?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mingw: unset PERL5LIB by default
[git/git-svn.git] / packfile.c
blobf2850a00b58cccfec57335fb8e0500b7b1cba093
1 #include "cache.h"
2 #include "list.h"
3 #include "pack.h"
4 #include "repository.h"
5 #include "dir.h"
6 #include "mergesort.h"
7 #include "packfile.h"
8 #include "delta.h"
9 #include "list.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
12 #include "commit.h"
13 #include "object.h"
14 #include "tag.h"
15 #include "tree-walk.h"
16 #include "tree.h"
17 #include "object-store.h"
18 #include "midx.h"
19
20 char *odb_pack_name(struct strbuf *buf,
21 const unsigned char *sha1,
22 const char *ext)
23 {
24 strbuf_reset(buf);
25 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1), ext);
27 return buf->buf;
28 }
29
30 char *sha1_pack_name(const unsigned char *sha1)
31 {
32 static struct strbuf buf = STRBUF_INIT;
33 return odb_pack_name(&buf, sha1, "pack");
34 }
35
36 char *sha1_pack_index_name(const unsigned char *sha1)
37 {
38 static struct strbuf buf = STRBUF_INIT;
39 return odb_pack_name(&buf, sha1, "idx");
40 }
41
42 static unsigned int pack_used_ctr;
43 static unsigned int pack_mmap_calls;
44 static unsigned int peak_pack_open_windows;
45 static unsigned int pack_open_windows;
46 static unsigned int pack_open_fds;
47 static unsigned int pack_max_fds;
48 static size_t peak_pack_mapped;
49 static size_t pack_mapped;
50
51 #define SZ_FMT PRIuMAX
52 static inline uintmax_t sz_fmt(size_t s) { return s; }
53
54 void pack_report(void)
55 {
56 fprintf(stderr,
57 "pack_report: getpagesize() = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size),
62 sz_fmt(packed_git_limit));
63 fprintf(stderr,
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT " / %10" SZ_FMT "\n",
69 pack_used_ctr,
70 pack_mmap_calls,
71 pack_open_windows, peak_pack_open_windows,
72 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
73 }
74
75 /*
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
78 * success.
79 */
80 static int check_packed_git_idx(const char *path, struct packed_git *p)
81 {
82 void *idx_map;
83 size_t idx_size;
84 int fd = git_open(path), ret;
85 struct stat st;
86 const unsigned int hashsz = the_hash_algo->rawsz;
87
88 if (fd < 0)
89 return -1;
90 if (fstat(fd, &st)) {
91 close(fd);
92 return -1;
93 }
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + hashsz + hashsz) {
96 close(fd);
97 return error("index file %s is too small", path);
98 }
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
101
102 ret = load_idx(path, hashsz, idx_map, idx_size, p);
103
104 if (ret)
105 munmap(idx_map, idx_size);
106
107 return ret;
108 }
109
110 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
111 size_t idx_size, struct packed_git *p)
112 {
113 struct pack_idx_header *hdr = idx_map;
114 uint32_t version, nr, i, *index;
115
116 if (idx_size < 4 * 256 + hashsz + hashsz)
117 return error("index file %s is too small", path);
118 if (idx_map == NULL)
119 return error("empty data");
120
121 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
122 version = ntohl(hdr->idx_version);
123 if (version < 2 || version > 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
127 path, version);
128 } else
129 version = 1;
130
131 nr = 0;
132 index = idx_map;
133 if (version > 1)
134 index += 2; /* skip index header */
135 for (i = 0; i < 256; i++) {
136 uint32_t n = ntohl(index[i]);
137 if (n < nr)
138 return error("non-monotonic index %s", path);
139 nr = n;
140 }
141
142 if (version == 1) {
143 /*
144 * Total size:
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
148 * - file checksum
149 */
150 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
151 return error("wrong index v1 file size in %s", path);
152 } else if (version == 2) {
153 /*
154 * Minimum size:
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
161 * - file checksum
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
165 */
166 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
167 unsigned long max_size = min_size;
168 if (nr)
169 max_size += (nr - 1)*8;
170 if (idx_size < min_size || idx_size > max_size)
171 return error("wrong index v2 file size in %s", path);
172 if (idx_size != min_size &&
173 /*
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
177 */
178 (sizeof(off_t) <= 4))
179 return error("pack too large for current definition of off_t in %s", path);
180 }
181
182 p->index_version = version;
183 p->index_data = idx_map;
184 p->index_size = idx_size;
185 p->num_objects = nr;
186 return 0;
187 }
188
189 int open_pack_index(struct packed_git *p)
190 {
191 char *idx_name;
192 size_t len;
193 int ret;
194
195 if (p->index_data)
196 return 0;
197
198 if (!strip_suffix(p->pack_name, ".pack", &len))
199 BUG("pack_name does not end in .pack");
200 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
201 ret = check_packed_git_idx(idx_name, p);
202 free(idx_name);
203 return ret;
204 }
205
206 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
207 {
208 const uint32_t *level1_ofs = p->index_data;
209
210 if (!level1_ofs) {
211 if (open_pack_index(p))
212 return 0;
213 level1_ofs = p->index_data;
214 }
215
216 if (p->index_version > 1) {
217 level1_ofs += 2;
218 }
219
220 return ntohl(level1_ofs[value]);
221 }
222
223 static struct packed_git *alloc_packed_git(int extra)
224 {
225 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
226 memset(p, 0, sizeof(*p));
227 p->pack_fd = -1;
228 return p;
229 }
230
231 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
232 {
233 const char *path = sha1_pack_name(sha1);
234 size_t alloc = st_add(strlen(path), 1);
235 struct packed_git *p = alloc_packed_git(alloc);
236
237 memcpy(p->pack_name, path, alloc); /* includes NUL */
238 hashcpy(p->sha1, sha1);
239 if (check_packed_git_idx(idx_path, p)) {
240 free(p);
241 return NULL;
242 }
243
244 return p;
245 }
246
247 static void scan_windows(struct packed_git *p,
248 struct packed_git **lru_p,
249 struct pack_window **lru_w,
250 struct pack_window **lru_l)
251 {
252 struct pack_window *w, *w_l;
253
254 for (w_l = NULL, w = p->windows; w; w = w->next) {
255 if (!w->inuse_cnt) {
256 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
257 *lru_p = p;
258 *lru_w = w;
259 *lru_l = w_l;
260 }
261 }
262 w_l = w;
263 }
264 }
265
266 static int unuse_one_window(struct packed_git *current)
267 {
268 struct packed_git *p, *lru_p = NULL;
269 struct pack_window *lru_w = NULL, *lru_l = NULL;
270
271 if (current)
272 scan_windows(current, &lru_p, &lru_w, &lru_l);
273 for (p = the_repository->objects->packed_git; p; p = p->next)
274 scan_windows(p, &lru_p, &lru_w, &lru_l);
275 if (lru_p) {
276 munmap(lru_w->base, lru_w->len);
277 pack_mapped -= lru_w->len;
278 if (lru_l)
279 lru_l->next = lru_w->next;
280 else
281 lru_p->windows = lru_w->next;
282 free(lru_w);
283 pack_open_windows--;
284 return 1;
285 }
286 return 0;
287 }
288
289 void release_pack_memory(size_t need)
290 {
291 size_t cur = pack_mapped;
292 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
293 ; /* nothing */
294 }
295
296 void close_pack_windows(struct packed_git *p)
297 {
298 while (p->windows) {
299 struct pack_window *w = p->windows;
300
301 if (w->inuse_cnt)
302 die("pack '%s' still has open windows to it",
303 p->pack_name);
304 munmap(w->base, w->len);
305 pack_mapped -= w->len;
306 pack_open_windows--;
307 p->windows = w->next;
308 free(w);
309 }
310 }
311
312 static int close_pack_fd(struct packed_git *p)
313 {
314 if (p->pack_fd < 0)
315 return 0;
316
317 close(p->pack_fd);
318 pack_open_fds--;
319 p->pack_fd = -1;
320
321 return 1;
322 }
323
324 void close_pack_index(struct packed_git *p)
325 {
326 if (p->index_data) {
327 munmap((void *)p->index_data, p->index_size);
328 p->index_data = NULL;
329 }
330 }
331
332 void close_pack(struct packed_git *p)
333 {
334 close_pack_windows(p);
335 close_pack_fd(p);
336 close_pack_index(p);
337 }
338
339 void close_all_packs(struct raw_object_store *o)
340 {
341 struct packed_git *p;
342
343 for (p = o->packed_git; p; p = p->next)
344 if (p->do_not_close)
345 BUG("want to close pack marked 'do-not-close'");
346 else
347 close_pack(p);
348 }
349
350 /*
351 * The LRU pack is the one with the oldest MRU window, preferring packs
352 * with no used windows, or the oldest mtime if it has no windows allocated.
