827 each_file_in_pack_dir_fn fn,
828 void *data)
829 {
830 struct strbuf path = STRBUF_INIT;
831 size_t dirnamelen;
832 DIR *dir;
833 struct dirent *de;
834
835 strbuf_addstr(&path, objdir);
836 strbuf_addstr(&path, "/pack");
837 dir = opendir(path.buf);
838 if (!dir) {
839 if (errno != ENOENT)
840 error_errno("unable to open object pack directory: %s",
841 path.buf);
842 strbuf_release(&path);
843 return;
844 }
845 strbuf_addch(&path, '/');
846 dirnamelen = path.len;
847 while ((de = readdir(dir)) != NULL) {
848 if (is_dot_or_dotdot(de->d_name))
849 continue;
850
851 strbuf_setlen(&path, dirnamelen);
852 strbuf_addstr(&path, de->d_name);
853
854 fn(path.buf, path.len, de->d_name, data);
855 }
856
857 closedir(dir);
858 strbuf_release(&path);
859 }
860
861 struct prepare_pack_data {
862 struct repository *r;
863 struct string_list *garbage;
864 int local;
865 struct multi_pack_index *m;
866 };
867
868 static void prepare_pack(const char *full_name, size_t full_name_len,
869 const char *file_name, void *_data)
870 {
871 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
872 struct packed_git *p;
873 size_t base_len = full_name_len;
874
875 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
876 !(data->m && midx_contains_pack(data->m, file_name))) {
877 /* Don't reopen a pack we already have. */
878 for (p = data->r->objects->packed_git; p; p = p->next) {
879 size_t len;
880 if (strip_suffix(p->pack_name, ".pack", &len) &&
881 len == base_len &&
882 !memcmp(p->pack_name, full_name, len))
883 break;
884 }
885
886 if (!p) {
887 p = add_packed_git(full_name, full_name_len, data->local);
888 if (p)
889 install_packed_git(data->r, p);
890 }
891 }
892
893 if (!report_garbage)
894 return;
895
896 if (!strcmp(file_name, "multi-pack-index"))
897 return;
898 if (ends_with(file_name, ".idx") ||
899 ends_with(file_name, ".pack") ||
900 ends_with(file_name, ".bitmap") ||
901 ends_with(file_name, ".keep") ||
902 ends_with(file_name, ".promisor"))
903 string_list_append(data->garbage, full_name);
904 else
905 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
906 }
907
908 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
909 {
910 struct prepare_pack_data data;
911 struct string_list garbage = STRING_LIST_INIT_DUP;
912
913 data.m = r->objects->multi_pack_index;
914
915 /* look for the multi-pack-index for this object directory */
916 while (data.m && strcmp(data.m->object_dir, objdir))
917 data.m = data.m->next;
918
919 data.r = r;
920 data.garbage = &garbage;
921 data.local = local;
922
923 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
924
925 report_pack_garbage(data.garbage);
926 string_list_clear(data.garbage, 0);
927 }
928
929 static void prepare_packed_git(struct repository *r);
930 /*
931 * Give a fast, rough count of the number of objects in the repository. This
932 * ignores loose objects completely. If you have a lot of them, then either
933 * you should repack because your performance will be awful, or they are
934 * all unreachable objects about to be pruned, in which case they're not really
935 * interesting as a measure of repo size in the first place.
936 */
937 unsigned long repo_approximate_object_count(struct repository *r)
938 {
939 if (!r->objects->approximate_object_count_valid) {
940 unsigned long count;
941 struct multi_pack_index *m;
942 struct packed_git *p;
943
944 prepare_packed_git(r);
945 count = 0;
946 for (m = get_multi_pack_index(r); m; m = m->next)
947 count += m->num_objects;
948 for (p = r->objects->packed_git; p; p = p->next) {
949 if (open_pack_index(p))
950 continue;
951 count += p->num_objects;
952 }
953 r->objects->approximate_object_count = count;
954 }
955 return r->objects->approximate_object_count;
956 }
957
958 static void *get_next_packed_git(const void *p)
959 {
960 return ((const struct packed_git *)p)->next;
961 }
962
963 static void set_next_packed_git(void *p, void *next)
964 {
965 ((struct packed_git *)p)->next = next;
966 }
967
968 static int sort_pack(const void *a_, const void *b_)
969 {
970 const struct packed_git *a = a_;
971 const struct packed_git *b = b_;
972 int st;
973
974 /*
975 * Local packs tend to contain objects specific to our
976 * variant of the project than remote ones. In addition,
977 * remote ones could be on a network mounted filesystem.
978 * Favor local ones for these reasons.
979 */
980 st = a->pack_local - b->pack_local;
981 if (st)
982 return -st;
983
984 /*
985 * Younger packs tend to contain more recent objects,
986 * and more recent objects tend to get accessed more
987 * often.