353 */
354 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
355 {
356 struct pack_window *w, *this_mru_w;
357 int has_windows_inuse = 0;
358
359 /*
360 * Reject this pack if it has windows and the previously selected
361 * one does not. If this pack does not have windows, reject
362 * it if the pack file is newer than the previously selected one.
363 */
364 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
365 return;
366
367 for (w = this_mru_w = p->windows; w; w = w->next) {
368 /*
369 * Reject this pack if any of its windows are in use,
370 * but the previously selected pack did not have any
371 * inuse windows. Otherwise, record that this pack
372 * has windows in use.
373 */
374 if (w->inuse_cnt) {
375 if (*accept_windows_inuse)
376 has_windows_inuse = 1;
377 else
378 return;
379 }
380
381 if (w->last_used > this_mru_w->last_used)
382 this_mru_w = w;
383
384 /*
385 * Reject this pack if it has windows that have been
386 * used more recently than the previously selected pack.
387 * If the previously selected pack had windows inuse and
388 * we have not encountered a window in this pack that is
389 * inuse, skip this check since we prefer a pack with no
390 * inuse windows to one that has inuse windows.
391 */
392 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
393 this_mru_w->last_used > (*mru_w)->last_used)
394 return;
395 }
396
397 /*
398 * Select this pack.
399 */
400 *mru_w = this_mru_w;
401 *lru_p = p;
402 *accept_windows_inuse = has_windows_inuse;
403 }
404
405 static int close_one_pack(void)
406 {
407 struct packed_git *p, *lru_p = NULL;
408 struct pack_window *mru_w = NULL;
409 int accept_windows_inuse = 1;
410
411 for (p = the_repository->objects->packed_git; p; p = p->next) {
412 if (p->pack_fd == -1)
413 continue;
414 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
415 }
416
417 if (lru_p)
418 return close_pack_fd(lru_p);
419
420 return 0;
421 }
422
423 static unsigned int get_max_fd_limit(void)
424 {
425 #ifdef RLIMIT_NOFILE
426 {
427 struct rlimit lim;
428
429 if (!getrlimit(RLIMIT_NOFILE, &lim))
430 return lim.rlim_cur;
431 }
432 #endif
433
434 #ifdef _SC_OPEN_MAX
435 {
436 long open_max = sysconf(_SC_OPEN_MAX);
437 if (0 < open_max)
438 return open_max;
439 /*
440 * Otherwise, we got -1 for one of the two
441 * reasons:
442 *
443 * (1) sysconf() did not understand _SC_OPEN_MAX
444 * and signaled an error with -1; or
445 * (2) sysconf() said there is no limit.
446 *
447 * We _could_ clear errno before calling sysconf() to
448 * tell these two cases apart and return a huge number
449 * in the latter case to let the caller cap it to a
450 * value that is not so selfish, but letting the
451 * fallback OPEN_MAX codepath take care of these cases
452 * is a lot simpler.
453 */
454 }
455 #endif
456
457 #ifdef OPEN_MAX
458 return OPEN_MAX;
459 #else
460 return 1; /* see the caller ;-) */
461 #endif
462 }
463
464 /*
465 * Do not call this directly as this leaks p->pack_fd on error return;
466 * call open_packed_git() instead.
467 */
468 static int open_packed_git_1(struct packed_git *p)
469 {
470 struct stat st;
471 struct pack_header hdr;
472 unsigned char hash[GIT_MAX_RAWSZ];
473 unsigned char *idx_hash;
474 long fd_flag;
475 ssize_t read_result;
476 const unsigned hashsz = the_hash_algo->rawsz;
477
478 if (!p->index_data) {
479 struct multi_pack_index *m;
480 const char *pack_name = strrchr(p->pack_name, '/');
481
482 for (m = the_repository->objects->multi_pack_index;
483 m; m = m->next) {
484 if (midx_contains_pack(m, pack_name))
485 break;
486 }
487
488 if (!m && open_pack_index(p))
489 return error("packfile %s index unavailable", p->pack_name);
490 }
491
492 if (!pack_max_fds) {
493 unsigned int max_fds = get_max_fd_limit();
494
495 /* Save 3 for stdin/stdout/stderr, 22 for work */
496 if (25 < max_fds)
497 pack_max_fds = max_fds - 25;
498 else
499 pack_max_fds = 1;
500 }
501
502 while (pack_max_fds <= pack_open_fds && close_one_pack())
503 ; /* nothing */
504
505 p->pack_fd = git_open(p->pack_name);
506 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
507 return -1;
508 pack_open_fds++;
509
510 /* If we created the struct before we had the pack we lack size. */
511 if (!p->pack_size) {
512 if (!S_ISREG(st.st_mode))
513 return error("packfile %s not a regular file", p->pack_name);
514 p->pack_size = st.st_size;
515 } else if (p->pack_size != st.st_size)
516 return error("packfile %s size changed", p->pack_name);
517
518 /* We leave these file descriptors open with sliding mmap;
519 * there is no point keeping them open across exec(), though.
520 */
521 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
522 if (fd_flag < 0)
523 return error("cannot determine file descriptor flags");
524 fd_flag |= FD_CLOEXEC;
525 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
526 return error("cannot set FD_CLOEXEC");
527
528 /* Verify we recognize this pack file format. */
529 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
530 if (read_result < 0)
531 return error_errno("error reading from %s", p->pack_name);
532 if (read_result != sizeof(hdr))
533 return error("file %s is far too short to be a packfile", p->pack_name);
534 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
535 return error("file %s is not a GIT packfile", p->pack_name);
536 if (!pack_version_ok(hdr.hdr_version))
537 return error("packfile %s is version %"PRIu32" and not"
538 " supported (try upgrading GIT to a newer version)",
539 p->pack_name, ntohl(hdr.hdr_version));
540
541 /* Skip index checking if in multi-pack-index */
542 if (!p->index_data)
543 return 0;
544
545 /* Verify the pack matches its index. */
546 if (p->num_objects != ntohl(hdr.hdr_entries))
547 return error("packfile %s claims to have %"PRIu32" objects"
548 " while index indicates %"PRIu32" objects",
549 p->pack_name, ntohl(hdr.hdr_entries),
550 p->num_objects);
551 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
552 return error("end of packfile %s is unavailable", p->pack_name);
553 read_result = read_in_full(p->pack_fd, hash, hashsz);
554 if (read_result < 0)
555 return error_errno("error reading from %s", p->pack_name);
556 if (read_result != hashsz)
557 return error("packfile %s signature is unavailable", p->pack_name);
558 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
559 if (!hasheq(hash, idx_hash))
560 return error("packfile %s does not match index", p->pack_name);
561 return 0;
562 }
563
564 static int open_packed_git(struct packed_git *p)
565 {
566 if (!open_packed_git_1(p))
567 return 0;
568 close_pack_fd(p);
569 return -1;
570 }
571
572 static int in_window(struct pack_window *win, off_t offset)
573 {
574 /* We must promise at least one full hash after the
575 * offset is available from this window, otherwise the offset
576 * is not actually in this window and a different window (which
577 * has that one hash excess) must be used. This is to support
578 * the object header and delta base parsing routines below.
579 */
580 off_t win_off = win->offset;
581 return win_off <= offset
582 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
583 }
584
585 unsigned char *use_pack(struct packed_git *p,
586 struct pack_window **w_cursor,
587 off_t offset,
588 unsigned long *left)
589 {
590 struct pack_window *win = *w_cursor;
591
592 /* Since packfiles end in a hash of their content and it's
593 * pointless to ask for an offset into the middle of that
594 * hash, and the in_window function above wouldn't match
595 * don't allow an offset too close to the end of the file.
596 */
597 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
598 die("packfile %s cannot be accessed", p->pack_name);
599 if (offset > (p->pack_size - the_hash_algo->rawsz))
600 die("offset beyond end of packfile (truncated pack?)");
601 if (offset < 0)
602 die(_("offset before end of packfile (broken .idx?)"));
603
604 if (!win || !in_window(win, offset)) {
605 if (win)
606 win->inuse_cnt--;
607 for (win = p->windows; win; win = win->next) {
608 if (in_window(win, offset))
609 break;
610 }
611 if (!win) {
612 size_t window_align = packed_git_window_size / 2;
613 off_t len;
614
615 if (p->pack_fd == -1 && open_packed_git(p))
616 die("packfile %s cannot be accessed", p->pack_name);
617
618 win = xcalloc(1, sizeof(*win));
619 win->offset = (offset / window_align) * window_align;
620 len = p->pack_size - win->offset;
621 if (len > packed_git_window_size)
622 len = packed_git_window_size;
623 win->len = (size_t)len;
624 pack_mapped += win->len;
625 while (packed_git_limit < pack_mapped
626 && unuse_one_window(p))
627 ; /* nothing */
628 win->base = xmmap(NULL, win->len,
629 PROT_READ, MAP_PRIVATE,
630 p->pack_fd, win->offset);
631 if (win->base == MAP_FAILED)
632 die_errno("packfile %s cannot be mapped",
633 p->pack_name);
634 if (!win->offset && win->len == p->pack_size
635 && !p->do_not_close)
636 close_pack_fd(p);
637 pack_mmap_calls++;
638 pack_open_windows++;
639 if (pack_mapped > peak_pack_mapped)
640 peak_pack_mapped = pack_mapped;
641 if (pack_open_windows > peak_pack_open_windows)
642 peak_pack_open_windows = pack_open_windows;
643 win->next = p->windows;
644 p->windows = win;
645 }
646 }
647 if (win != *w_cursor) {
648 win->last_used = pack_used_ctr++;
649 win->inuse_cnt++;
650 *w_cursor = win;
651 }
652 offset -= win->offset;
653 if (left)
654 *left = win->len - xsize_t(offset);
655 return win->base + offset;
656 }
657
658 void unuse_pack(struct pack_window **w_cursor)
659 {
660 struct pack_window *w = *w_cursor;
661 if (w) {
662 w->inuse_cnt--;
663 *w_cursor = NULL;
664 }
665 }
666
667 static void try_to_free_pack_memory(size_t size)
668 {
669 release_pack_memory(size);
670 }
671
672 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
673 {
674 static int have_set_try_to_free_routine;
675 struct stat st;
676 size_t alloc;
677 struct packed_git *p;
678
679 if (!have_set_try_to_free_routine) {
680 have_set_try_to_free_routine = 1;
681 set_try_to_free_routine(try_to_free_pack_memory);
682 }
683
684 /*
685 * Make sure a corresponding .pack file exists and that
686 * the index looks sane.
687 */
688 if (!strip_suffix_mem(path, &path_len, ".idx"))
689 return NULL;
690
691 /*
692 * ".promisor" is long enough to hold any suffix we're adding (and
693 * the use xsnprintf double-checks that)
694 */
695 alloc = st_add3(path_len, strlen(".promisor"), 1);
696 p = alloc_packed_git(alloc);
697 memcpy(p->pack_name, path, path_len);
698
699 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
700 if (!access(p->pack_name, F_OK))
701 p->pack_keep = 1;
702
703 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
704 if (!access(p->pack_name, F_OK))
705 p->pack_promisor = 1;
706
707 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
708 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
709 free(p);
710 return NULL;
711 }
712
713 /* ok, it looks sane as far as we can check without
714 * actually mapping the pack file.