988 */
989 if (a->mtime < b->mtime)
990 return 1;
991 else if (a->mtime == b->mtime)
992 return 0;
993 return -1;
994 }
995
996 static void rearrange_packed_git(struct repository *r)
997 {
998 r->objects->packed_git = llist_mergesort(
999 r->objects->packed_git, get_next_packed_git,
1000 set_next_packed_git, sort_pack);
1001 }
1002
1003 static void prepare_packed_git_mru(struct repository *r)
1004 {
1005 struct packed_git *p;
1006
1007 INIT_LIST_HEAD(&r->objects->packed_git_mru);
1008
1009 for (p = r->objects->packed_git; p; p = p->next)
1010 list_add_tail(&p->mru, &r->objects->packed_git_mru);
1011 }
1012
1013 static void prepare_packed_git(struct repository *r)
1014 {
1015 struct object_directory *odb;
1016
1017 if (r->objects->packed_git_initialized)
1018 return;
1019
1020 prepare_alt_odb(r);
1021 for (odb = r->objects->odb; odb; odb = odb->next) {
1022 int local = (odb == r->objects->odb);
1023 prepare_multi_pack_index_one(r, odb->path, local);
1024 prepare_packed_git_one(r, odb->path, local);
1025 }
1026 rearrange_packed_git(r);
1027
1028 prepare_packed_git_mru(r);
1029 r->objects->packed_git_initialized = 1;
1030 }
1031
1032 void reprepare_packed_git(struct repository *r)
1033 {
1034 struct object_directory *odb;
1035
1036 for (odb = r->objects->odb; odb; odb = odb->next)
1037 odb_clear_loose_cache(odb);
1038
1039 r->objects->approximate_object_count_valid = 0;
1040 r->objects->packed_git_initialized = 0;
1041 prepare_packed_git(r);
1042 }
1043
1044 struct packed_git *get_packed_git(struct repository *r)
1045 {
1046 prepare_packed_git(r);
1047 return r->objects->packed_git;
1048 }
1049
1050 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1051 {
1052 prepare_packed_git(r);
1053 return r->objects->multi_pack_index;
1054 }
1055
1056 struct packed_git *get_all_packs(struct repository *r)
1057 {
1058 struct multi_pack_index *m;
1059
1060 prepare_packed_git(r);
1061 for (m = r->objects->multi_pack_index; m; m = m->next) {
1062 uint32_t i;
1063 for (i = 0; i < m->num_packs; i++)
1064 prepare_midx_pack(r, m, i);
1065 }
1066
1067 return r->objects->packed_git;
1068 }
1069
1070 struct list_head *get_packed_git_mru(struct repository *r)
1071 {
1072 prepare_packed_git(r);
1073 return &r->objects->packed_git_mru;
1074 }
1075
1076 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1077 unsigned long len, enum object_type *type, unsigned long *sizep)
1078 {
1079 unsigned shift;
1080 unsigned long size, c;
1081 unsigned long used = 0;
1082
1083 c = buf[used++];
1084 *type = (c >> 4) & 7;
1085 size = c & 15;
1086 shift = 4;
1087 while (c & 0x80) {
1088 if (len <= used || bitsizeof(long) <= shift) {
1089 error("bad object header");
1090 size = used = 0;
1091 break;
1092 }
1093 c = buf[used++];
1094 size += (c & 0x7f) << shift;
1095 shift += 7;
1096 }
1097 *sizep = size;
1098 return used;
1099 }
1100
1101 unsigned long get_size_from_delta(struct packed_git *p,
1102 struct pack_window **w_curs,
1103 off_t curpos)
1104 {
1105 const unsigned char *data;
1106 unsigned char delta_head[20], *in;
1107 git_zstream stream;
1108 int st;
1109
1110 memset(&stream, 0, sizeof(stream));
1111 stream.next_out = delta_head;
1112 stream.avail_out = sizeof(delta_head);
1113
1114 git_inflate_init(&stream);
1115 do {
1116 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1117 stream.next_in = in;
1118 st = git_inflate(&stream, Z_FINISH);
1119 curpos += stream.next_in - in;
1120 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1121 stream.total_out < sizeof(delta_head));
1122 git_inflate_end(&stream);
1123 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1124 error("delta data unpack-initial failed");
1125 return 0;
1126 }
1127
1128 /* Examine the initial part of the delta to figure out
1129 * the result size.
1130 */
1131 data = delta_head;
1132
1133 /* ignore base size */
1134 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1135
1136 /* Read the result size */
1137 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1138 }
1139
1140 int unpack_object_header(struct packed_git *p,
1141 struct pack_window **w_curs,
1142 off_t *curpos,
1143 unsigned long *sizep)
1144 {
1145 unsigned char *base;
1146 unsigned long left;
1147 unsigned long used;
1148 enum object_type type;
1149
1150 /* use_pack() assures us we have [base, base + 20) available
1151 * as a range that we can look at. (Its actually the hash
1152 * size that is assured.) With our object header encoding
1153 * the maximum deflated object size is 2^137, which is just
1154 * insane, so we know won't exceed what we have been given.