715 */
716 p->pack_size = st.st_size;
717 p->pack_local = local;
718 p->mtime = st.st_mtime;
719 if (path_len < the_hash_algo->hexsz ||
720 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->sha1))
721 hashclr(p->sha1);
722 return p;
723 }
724
725 void install_packed_git(struct repository *r, struct packed_git *pack)
726 {
727 if (pack->pack_fd != -1)
728 pack_open_fds++;
729
730 pack->next = r->objects->packed_git;
731 r->objects->packed_git = pack;
732 }
733
734 void (*report_garbage)(unsigned seen_bits, const char *path);
735
736 static void report_helper(const struct string_list *list,
737 int seen_bits, int first, int last)
738 {
739 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
740 return;
741
742 for (; first < last; first++)
743 report_garbage(seen_bits, list->items[first].string);
744 }
745
746 static void report_pack_garbage(struct string_list *list)
747 {
748 int i, baselen = -1, first = 0, seen_bits = 0;
749
750 if (!report_garbage)
751 return;
752
753 string_list_sort(list);
754
755 for (i = 0; i < list->nr; i++) {
756 const char *path = list->items[i].string;
757 if (baselen != -1 &&
758 strncmp(path, list->items[first].string, baselen)) {
759 report_helper(list, seen_bits, first, i);
760 baselen = -1;
761 seen_bits = 0;
762 }
763 if (baselen == -1) {
764 const char *dot = strrchr(path, '.');
765 if (!dot) {
766 report_garbage(PACKDIR_FILE_GARBAGE, path);
767 continue;
768 }
769 baselen = dot - path + 1;
770 first = i;
771 }
772 if (!strcmp(path + baselen, "pack"))
773 seen_bits |= 1;
774 else if (!strcmp(path + baselen, "idx"))
775 seen_bits |= 2;
776 }
777 report_helper(list, seen_bits, first, list->nr);
778 }
779
780 void for_each_file_in_pack_dir(const char *objdir,
781 each_file_in_pack_dir_fn fn,
782 void *data)
783 {
784 struct strbuf path = STRBUF_INIT;
785 size_t dirnamelen;
786 DIR *dir;
787 struct dirent *de;
788
789 strbuf_addstr(&path, objdir);
790 strbuf_addstr(&path, "/pack");
791 dir = opendir(path.buf);
792 if (!dir) {
793 if (errno != ENOENT)
794 error_errno("unable to open object pack directory: %s",
795 path.buf);
796 strbuf_release(&path);
797 return;
798 }
799 strbuf_addch(&path, '/');
800 dirnamelen = path.len;
801 while ((de = readdir(dir)) != NULL) {
802 if (is_dot_or_dotdot(de->d_name))
803 continue;
804
805 strbuf_setlen(&path, dirnamelen);
806 strbuf_addstr(&path, de->d_name);
807
808 fn(path.buf, path.len, de->d_name, data);
809 }
810
811 closedir(dir);
812 strbuf_release(&path);
813 }
814
815 struct prepare_pack_data {
816 struct repository *r;
817 struct string_list *garbage;
818 int local;
819 struct multi_pack_index *m;
820 };
821
822 static void prepare_pack(const char *full_name, size_t full_name_len,
823 const char *file_name, void *_data)
824 {
825 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
826 struct packed_git *p;
827 size_t base_len = full_name_len;
828
829 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
830 !(data->m && midx_contains_pack(data->m, file_name))) {
831 /* Don't reopen a pack we already have. */
832 for (p = data->r->objects->packed_git; p; p = p->next) {
833 size_t len;
834 if (strip_suffix(p->pack_name, ".pack", &len) &&
835 len == base_len &&
836 !memcmp(p->pack_name, full_name, len))
837 break;
838 }
839
840 if (!p) {
841 p = add_packed_git(full_name, full_name_len, data->local);
842 if (p)
843 install_packed_git(data->r, p);
844 }
845 }
846
847 if (!report_garbage)
848 return;
849
850 if (!strcmp(file_name, "multi-pack-index"))
851 return;
852 if (ends_with(file_name, ".idx") ||
853 ends_with(file_name, ".pack") ||
854 ends_with(file_name, ".bitmap") ||
855 ends_with(file_name, ".keep") ||
856 ends_with(file_name, ".promisor"))
857 string_list_append(data->garbage, full_name);
858 else
859 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
860 }
861
862 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
863 {
864 struct prepare_pack_data data;
865 struct string_list garbage = STRING_LIST_INIT_DUP;
866
867 data.m = r->objects->multi_pack_index;
868
869 /* look for the multi-pack-index for this object directory */
870 while (data.m && strcmp(data.m->object_dir, objdir))
871 data.m = data.m->next;
872
873 data.r = r;
874 data.garbage = &garbage;
875 data.local = local;
876
877 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
878
879 report_pack_garbage(data.garbage);
880 string_list_clear(data.garbage, 0);
881 }
882
883 static void prepare_packed_git(struct repository *r);
884 /*
885 * Give a fast, rough count of the number of objects in the repository. This
886 * ignores loose objects completely. If you have a lot of them, then either
887 * you should repack because your performance will be awful, or they are
888 * all unreachable objects about to be pruned, in which case they're not really
889 * interesting as a measure of repo size in the first place.
890 */
891 unsigned long approximate_object_count(void)
892 {
893 if (!the_repository->objects->approximate_object_count_valid) {
894 unsigned long count;
895 struct multi_pack_index *m;
896 struct packed_git *p;
897
898 prepare_packed_git(the_repository);
899 count = 0;
900 for (m = get_multi_pack_index(the_repository); m; m = m->next)
901 count += m->num_objects;
902 for (p = the_repository->objects->packed_git; p; p = p->next) {
903 if (open_pack_index(p))
904 continue;
905 count += p->num_objects;
906 }
907 the_repository->objects->approximate_object_count = count;
908 }
909 return the_repository->objects->approximate_object_count;
910 }
911
912 static void *get_next_packed_git(const void *p)
913 {
914 return ((const struct packed_git *)p)->next;
915 }
916
917 static void set_next_packed_git(void *p, void *next)
918 {
919 ((struct packed_git *)p)->next = next;
920 }
921
922 static int sort_pack(const void *a_, const void *b_)
923 {
924 const struct packed_git *a = a_;
925 const struct packed_git *b = b_;
926 int st;
927
928 /*
929 * Local packs tend to contain objects specific to our
930 * variant of the project than remote ones. In addition,
931 * remote ones could be on a network mounted filesystem.
932 * Favor local ones for these reasons.
933 */
934 st = a->pack_local - b->pack_local;
935 if (st)
936 return -st;
937
938 /*
939 * Younger packs tend to contain more recent objects,
940 * and more recent objects tend to get accessed more
941 * often.
942 */
943 if (a->mtime < b->mtime)
944 return 1;
945 else if (a->mtime == b->mtime)
946 return 0;
947 return -1;
948 }
949
950 static void rearrange_packed_git(struct repository *r)
951 {
952 r->objects->packed_git = llist_mergesort(
953 r->objects->packed_git, get_next_packed_git,
954 set_next_packed_git, sort_pack);
955 }
956
957 static void prepare_packed_git_mru(struct repository *r)
958 {
959 struct packed_git *p;
960
961 INIT_LIST_HEAD(&r->objects->packed_git_mru);
962
963 for (p = r->objects->packed_git; p; p = p->next)
964 list_add_tail(&p->mru, &r->objects->packed_git_mru);
965 }
966
967 static void prepare_packed_git(struct repository *r)
968 {
969 struct alternate_object_database *alt;
970
971 if (r->objects->packed_git_initialized)
972 return;
973 prepare_multi_pack_index_one(r, r->objects->objectdir, 1);
974 prepare_packed_git_one(r, r->objects->objectdir, 1);
975 prepare_alt_odb(r);
976 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
977 prepare_multi_pack_index_one(r, alt->path, 0);
978 prepare_packed_git_one(r, alt->path, 0);
979 }
980 rearrange_packed_git(r);
981
982 r->objects->all_packs = NULL;
983
984 prepare_packed_git_mru(r);
985 r->objects->packed_git_initialized = 1;
986 }
987
988 void reprepare_packed_git(struct repository *r)
989 {
990 r->objects->approximate_object_count_valid = 0;
991 r->objects->packed_git_initialized = 0;
992 prepare_packed_git(r);
993 }
994
995 struct packed_git *get_packed_git(struct repository *r)
996 {
997 prepare_packed_git(r);
998 return r->objects->packed_git;
999 }
1000
1001 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1002 {
1003 prepare_packed_git(r);
1004 return r->objects->multi_pack_index;
1005 }
1006
1007 struct packed_git *get_all_packs(struct repository *r)
1008 {
1009 prepare_packed_git(r);
1010
1011 if (!r->objects->all_packs) {
1012 struct packed_git *p = r->objects->packed_git;
1013 struct multi_pack_index *m;
1014
1015 for (m = r->objects->multi_pack_index; m; m = m->next) {
1016 uint32_t i;
1017 for (i = 0; i < m->num_packs; i++) {
1018 if (!prepare_midx_pack(m, i)) {
1019 m->packs[i]->next = p;
1020 p = m->packs[i];
1021 }
1022 }
1023 }
1024
1025 r->objects->all_packs = p;
1026 }
1027
1028 return r->objects->all_packs;
1029 }
1030
1031 struct list_head *get_packed_git_mru(struct repository *r)
1032 {
1033 prepare_packed_git(r);
1034 return &r->objects->packed_git_mru;
1035 }
1036
1037 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1038 unsigned long len, enum object_type *type, unsigned long *sizep)
1039 {
1040 unsigned shift;
1041 unsigned long size, c;
1042 unsigned long used = 0;
1043
1044 c = buf[used++];
1045 *type = (c >> 4) & 7;
1046 size = c & 15;
1047 shift = 4;
1048 while (c & 0x80) {
1049 if (len <= used || bitsizeof(long) <= shift) {
1050 error("bad object header");
1051 size = used = 0;
1052 break;
1053 }
1054 c = buf[used++];
1055 size += (c & 0x7f) << shift;
1056 shift += 7;
1057 }
1058 *sizep = size;
1059 return used;
1060 }
1061
1062 unsigned long get_size_from_delta(struct packed_git *p,
1063 struct pack_window **w_curs,
1064 off_t curpos)
1065 {
1066 const unsigned char *data;
1067 unsigned char delta_head[20], *in;
1068 git_zstream stream;
1069 int st;
1070
1071 memset(&stream, 0, sizeof(stream));
1072 stream.next_out = delta_head;
1073 stream.avail_out = sizeof(delta_head);
1074
1075 git_inflate_init(&stream);
1076 do {
1077 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1078 stream.next_in = in;
1079 st = git_inflate(&stream, Z_FINISH);
1080 curpos += stream.next_in - in;
1081 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1082 stream.total_out < sizeof(delta_head));
1083 git_inflate_end(&stream);
1084 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1085 error("delta data unpack-initial failed");
1086 return 0;
1087 }
1088
1089 /* Examine the initial part of the delta to figure out
1090 * the result size.