1155 */
1156 base = use_pack(p, w_curs, *curpos, &left);
1157 used = unpack_object_header_buffer(base, left, &type, sizep);
1158 if (!used) {
1159 type = OBJ_BAD;
1160 } else
1161 *curpos += used;
1162
1163 return type;
1164 }
1165
1166 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1167 {
1168 unsigned i;
1169 const unsigned hashsz = the_hash_algo->rawsz;
1170 for (i = 0; i < p->num_bad_objects; i++)
1171 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1172 return;
1173 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1174 st_mult(GIT_MAX_RAWSZ,
1175 st_add(p->num_bad_objects, 1)));
1176 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1177 p->num_bad_objects++;
1178 }
1179
1180 const struct packed_git *has_packed_and_bad(struct repository *r,
1181 const unsigned char *sha1)
1182 {
1183 struct packed_git *p;
1184 unsigned i;
1185
1186 for (p = r->objects->packed_git; p; p = p->next)
1187 for (i = 0; i < p->num_bad_objects; i++)
1188 if (hasheq(sha1,
1189 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1190 return p;
1191 return NULL;
1192 }
1193
1194 static off_t get_delta_base(struct packed_git *p,
1195 struct pack_window **w_curs,
1196 off_t *curpos,
1197 enum object_type type,
1198 off_t delta_obj_offset)
1199 {
1200 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1201 off_t base_offset;
1202
1203 /* use_pack() assured us we have [base_info, base_info + 20)
1204 * as a range that we can look at without walking off the
1205 * end of the mapped window. Its actually the hash size
1206 * that is assured. An OFS_DELTA longer than the hash size
1207 * is stupid, as then a REF_DELTA would be smaller to store.
1208 */
1209 if (type == OBJ_OFS_DELTA) {
1210 unsigned used = 0;
1211 unsigned char c = base_info[used++];
1212 base_offset = c & 127;
1213 while (c & 128) {
1214 base_offset += 1;
1215 if (!base_offset || MSB(base_offset, 7))
1216 return 0; /* overflow */
1217 c = base_info[used++];
1218 base_offset = (base_offset << 7) + (c & 127);
1219 }
1220 base_offset = delta_obj_offset - base_offset;
1221 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1222 return 0; /* out of bound */
1223 *curpos += used;
1224 } else if (type == OBJ_REF_DELTA) {
1225 /* The base entry _must_ be in the same pack */
1226 base_offset = find_pack_entry_one(base_info, p);
1227 *curpos += the_hash_algo->rawsz;
1228 } else
1229 die("I am totally screwed");
1230 return base_offset;
1231 }
1232
1233 /*
1234 * Like get_delta_base above, but we return the sha1 instead of the pack
1235 * offset. This means it is cheaper for REF deltas (we do not have to do
1236 * the final object lookup), but more expensive for OFS deltas (we
1237 * have to load the revidx to convert the offset back into a sha1).
1238 */
1239 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1240 struct pack_window **w_curs,
1241 off_t curpos,
1242 enum object_type type,
1243 off_t delta_obj_offset)
1244 {
1245 if (type == OBJ_REF_DELTA) {
1246 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1247 return base;
1248 } else if (type == OBJ_OFS_DELTA) {
1249 struct revindex_entry *revidx;
1250 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1251 type, delta_obj_offset);
1252
1253 if (!base_offset)
1254 return NULL;
1255
1256 revidx = find_pack_revindex(p, base_offset);
1257 if (!revidx)
1258 return NULL;
1259
1260 return nth_packed_object_sha1(p, revidx->nr);
1261 } else
1262 return NULL;
1263 }
1264
1265 static int retry_bad_packed_offset(struct repository *r,
1266 struct packed_git *p,
1267 off_t obj_offset)
1268 {
1269 int type;
1270 struct revindex_entry *revidx;
1271 struct object_id oid;
1272 revidx = find_pack_revindex(p, obj_offset);
1273 if (!revidx)
1274 return OBJ_BAD;
1275 nth_packed_object_oid(&oid, p, revidx->nr);
1276 mark_bad_packed_object(p, oid.hash);
1277 type = oid_object_info(r, &oid, NULL);
1278 if (type <= OBJ_NONE)
1279 return OBJ_BAD;
1280 return type;
1281 }
1282
1283 #define POI_STACK_PREALLOC 64
1284
1285 static enum object_type packed_to_object_type(struct repository *r,
1286 struct packed_git *p,
1287 off_t obj_offset,
1288 enum object_type type,
1289 struct pack_window **w_curs,
1290 off_t curpos)
1291 {
1292 off_t small_poi_stack[POI_STACK_PREALLOC];
1293 off_t *poi_stack = small_poi_stack;
1294 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1295
1296 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1297 off_t base_offset;
1298 unsigned long size;
1299 /* Push the object we're going to leave behind */
1300 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1301 poi_stack_alloc = alloc_nr(poi_stack_nr);
1302 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1303 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1304 } else {
1305 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1306 }
1307 poi_stack[poi_stack_nr++] = obj_offset;
1308 /* If parsing the base offset fails, just unwind */
1309 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1310 if (!base_offset)
1311 goto unwind;
1312 curpos = obj_offset = base_offset;
1313 type = unpack_object_header(p, w_curs, &curpos, &size);
1314 if (type <= OBJ_NONE) {
1315 /* If getting the base itself fails, we first
1316 * retry the base, otherwise unwind */