1091 */
1092 data = delta_head;
1093
1094 /* ignore base size */
1095 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1096
1097 /* Read the result size */
1098 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1099 }
1100
1101 int unpack_object_header(struct packed_git *p,
1102 struct pack_window **w_curs,
1103 off_t *curpos,
1104 unsigned long *sizep)
1105 {
1106 unsigned char *base;
1107 unsigned long left;
1108 unsigned long used;
1109 enum object_type type;
1110
1111 /* use_pack() assures us we have [base, base + 20) available
1112 * as a range that we can look at. (Its actually the hash
1113 * size that is assured.) With our object header encoding
1114 * the maximum deflated object size is 2^137, which is just
1115 * insane, so we know won't exceed what we have been given.
1116 */
1117 base = use_pack(p, w_curs, *curpos, &left);
1118 used = unpack_object_header_buffer(base, left, &type, sizep);
1119 if (!used) {
1120 type = OBJ_BAD;
1121 } else
1122 *curpos += used;
1123
1124 return type;
1125 }
1126
1127 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1128 {
1129 unsigned i;
1130 const unsigned hashsz = the_hash_algo->rawsz;
1131 for (i = 0; i < p->num_bad_objects; i++)
1132 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1133 return;
1134 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1135 st_mult(GIT_MAX_RAWSZ,
1136 st_add(p->num_bad_objects, 1)));
1137 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1138 p->num_bad_objects++;
1139 }
1140
1141 const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1142 {
1143 struct packed_git *p;
1144 unsigned i;
1145
1146 for (p = the_repository->objects->packed_git; p; p = p->next)
1147 for (i = 0; i < p->num_bad_objects; i++)
1148 if (hasheq(sha1,
1149 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1150 return p;
1151 return NULL;
1152 }
1153
1154 static off_t get_delta_base(struct packed_git *p,
1155 struct pack_window **w_curs,
1156 off_t *curpos,
1157 enum object_type type,
1158 off_t delta_obj_offset)
1159 {
1160 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1161 off_t base_offset;
1162
1163 /* use_pack() assured us we have [base_info, base_info + 20)
1164 * as a range that we can look at without walking off the
1165 * end of the mapped window. Its actually the hash size
1166 * that is assured. An OFS_DELTA longer than the hash size
1167 * is stupid, as then a REF_DELTA would be smaller to store.
1168 */
1169 if (type == OBJ_OFS_DELTA) {
1170 unsigned used = 0;
1171 unsigned char c = base_info[used++];
1172 base_offset = c & 127;
1173 while (c & 128) {
1174 base_offset += 1;
1175 if (!base_offset || MSB(base_offset, 7))
1176 return 0; /* overflow */
1177 c = base_info[used++];
1178 base_offset = (base_offset << 7) + (c & 127);
1179 }
1180 base_offset = delta_obj_offset - base_offset;
1181 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1182 return 0; /* out of bound */
1183 *curpos += used;
1184 } else if (type == OBJ_REF_DELTA) {
1185 /* The base entry _must_ be in the same pack */
1186 base_offset = find_pack_entry_one(base_info, p);
1187 *curpos += the_hash_algo->rawsz;
1188 } else
1189 die("I am totally screwed");
1190 return base_offset;
1191 }
1192
1193 /*
1194 * Like get_delta_base above, but we return the sha1 instead of the pack
1195 * offset. This means it is cheaper for REF deltas (we do not have to do
1196 * the final object lookup), but more expensive for OFS deltas (we
1197 * have to load the revidx to convert the offset back into a sha1).