1317 type = retry_bad_packed_offset(r, p, base_offset);
1318 if (type > OBJ_NONE)
1319 goto out;
1320 goto unwind;
1321 }
1322 }
1323
1324 switch (type) {
1325 case OBJ_BAD:
1326 case OBJ_COMMIT:
1327 case OBJ_TREE:
1328 case OBJ_BLOB:
1329 case OBJ_TAG:
1330 break;
1331 default:
1332 error("unknown object type %i at offset %"PRIuMAX" in %s",
1333 type, (uintmax_t)obj_offset, p->pack_name);
1334 type = OBJ_BAD;
1335 }
1336
1337 out:
1338 if (poi_stack != small_poi_stack)
1339 free(poi_stack);
1340 return type;
1341
1342 unwind:
1343 while (poi_stack_nr) {
1344 obj_offset = poi_stack[--poi_stack_nr];
1345 type = retry_bad_packed_offset(r, p, obj_offset);
1346 if (type > OBJ_NONE)
1347 goto out;
1348 }
1349 type = OBJ_BAD;
1350 goto out;
1351 }
1352
1353 static struct hashmap delta_base_cache;
1354 static size_t delta_base_cached;
1355
1356 static LIST_HEAD(delta_base_cache_lru);
1357
1358 struct delta_base_cache_key {
1359 struct packed_git *p;
1360 off_t base_offset;
1361 };
1362
1363 struct delta_base_cache_entry {
1364 struct hashmap hash;
1365 struct delta_base_cache_key key;
1366 struct list_head lru;
1367 void *data;
1368 unsigned long size;
1369 enum object_type type;
1370 };
1371
1372 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1373 {
1374 unsigned int hash;
1375
1376 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1377 hash += (hash >> 8) + (hash >> 16);
1378 return hash;
1379 }
1380
1381 static struct delta_base_cache_entry *
1382 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1383 {
1384 struct hashmap_entry entry;
1385 struct delta_base_cache_key key;
1386
1387 if (!delta_base_cache.cmpfn)
1388 return NULL;
1389
1390 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1391 key.p = p;
1392 key.base_offset = base_offset;
1393 return hashmap_get(&delta_base_cache, &entry, &key);
1394 }
1395
1396 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1397 const struct delta_base_cache_key *b)
1398 {
1399 return a->p == b->p && a->base_offset == b->base_offset;
1400 }
1401
1402 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1403 const void *va, const void *vb,
1404 const void *vkey)
1405 {
1406 const struct delta_base_cache_entry *a = va, *b = vb;
1407 const struct delta_base_cache_key *key = vkey;
1408 if (key)
1409 return !delta_base_cache_key_eq(&a->key, key);
1410 else
1411 return !delta_base_cache_key_eq(&a->key, &b->key);
1412 }
1413
1414 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1415 {
1416 return !!get_delta_base_cache_entry(p, base_offset);
1417 }
1418
1419 /*
1420 * Remove the entry from the cache, but do _not_ free the associated
1421 * entry data. The caller takes ownership of the "data" buffer, and
1422 * should copy out any fields it wants before detaching.
1423 */
1424 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1425 {
1426 hashmap_remove(&delta_base_cache, ent, &ent->key);
1427 list_del(&ent->lru);
1428 delta_base_cached -= ent->size;
1429 free(ent);
1430 }
1431
1432 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1433 off_t base_offset, unsigned long *base_size,
1434 enum object_type *type)
1435 {
1436 struct delta_base_cache_entry *ent;
1437
1438 ent = get_delta_base_cache_entry(p, base_offset);
1439 if (!ent)
1440 return unpack_entry(r, p, base_offset, type, base_size);
1441
1442 if (type)
1443 *type = ent->type;
1444 if (base_size)
1445 *base_size = ent->size;
1446 return xmemdupz(ent->data, ent->size);
1447 }
1448
1449 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1450 {
1451 free(ent->data);
1452 detach_delta_base_cache_entry(ent);
1453 }
1454
1455 void clear_delta_base_cache(void)
1456 {
1457 struct list_head *lru, *tmp;
1458 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1459 struct delta_base_cache_entry *entry =
1460 list_entry(lru, struct delta_base_cache_entry, lru);
1461 release_delta_base_cache(entry);
1462 }
1463 }
1464
1465 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1466 void *base, unsigned long base_size, enum object_type type)
1467 {
1468 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1469 struct list_head *lru, *tmp;
1470
1471 delta_base_cached += base_size;
1472
1473 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1474 struct delta_base_cache_entry *f =
1475 list_entry(lru, struct delta_base_cache_entry, lru);
1476 if (delta_base_cached <= delta_base_cache_limit)
1477 break;
1478 release_delta_base_cache(f);
1479 }
1480
1481 ent->key.p = p;
1482 ent->key.base_offset = base_offset;
1483 ent->type = type;
1484 ent->data = base;
1485 ent->size = base_size;
1486 list_add_tail(&ent->lru, &delta_base_cache_lru);
1487
1488 if (!delta_base_cache.cmpfn)
1489 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1490 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1491 hashmap_add(&delta_base_cache, ent);
1492 }
1493
1494 int packed_object_info(struct repository *r, struct packed_git *p,
1495 off_t obj_offset, struct object_info *oi)
1496 {
1497 struct pack_window *w_curs = NULL;
1498 unsigned long size;
1499 off_t curpos = obj_offset;
1500 enum object_type type;
1501
1502 /*
1503 * We always get the representation type, but only convert it to
1504 * a "real" type later if the caller is interested.
1505 */
1506 if (oi->contentp) {
1507 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1508 &type);
1509 if (!*oi->contentp)
1510 type = OBJ_BAD;
1511 } else {
1512 type = unpack_object_header(p, &w_curs, &curpos, &size);
1513 }
1514
1515 if (!oi->contentp && oi->sizep) {
1516 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1517 off_t tmp_pos = curpos;
1518 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1519 type, obj_offset);
1520 if (!base_offset) {
1521 type = OBJ_BAD;
1522 goto out;
1523 }
1524 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1525 if (*oi->sizep == 0) {
1526 type = OBJ_BAD;
1527 goto out;
1528 }
1529 } else {
1530 *oi->sizep = size;
1531 }
1532 }
1533
1534 if (oi->disk_sizep) {
1535 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1536 *oi->disk_sizep = revidx[1].offset - obj_offset;
1537 }
1538
1539 if (oi->typep || oi->type_name) {
1540 enum object_type ptot;
1541 ptot = packed_to_object_type(r, p, obj_offset,
1542 type, &w_curs, curpos);
1543 if (oi->typep)
1544 *oi->typep = ptot;
1545 if (oi->type_name) {
1546 const char *tn = type_name(ptot);
1547 if (tn)
1548 strbuf_addstr(oi->type_name, tn);
1549 }
1550 if (ptot < 0) {
1551 type = OBJ_BAD;
1552 goto out;
1553 }
1554 }
1555
1556 if (oi->delta_base_sha1) {
1557 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1558 const unsigned char *base;
1559
1560 base = get_delta_base_sha1(p, &w_curs, curpos,
1561 type, obj_offset);
1562 if (!base) {
1563 type = OBJ_BAD;
1564 goto out;
1565 }
1566
1567 hashcpy(oi->delta_base_sha1, base);
1568 } else
1569 hashclr(oi->delta_base_sha1);
1570 }
1571
1572 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1573 OI_PACKED;
1574
1575 out:
1576 unuse_pack(&w_curs);
1577 return type;
1578 }
1579
1580 static void *unpack_compressed_entry(struct packed_git *p,
1581 struct pack_window **w_curs,
1582 off_t curpos,
1583 unsigned long size)
1584 {
1585 int st;
1586 git_zstream stream;
1587 unsigned char *buffer, *in;
1588
1589 buffer = xmallocz_gently(size);
1590 if (!buffer)
1591 return NULL;
1592 memset(&stream, 0, sizeof(stream));
1593 stream.next_out = buffer;
1594 stream.avail_out = size + 1;
1595
1596 git_inflate_init(&stream);
1597 do {
1598 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1599 stream.next_in = in;
1600 st = git_inflate(&stream, Z_FINISH);
1601 if (!stream.avail_out)
1602 break; /* the payload is larger than it should be */
1603 curpos += stream.next_in - in;
1604 } while (st == Z_OK || st == Z_BUF_ERROR);
1605 git_inflate_end(&stream);
1606 if ((st != Z_STREAM_END) || stream.total_out != size) {
1607 free(buffer);
1608 return NULL;
1609 }
1610
1611 /* versions of zlib can clobber unconsumed portion of outbuf */
1612 buffer[size] = '\0';
1613
1614 return buffer;
1615 }
1616
1617 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1618 {
1619 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1620 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1621 p->pack_name, (uintmax_t)obj_offset);
1622 }
1623
1624 int do_check_packed_object_crc;
1625
1626 #define UNPACK_ENTRY_STACK_PREALLOC 64
1627 struct unpack_entry_stack_ent {
1628 off_t obj_offset;
1629 off_t curpos;
1630 unsigned long size;
1631 };
1632
1633 static void *read_object(struct repository *r,
1634 const struct object_id *oid,
1635 enum object_type *type,
1636 unsigned long *size)
1637 {
1638 struct object_info oi = OBJECT_INFO_INIT;
1639 void *content;
1640 oi.typep = type;
1641 oi.sizep = size;
1642 oi.contentp = &content;
1643
1644 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1645 return NULL;
1646 return content;
1647 }
1648
1649 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1650 enum object_type *final_type, unsigned long *final_size)
1651 {
1652 struct pack_window *w_curs = NULL;
1653 off_t curpos = obj_offset;
1654 void *data = NULL;
1655 unsigned long size;
1656 enum object_type type;
1657 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1658 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1659 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1660 int base_from_cache = 0;
1661
1662 write_pack_access_log(p, obj_offset);
1663
1664 /* PHASE 1: drill down to the innermost base object */
1665 for (;;) {
1666 off_t base_offset;
1667 int i;
1668 struct delta_base_cache_entry *ent;
1669
1670 ent = get_delta_base_cache_entry(p, curpos);