1198 */
1199 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1200 struct pack_window **w_curs,
1201 off_t curpos,
1202 enum object_type type,
1203 off_t delta_obj_offset)
1204 {
1205 if (type == OBJ_REF_DELTA) {
1206 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1207 return base;
1208 } else if (type == OBJ_OFS_DELTA) {
1209 struct revindex_entry *revidx;
1210 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1211 type, delta_obj_offset);
1212
1213 if (!base_offset)
1214 return NULL;
1215
1216 revidx = find_pack_revindex(p, base_offset);
1217 if (!revidx)
1218 return NULL;
1219
1220 return nth_packed_object_sha1(p, revidx->nr);
1221 } else
1222 return NULL;
1223 }
1224
1225 static int retry_bad_packed_offset(struct repository *r,
1226 struct packed_git *p,
1227 off_t obj_offset)
1228 {
1229 int type;
1230 struct revindex_entry *revidx;
1231 struct object_id oid;
1232 revidx = find_pack_revindex(p, obj_offset);
1233 if (!revidx)
1234 return OBJ_BAD;
1235 nth_packed_object_oid(&oid, p, revidx->nr);
1236 mark_bad_packed_object(p, oid.hash);
1237 type = oid_object_info(r, &oid, NULL);
1238 if (type <= OBJ_NONE)
1239 return OBJ_BAD;
1240 return type;
1241 }
1242
1243 #define POI_STACK_PREALLOC 64
1244
1245 static enum object_type packed_to_object_type(struct repository *r,
1246 struct packed_git *p,
1247 off_t obj_offset,
1248 enum object_type type,
1249 struct pack_window **w_curs,
1250 off_t curpos)
1251 {
1252 off_t small_poi_stack[POI_STACK_PREALLOC];
1253 off_t *poi_stack = small_poi_stack;
1254 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1255
1256 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1257 off_t base_offset;
1258 unsigned long size;
1259 /* Push the object we're going to leave behind */
1260 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1261 poi_stack_alloc = alloc_nr(poi_stack_nr);
1262 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1263 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1264 } else {
1265 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1266 }
1267 poi_stack[poi_stack_nr++] = obj_offset;
1268 /* If parsing the base offset fails, just unwind */
1269 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1270 if (!base_offset)
1271 goto unwind;
1272 curpos = obj_offset = base_offset;
1273 type = unpack_object_header(p, w_curs, &curpos, &size);
1274 if (type <= OBJ_NONE) {
1275 /* If getting the base itself fails, we first
1276 * retry the base, otherwise unwind */
1277 type = retry_bad_packed_offset(r, p, base_offset);
1278 if (type > OBJ_NONE)
1279 goto out;
1280 goto unwind;
1281 }
1282 }
1283
1284 switch (type) {
1285 case OBJ_BAD:
1286 case OBJ_COMMIT:
1287 case OBJ_TREE:
1288 case OBJ_BLOB:
1289 case OBJ_TAG:
1290 break;
1291 default:
1292 error("unknown object type %i at offset %"PRIuMAX" in %s",
1293 type, (uintmax_t)obj_offset, p->pack_name);
1294 type = OBJ_BAD;
1295 }
1296
1297 out:
1298 if (poi_stack != small_poi_stack)
1299 free(poi_stack);
1300 return type;
1301
1302 unwind:
1303 while (poi_stack_nr) {
1304 obj_offset = poi_stack[--poi_stack_nr];
1305 type = retry_bad_packed_offset(r, p, obj_offset);
1306 if (type > OBJ_NONE)
1307 goto out;
1308 }
1309 type = OBJ_BAD;
1310 goto out;
1311 }
1312
1313 static struct hashmap delta_base_cache;
1314 static size_t delta_base_cached;
1315
1316 static LIST_HEAD(delta_base_cache_lru);
1317
1318 struct delta_base_cache_key {
1319 struct packed_git *p;
1320 off_t base_offset;
1321 };
1322
1323 struct delta_base_cache_entry {
1324 struct hashmap hash;
1325 struct delta_base_cache_key key;
1326 struct list_head lru;
1327 void *data;
1328 unsigned long size;
1329 enum object_type type;
1330 };
1331
1332 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1333 {
1334 unsigned int hash;
1335
1336 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1337 hash += (hash >> 8) + (hash >> 16);
1338 return hash;
1339 }
1340
1341 static struct delta_base_cache_entry *
1342 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1343 {
1344 struct hashmap_entry entry;
1345 struct delta_base_cache_key key;
1346
1347 if (!delta_base_cache.cmpfn)
1348 return NULL;
1349
1350 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1351 key.p = p;
1352 key.base_offset = base_offset;
1353 return hashmap_get(&delta_base_cache, &entry, &key);
1354 }
1355
1356 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1357 const struct delta_base_cache_key *b)
1358 {
1359 return a->p == b->p && a->base_offset == b->base_offset;
1360 }
1361
1362 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1363 const void *va, const void *vb,
1364 const void *vkey)
1365 {
1366 const struct delta_base_cache_entry *a = va, *b = vb;
1367 const struct delta_base_cache_key *key = vkey;
1368 if (key)
1369 return !delta_base_cache_key_eq(&a->key, key);
1370 else
1371 return !delta_base_cache_key_eq(&a->key, &b->key);
1372 }
1373
1374 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1375 {
1376 return !!get_delta_base_cache_entry(p, base_offset);
1377 }
1378
1379 /*
1380 * Remove the entry from the cache, but do _not_ free the associated
1381 * entry data. The caller takes ownership of the "data" buffer, and
1382 * should copy out any fields it wants before detaching.
1383 */
1384 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1385 {
1386 hashmap_remove(&delta_base_cache, ent, &ent->key);
1387 list_del(&ent->lru);
1388 delta_base_cached -= ent->size;
1389 free(ent);
1390 }
1391
1392 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1393 off_t base_offset, unsigned long *base_size,
1394 enum object_type *type)
1395 {
1396 struct delta_base_cache_entry *ent;
1397
1398 ent = get_delta_base_cache_entry(p, base_offset);
1399 if (!ent)
1400 return unpack_entry(r, p, base_offset, type, base_size);
1401
1402 if (type)
1403 *type = ent->type;
1404 if (base_size)
1405 *base_size = ent->size;
1406 return xmemdupz(ent->data, ent->size);
1407 }
1408
1409 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1410 {
1411 free(ent->data);
1412 detach_delta_base_cache_entry(ent);
1413 }
1414
1415 void clear_delta_base_cache(void)
1416 {
1417 struct list_head *lru, *tmp;
1418 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1419 struct delta_base_cache_entry *entry =
1420 list_entry(lru, struct delta_base_cache_entry, lru);
1421 release_delta_base_cache(entry);
1422 }
1423 }
1424
1425 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1426 void *base, unsigned long base_size, enum object_type type)
1427 {
1428 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1429 struct list_head *lru, *tmp;
1430
1431 delta_base_cached += base_size;
1432
1433 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1434 struct delta_base_cache_entry *f =
1435 list_entry(lru, struct delta_base_cache_entry, lru);
1436 if (delta_base_cached <= delta_base_cache_limit)
1437 break;
1438 release_delta_base_cache(f);
1439 }
1440
1441 ent->key.p = p;
1442 ent->key.base_offset = base_offset;
1443 ent->type = type;
1444 ent->data = base;
1445 ent->size = base_size;
1446 list_add_tail(&ent->lru, &delta_base_cache_lru);
1447
1448 if (!delta_base_cache.cmpfn)
1449 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1450 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1451 hashmap_add(&delta_base_cache, ent);
1452 }
1453
1454 int packed_object_info(struct repository *r, struct packed_git *p,
1455 off_t obj_offset, struct object_info *oi)
1456 {
1457 struct pack_window *w_curs = NULL;
1458 unsigned long size;
1459 off_t curpos = obj_offset;
1460 enum object_type type;
1461
1462 /*
1463 * We always get the representation type, but only convert it to
1464 * a "real" type later if the caller is interested.