1671 if (ent) {
1672 type = ent->type;
1673 data = ent->data;
1674 size = ent->size;
1675 detach_delta_base_cache_entry(ent);
1676 base_from_cache = 1;
1677 break;
1678 }
1679
1680 if (do_check_packed_object_crc && p->index_version > 1) {
1681 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1682 off_t len = revidx[1].offset - obj_offset;
1683 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1684 struct object_id oid;
1685 nth_packed_object_oid(&oid, p, revidx->nr);
1686 error("bad packed object CRC for %s",
1687 oid_to_hex(&oid));
1688 mark_bad_packed_object(p, oid.hash);
1689 data = NULL;
1690 goto out;
1691 }
1692 }
1693
1694 type = unpack_object_header(p, &w_curs, &curpos, &size);
1695 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1696 break;
1697
1698 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1699 if (!base_offset) {
1700 error("failed to validate delta base reference "
1701 "at offset %"PRIuMAX" from %s",
1702 (uintmax_t)curpos, p->pack_name);
1703 /* bail to phase 2, in hopes of recovery */
1704 data = NULL;
1705 break;
1706 }
1707
1708 /* push object, proceed to base */
1709 if (delta_stack_nr >= delta_stack_alloc
1710 && delta_stack == small_delta_stack) {
1711 delta_stack_alloc = alloc_nr(delta_stack_nr);
1712 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1713 COPY_ARRAY(delta_stack, small_delta_stack,
1714 delta_stack_nr);
1715 } else {
1716 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1717 }
1718 i = delta_stack_nr++;
1719 delta_stack[i].obj_offset = obj_offset;
1720 delta_stack[i].curpos = curpos;
1721 delta_stack[i].size = size;
1722
1723 curpos = obj_offset = base_offset;
1724 }
1725
1726 /* PHASE 2: handle the base */
1727 switch (type) {
1728 case OBJ_OFS_DELTA:
1729 case OBJ_REF_DELTA:
1730 if (data)
1731 BUG("unpack_entry: left loop at a valid delta");
1732 break;
1733 case OBJ_COMMIT:
1734 case OBJ_TREE:
1735 case OBJ_BLOB:
1736 case OBJ_TAG:
1737 if (!base_from_cache)
1738 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1739 break;
1740 default:
1741 data = NULL;
1742 error("unknown object type %i at offset %"PRIuMAX" in %s",
1743 type, (uintmax_t)obj_offset, p->pack_name);
1744 }
1745
1746 /* PHASE 3: apply deltas in order */
1747
1748 /* invariants:
1749 * 'data' holds the base data, or NULL if there was corruption
1750 */
1751 while (delta_stack_nr) {
1752 void *delta_data;
1753 void *base = data;
1754 void *external_base = NULL;
1755 unsigned long delta_size, base_size = size;
1756 int i;
1757
1758 data = NULL;
1759
1760 if (base)
1761 add_delta_base_cache(p, obj_offset, base, base_size, type);
1762
1763 if (!base) {
1764 /*
1765 * We're probably in deep shit, but let's try to fetch
1766 * the required base anyway from another pack or loose.
1767 * This is costly but should happen only in the presence
1768 * of a corrupted pack, and is better than failing outright.
1769 */
1770 struct revindex_entry *revidx;
1771 struct object_id base_oid;
1772 revidx = find_pack_revindex(p, obj_offset);
1773 if (revidx) {
1774 nth_packed_object_oid(&base_oid, p, revidx->nr);
1775 error("failed to read delta base object %s"
1776 " at offset %"PRIuMAX" from %s",
1777 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1778 p->pack_name);
1779 mark_bad_packed_object(p, base_oid.hash);
1780 base = read_object(r, &base_oid, &type, &base_size);
1781 external_base = base;
1782 }
1783 }
1784
1785 i = --delta_stack_nr;
1786 obj_offset = delta_stack[i].obj_offset;
1787 curpos = delta_stack[i].curpos;
1788 delta_size = delta_stack[i].size;
1789
1790 if (!base)
1791 continue;
1792
1793 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1794
1795 if (!delta_data) {
1796 error("failed to unpack compressed delta "
1797 "at offset %"PRIuMAX" from %s",
1798 (uintmax_t)curpos, p->pack_name);
1799 data = NULL;
1800 free(external_base);
1801 continue;
1802 }
1803
1804 data = patch_delta(base, base_size,
1805 delta_data, delta_size,
1806 &size);
1807
1808 /*
1809 * We could not apply the delta; warn the user, but keep going.
1810 * Our failure will be noticed either in the next iteration of
1811 * the loop, or if this is the final delta, in the caller when
1812 * we return NULL. Those code paths will take care of making
1813 * a more explicit warning and retrying with another copy of
1814 * the object.