1465 */
1466 if (oi->contentp) {
1467 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1468 &type);
1469 if (!*oi->contentp)
1470 type = OBJ_BAD;
1471 } else {
1472 type = unpack_object_header(p, &w_curs, &curpos, &size);
1473 }
1474
1475 if (!oi->contentp && oi->sizep) {
1476 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1477 off_t tmp_pos = curpos;
1478 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1479 type, obj_offset);
1480 if (!base_offset) {
1481 type = OBJ_BAD;
1482 goto out;
1483 }
1484 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1485 if (*oi->sizep == 0) {
1486 type = OBJ_BAD;
1487 goto out;
1488 }
1489 } else {
1490 *oi->sizep = size;
1491 }
1492 }
1493
1494 if (oi->disk_sizep) {
1495 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1496 *oi->disk_sizep = revidx[1].offset - obj_offset;
1497 }
1498
1499 if (oi->typep || oi->type_name) {
1500 enum object_type ptot;
1501 ptot = packed_to_object_type(r, p, obj_offset,
1502 type, &w_curs, curpos);
1503 if (oi->typep)
1504 *oi->typep = ptot;
1505 if (oi->type_name) {
1506 const char *tn = type_name(ptot);
1507 if (tn)
1508 strbuf_addstr(oi->type_name, tn);
1509 }
1510 if (ptot < 0) {
1511 type = OBJ_BAD;
1512 goto out;
1513 }
1514 }
1515
1516 if (oi->delta_base_sha1) {
1517 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1518 const unsigned char *base;
1519
1520 base = get_delta_base_sha1(p, &w_curs, curpos,
1521 type, obj_offset);
1522 if (!base) {
1523 type = OBJ_BAD;
1524 goto out;
1525 }
1526
1527 hashcpy(oi->delta_base_sha1, base);
1528 } else
1529 hashclr(oi->delta_base_sha1);
1530 }
1531
1532 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1533 OI_PACKED;
1534
1535 out:
1536 unuse_pack(&w_curs);
1537 return type;
1538 }
1539
1540 static void *unpack_compressed_entry(struct packed_git *p,
1541 struct pack_window **w_curs,
1542 off_t curpos,
1543 unsigned long size)
1544 {
1545 int st;
1546 git_zstream stream;
1547 unsigned char *buffer, *in;
1548
1549 buffer = xmallocz_gently(size);
1550 if (!buffer)
1551 return NULL;
1552 memset(&stream, 0, sizeof(stream));
1553 stream.next_out = buffer;
1554 stream.avail_out = size + 1;
1555
1556 git_inflate_init(&stream);
1557 do {
1558 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1559 stream.next_in = in;
1560 st = git_inflate(&stream, Z_FINISH);
1561 if (!stream.avail_out)
1562 break; /* the payload is larger than it should be */
1563 curpos += stream.next_in - in;
1564 } while (st == Z_OK || st == Z_BUF_ERROR);
1565 git_inflate_end(&stream);
1566 if ((st != Z_STREAM_END) || stream.total_out != size) {
1567 free(buffer);
1568 return NULL;
1569 }
1570
1571 /* versions of zlib can clobber unconsumed portion of outbuf */
1572 buffer[size] = '\0';
1573
1574 return buffer;
1575 }
1576
1577 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1578 {
1579 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1580 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1581 p->pack_name, (uintmax_t)obj_offset);
1582 }
1583
1584 int do_check_packed_object_crc;
1585
1586 #define UNPACK_ENTRY_STACK_PREALLOC 64
1587 struct unpack_entry_stack_ent {
1588 off_t obj_offset;
1589 off_t curpos;
1590 unsigned long size;
1591 };
1592
1593 static void *read_object(struct repository *r,
1594 const struct object_id *oid,
1595 enum object_type *type,
1596 unsigned long *size)
1597 {
1598 struct object_info oi = OBJECT_INFO_INIT;
1599 void *content;
1600 oi.typep = type;
1601 oi.sizep = size;
1602 oi.contentp = &content;
1603
1604 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1605 return NULL;
1606 return content;
1607 }
1608
1609 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1610 enum object_type *final_type, unsigned long *final_size)
1611 {
1612 struct pack_window *w_curs = NULL;
1613 off_t curpos = obj_offset;
1614 void *data = NULL;
1615 unsigned long size;
1616 enum object_type type;
1617 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1618 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1619 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1620 int base_from_cache = 0;
1621
1622 write_pack_access_log(p, obj_offset);
1623
1624 /* PHASE 1: drill down to the innermost base object */
1625 for (;;) {
1626 off_t base_offset;
1627 int i;
1628 struct delta_base_cache_entry *ent;
1629
1630 ent = get_delta_base_cache_entry(p, curpos);
1631 if (ent) {
1632 type = ent->type;
1633 data = ent->data;
1634 size = ent->size;
1635 detach_delta_base_cache_entry(ent);
1636 base_from_cache = 1;
1637 break;
1638 }
1639
1640 if (do_check_packed_object_crc && p->index_version > 1) {
1641 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1642 off_t len = revidx[1].offset - obj_offset;
1643 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1644 struct object_id oid;
1645 nth_packed_object_oid(&oid, p, revidx->nr);
1646 error("bad packed object CRC for %s",
1647 oid_to_hex(&oid));
1648 mark_bad_packed_object(p, oid.hash);
1649 data = NULL;
1650 goto out;
1651 }
1652 }
1653
1654 type = unpack_object_header(p, &w_curs, &curpos, &size);
1655 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1656 break;
1657
1658 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1659 if (!base_offset) {
1660 error("failed to validate delta base reference "
1661 "at offset %"PRIuMAX" from %s",
1662 (uintmax_t)curpos, p->pack_name);
1663 /* bail to phase 2, in hopes of recovery */
1664 data = NULL;
1665 break;
1666 }
1667
1668 /* push object, proceed to base */
1669 if (delta_stack_nr >= delta_stack_alloc
1670 && delta_stack == small_delta_stack) {
1671 delta_stack_alloc = alloc_nr(delta_stack_nr);
1672 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1673 memcpy(delta_stack, small_delta_stack,
1674 sizeof(*delta_stack)*delta_stack_nr);
1675 } else {
1676 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1677 }
1678 i = delta_stack_nr++;
1679 delta_stack[i].obj_offset = obj_offset;
1680 delta_stack[i].curpos = curpos;
1681 delta_stack[i].size = size;
1682
1683 curpos = obj_offset = base_offset;
1684 }
1685
1686 /* PHASE 2: handle the base */
1687 switch (type) {
1688 case OBJ_OFS_DELTA:
1689 case OBJ_REF_DELTA:
1690 if (data)
1691 BUG("unpack_entry: left loop at a valid delta");
1692 break;
1693 case OBJ_COMMIT:
1694 case OBJ_TREE:
1695 case OBJ_BLOB:
1696 case OBJ_TAG:
1697 if (!base_from_cache)
1698 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1699 break;
1700 default:
1701 data = NULL;
1702 error("unknown object type %i at offset %"PRIuMAX" in %s",
1703 type, (uintmax_t)obj_offset, p->pack_name);
1704 }
1705
1706 /* PHASE 3: apply deltas in order */
1707
1708 /* invariants:
1709 * 'data' holds the base data, or NULL if there was corruption
1710 */
1711 while (delta_stack_nr) {
1712 void *delta_data;
1713 void *base = data;
1714 void *external_base = NULL;
1715 unsigned long delta_size, base_size = size;
1716 int i;
1717
1718 data = NULL;
1719
1720 if (base)
1721 add_delta_base_cache(p, obj_offset, base, base_size, type);
1722
1723 if (!base) {
1724 /*
1725 * We're probably in deep shit, but let's try to fetch
1726 * the required base anyway from another pack or loose.
1727 * This is costly but should happen only in the presence
1728 * of a corrupted pack, and is better than failing outright.
1729 */
1730 struct revindex_entry *revidx;
1731 struct object_id base_oid;
1732 revidx = find_pack_revindex(p, obj_offset);
1733 if (revidx) {
1734 nth_packed_object_oid(&base_oid, p, revidx->nr);
1735 error("failed to read delta base object %s"
1736 " at offset %"PRIuMAX" from %s",
1737 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1738 p->pack_name);
1739 mark_bad_packed_object(p, base_oid.hash);
1740 base = read_object(r, &base_oid, &type, &base_size);
1741 external_base = base;
1742 }
1743 }
1744
1745 i = --delta_stack_nr;
1746 obj_offset = delta_stack[i].obj_offset;
1747 curpos = delta_stack[i].curpos;
1748 delta_size = delta_stack[i].size;
1749
1750 if (!base)
1751 continue;
1752
1753 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1754
1755 if (!delta_data) {
1756 error("failed to unpack compressed delta "
1757 "at offset %"PRIuMAX" from %s",
1758 (uintmax_t)curpos, p->pack_name);
1759 data = NULL;
1760 free(external_base);
1761 continue;
1762 }
1763
1764 data = patch_delta(base, base_size,
1765 delta_data, delta_size,
1766 &size);
1767
1768 /*
1769 * We could not apply the delta; warn the user, but keep going.