1815 */
1816 if (!data)
1817 error("failed to apply delta");
1818
1819 free(delta_data);
1820 free(external_base);
1821 }
1822
1823 if (final_type)
1824 *final_type = type;
1825 if (final_size)
1826 *final_size = size;
1827
1828 out:
1829 unuse_pack(&w_curs);
1830
1831 if (delta_stack != small_delta_stack)
1832 free(delta_stack);
1833
1834 return data;
1835 }
1836
1837 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1838 {
1839 const unsigned char *index_fanout = p->index_data;
1840 const unsigned char *index_lookup;
1841 const unsigned int hashsz = the_hash_algo->rawsz;
1842 int index_lookup_width;
1843
1844 if (!index_fanout)
1845 BUG("bsearch_pack called without a valid pack-index");
1846
1847 index_lookup = index_fanout + 4 * 256;
1848 if (p->index_version == 1) {
1849 index_lookup_width = hashsz + 4;
1850 index_lookup += 4;
1851 } else {
1852 index_lookup_width = hashsz;
1853 index_fanout += 8;
1854 index_lookup += 8;
1855 }
1856
1857 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1858 index_lookup, index_lookup_width, result);
1859 }
1860
1861 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1862 uint32_t n)
1863 {
1864 const unsigned char *index = p->index_data;
1865 const unsigned int hashsz = the_hash_algo->rawsz;
1866 if (!index) {
1867 if (open_pack_index(p))
1868 return NULL;
1869 index = p->index_data;
1870 }
1871 if (n >= p->num_objects)
1872 return NULL;
1873 index += 4 * 256;
1874 if (p->index_version == 1) {
1875 return index + (hashsz + 4) * n + 4;
1876 } else {
1877 index += 8;
1878 return index + hashsz * n;
1879 }
1880 }
1881
1882 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1883 struct packed_git *p,
1884 uint32_t n)
1885 {
1886 const unsigned char *hash = nth_packed_object_sha1(p, n);
1887 if (!hash)
1888 return NULL;
1889 hashcpy(oid->hash, hash);
1890 return oid;
1891 }
1892
1893 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1894 {
1895 const unsigned char *ptr = vptr;
1896 const unsigned char *start = p->index_data;
1897 const unsigned char *end = start + p->index_size;
1898 if (ptr < start)
1899 die(_("offset before start of pack index for %s (corrupt index?)"),
1900 p->pack_name);
1901 /* No need to check for underflow; .idx files must be at least 8 bytes */
1902 if (ptr >= end - 8)
1903 die(_("offset beyond end of pack index for %s (truncated index?)"),
1904 p->pack_name);
1905 }
1906
1907 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1908 {
1909 const unsigned char *index = p->index_data;
1910 const unsigned int hashsz = the_hash_algo->rawsz;
1911 index += 4 * 256;
1912 if (p->index_version == 1) {
1913 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1914 } else {
1915 uint32_t off;
1916 index += 8 + p->num_objects * (hashsz + 4);
1917 off = ntohl(*((uint32_t *)(index + 4 * n)));
1918 if (!(off & 0x80000000))
1919 return off;
1920 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1921 check_pack_index_ptr(p, index);
1922 return get_be64(index);
1923 }
1924 }
1925
1926 off_t find_pack_entry_one(const unsigned char *sha1,
1927 struct packed_git *p)
1928 {
1929 const unsigned char *index = p->index_data;
1930 struct object_id oid;
1931 uint32_t result;
1932
1933 if (!index) {
1934 if (open_pack_index(p))
1935 return 0;
1936 }
1937
1938 hashcpy(oid.hash, sha1);
1939 if (bsearch_pack(&oid, p, &result))
1940 return nth_packed_object_offset(p, result);
1941 return 0;
1942 }
1943
1944 int is_pack_valid(struct packed_git *p)
1945 {
1946 /* An already open pack is known to be valid. */
1947 if (p->pack_fd != -1)
1948 return 1;
1949
1950 /* If the pack has one window completely covering the
1951 * file size, the pack is known to be valid even if
1952 * the descriptor is not currently open.
1953 */
1954 if (p->windows) {
1955 struct pack_window *w = p->windows;
1956
1957 if (!w->offset && w->len == p->pack_size)
1958 return 1;
1959 }
1960
1961 /* Force the pack to open to prove its valid. */
1962 return !open_packed_git(p);
1963 }
1964
1965 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1966 struct packed_git *packs)
1967 {
1968 struct packed_git *p;
1969
1970 for (p = packs; p; p = p->next) {
1971 if (find_pack_entry_one(sha1, p))
1972 return p;
1973 }
1974 return NULL;
1975
1976 }
1977
1978 static int fill_pack_entry(const struct object_id *oid,
1979 struct pack_entry *e,
1980 struct packed_git *p)
1981 {
1982 off_t offset;
1983
1984 if (p->num_bad_objects) {
1985 unsigned i;
1986 for (i = 0; i < p->num_bad_objects; i++)
1987 if (hasheq(oid->hash,
1988 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1989 return 0;
1990 }
1991
1992 offset = find_pack_entry_one(oid->hash, p);
1993 if (!offset)
1994 return 0;
1995
1996 /*
1997 * We are about to tell the caller where they can locate the
1998 * requested object. We better make sure the packfile is
1999 * still here and can be accessed before supplying that
2000 * answer, as it may have been deleted since the index was
2001 * loaded!
2002 */
2003 if (!is_pack_valid(p))
2004 return 0;
2005 e->offset = offset;
2006 e->p = p;
2007 return 1;
2008 }
2009
2010 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2011 {
2012 struct list_head *pos;
2013 struct multi_pack_index *m;
2014
2015 prepare_packed_git(r);
2016 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2017 return 0;
2018
2019 for (m = r->objects->multi_pack_index; m; m = m->next) {
2020 if (fill_midx_entry(r, oid, e, m))
2021 return 1;
2022 }
2023
2024 list_for_each(pos, &r->objects->packed_git_mru) {
2025 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2026 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2027 list_move(&p->mru, &r->objects->packed_git_mru);
2028 return 1;
2029 }
2030 }
2031 return 0;
2032 }
2033
2034 int has_object_pack(const struct object_id *oid)
2035 {
2036 struct pack_entry e;
2037 return find_pack_entry(the_repository, oid, &e);
2038 }
2039
2040 int has_pack_index(const unsigned char *sha1)
2041 {
2042 struct stat st;
2043 if (stat(sha1_pack_index_name(sha1), &st))
2044 return 0;
2045 return 1;
2046 }
2047
2048 int for_each_object_in_pack(struct packed_git *p,
2049 each_packed_object_fn cb, void *data,
2050 enum for_each_object_flags flags)
2051 {
2052 uint32_t i;
2053 int r = 0;
2054
2055 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2056 if (load_pack_revindex(p))
2057 return -1;
2058 }
2059
2060 for (i = 0; i < p->num_objects; i++) {
2061 uint32_t pos;
2062 struct object_id oid;
2063
2064 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2065 pos = p->revindex[i].nr;
2066 else
2067 pos = i;
2068
2069 if (!nth_packed_object_oid(&oid, p, pos))
2070 return error("unable to get sha1 of object %u in %s",
2071 pos, p->pack_name);
2072
2073 r = cb(&oid, p, pos, data);
2074 if (r)
2075 break;
2076 }
2077 return r;
2078 }
2079
2080 int for_each_packed_object(each_packed_object_fn cb, void *data,
2081 enum for_each_object_flags flags)
2082 {
2083 struct packed_git *p;
2084 int r = 0;
2085 int pack_errors = 0;
2086
2087 prepare_packed_git(the_repository);
2088 for (p = get_all_packs(the_repository); p; p = p->next) {
2089 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2090 continue;
2091 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2092 !p->pack_promisor)
2093 continue;
2094 if (open_pack_index(p)) {
2095 pack_errors = 1;
2096 continue;
2097 }
2098 r = for_each_object_in_pack(p, cb, data, flags);
2099 if (r)
2100 break;
2101 }
2102 return r ? r : pack_errors;
2103 }
2104
2105 static int add_promisor_object(const struct object_id *oid,
2106 struct packed_git *pack,
2107 uint32_t pos,
2108 void *set_)
2109 {
2110 struct oidset *set = set_;
2111 struct object *obj = parse_object(the_repository, oid);
2112 if (!obj)
2113 return 1;
2114
2115 oidset_insert(set, oid);
2116
2117 /*
2118 * If this is a tree, commit, or tag, the objects it refers
2119 * to are also promisor objects. (Blobs refer to no objects->)
2120 */
2121 if (obj->type == OBJ_TREE) {
2122 struct tree *tree = (struct tree *)obj;
2123 struct tree_desc desc;
2124 struct name_entry entry;
2125 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2126 /*
2127 * Error messages are given when packs are
2128 * verified, so do not print any here.
2129 */
2130 return 0;
2131 while (tree_entry_gently(&desc, &entry))
2132 oidset_insert(set, &entry.oid);
2133 } else if (obj->type == OBJ_COMMIT) {
2134 struct commit *commit = (struct commit *) obj;
2135 struct commit_list *parents = commit->parents;
2136
2137 oidset_insert(set, get_commit_tree_oid(commit));
2138 for (; parents; parents = parents->next)
2139 oidset_insert(set, &parents->item->object.oid);
2140 } else if (obj->type == OBJ_TAG) {
2141 struct tag *tag = (struct tag *) obj;
2142 oidset_insert(set, &tag->tagged->oid);
2143 }
2144 return 0;
2145 }
2146
2147 int is_promisor_object(const struct object_id *oid)
2148 {
2149 static struct oidset promisor_objects;
2150 static int promisor_objects_prepared;
2151
2152 if (!promisor_objects_prepared) {
2153 if (repository_format_partial_clone) {
2154 for_each_packed_object(add_promisor_object,
2155 &promisor_objects,
2156 FOR_EACH_OBJECT_PROMISOR_ONLY);
2157 }
2158 promisor_objects_prepared = 1;
2159 }
2160 return oidset_contains(&promisor_objects, oid);
2161 }