1770 * Our failure will be noticed either in the next iteration of
1771 * the loop, or if this is the final delta, in the caller when
1772 * we return NULL. Those code paths will take care of making
1773 * a more explicit warning and retrying with another copy of
1774 * the object.
1775 */
1776 if (!data)
1777 error("failed to apply delta");
1778
1779 free(delta_data);
1780 free(external_base);
1781 }
1782
1783 if (final_type)
1784 *final_type = type;
1785 if (final_size)
1786 *final_size = size;
1787
1788 out:
1789 unuse_pack(&w_curs);
1790
1791 if (delta_stack != small_delta_stack)
1792 free(delta_stack);
1793
1794 return data;
1795 }
1796
1797 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1798 {
1799 const unsigned char *index_fanout = p->index_data;
1800 const unsigned char *index_lookup;
1801 const unsigned int hashsz = the_hash_algo->rawsz;
1802 int index_lookup_width;
1803
1804 if (!index_fanout)
1805 BUG("bsearch_pack called without a valid pack-index");
1806
1807 index_lookup = index_fanout + 4 * 256;
1808 if (p->index_version == 1) {
1809 index_lookup_width = hashsz + 4;
1810 index_lookup += 4;
1811 } else {
1812 index_lookup_width = hashsz;
1813 index_fanout += 8;
1814 index_lookup += 8;
1815 }
1816
1817 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1818 index_lookup, index_lookup_width, result);
1819 }
1820
1821 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1822 uint32_t n)
1823 {
1824 const unsigned char *index = p->index_data;
1825 const unsigned int hashsz = the_hash_algo->rawsz;
1826 if (!index) {
1827 if (open_pack_index(p))
1828 return NULL;
1829 index = p->index_data;
1830 }
1831 if (n >= p->num_objects)
1832 return NULL;
1833 index += 4 * 256;
1834 if (p->index_version == 1) {
1835 return index + (hashsz + 4) * n + 4;
1836 } else {
1837 index += 8;
1838 return index + hashsz * n;
1839 }
1840 }
1841
1842 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1843 struct packed_git *p,
1844 uint32_t n)
1845 {
1846 const unsigned char *hash = nth_packed_object_sha1(p, n);
1847 if (!hash)
1848 return NULL;
1849 hashcpy(oid->hash, hash);
1850 return oid;
1851 }
1852
1853 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1854 {
1855 const unsigned char *ptr = vptr;
1856 const unsigned char *start = p->index_data;
1857 const unsigned char *end = start + p->index_size;
1858 if (ptr < start)
1859 die(_("offset before start of pack index for %s (corrupt index?)"),
1860 p->pack_name);
1861 /* No need to check for underflow; .idx files must be at least 8 bytes */
1862 if (ptr >= end - 8)
1863 die(_("offset beyond end of pack index for %s (truncated index?)"),
1864 p->pack_name);
1865 }
1866
1867 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1868 {
1869 const unsigned char *index = p->index_data;
1870 const unsigned int hashsz = the_hash_algo->rawsz;
1871 index += 4 * 256;
1872 if (p->index_version == 1) {
1873 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1874 } else {
1875 uint32_t off;
1876 index += 8 + p->num_objects * (hashsz + 4);
1877 off = ntohl(*((uint32_t *)(index + 4 * n)));
1878 if (!(off & 0x80000000))
1879 return off;
1880 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1881 check_pack_index_ptr(p, index);
1882 return get_be64(index);
1883 }
1884 }
1885
1886 off_t find_pack_entry_one(const unsigned char *sha1,
1887 struct packed_git *p)
1888 {
1889 const unsigned char *index = p->index_data;
1890 struct object_id oid;
1891 uint32_t result;
1892
1893 if (!index) {
1894 if (open_pack_index(p))
1895 return 0;
1896 }
1897
1898 hashcpy(oid.hash, sha1);
1899 if (bsearch_pack(&oid, p, &result))
1900 return nth_packed_object_offset(p, result);
1901 return 0;
1902 }
1903
1904 int is_pack_valid(struct packed_git *p)
1905 {
1906 /* An already open pack is known to be valid. */
1907 if (p->pack_fd != -1)
1908 return 1;
1909
1910 /* If the pack has one window completely covering the
1911 * file size, the pack is known to be valid even if
1912 * the descriptor is not currently open.
1913 */
1914 if (p->windows) {
1915 struct pack_window *w = p->windows;
1916
1917 if (!w->offset && w->len == p->pack_size)
1918 return 1;
1919 }
1920
1921 /* Force the pack to open to prove its valid. */
1922 return !open_packed_git(p);
1923 }
1924
1925 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1926 struct packed_git *packs)
1927 {
1928 struct packed_git *p;
1929
1930 for (p = packs; p; p = p->next) {
1931 if (find_pack_entry_one(sha1, p))
1932 return p;
1933 }
1934 return NULL;
1935
1936 }
1937
1938 static int fill_pack_entry(const struct object_id *oid,
1939 struct pack_entry *e,
1940 struct packed_git *p)
1941 {
1942 off_t offset;
1943
1944 if (p->num_bad_objects) {
1945 unsigned i;
1946 for (i = 0; i < p->num_bad_objects; i++)
1947 if (hasheq(oid->hash,
1948 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1949 return 0;
1950 }
1951
1952 offset = find_pack_entry_one(oid->hash, p);
1953 if (!offset)
1954 return 0;
1955
1956 /*
1957 * We are about to tell the caller where they can locate the
1958 * requested object. We better make sure the packfile is
1959 * still here and can be accessed before supplying that
1960 * answer, as it may have been deleted since the index was
1961 * loaded!
1962 */
1963 if (!is_pack_valid(p))
1964 return 0;
1965 e->offset = offset;
1966 e->p = p;
1967 return 1;
1968 }
1969
1970 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1971 {
1972 struct list_head *pos;
1973 struct multi_pack_index *m;
1974
1975 prepare_packed_git(r);
1976 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1977 return 0;
1978
1979 for (m = r->objects->multi_pack_index; m; m = m->next) {
1980 if (fill_midx_entry(oid, e, m))
1981 return 1;
1982 }
1983
1984 list_for_each(pos, &r->objects->packed_git_mru) {
1985 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1986 if (fill_pack_entry(oid, e, p)) {
1987 list_move(&p->mru, &r->objects->packed_git_mru);
1988 return 1;
1989 }
1990 }
1991 return 0;
1992 }
1993
1994 int has_object_pack(const struct object_id *oid)
1995 {
1996 struct pack_entry e;
1997 return find_pack_entry(the_repository, oid, &e);
1998 }
1999
2000 int has_pack_index(const unsigned char *sha1)
2001 {
2002 struct stat st;
2003 if (stat(sha1_pack_index_name(sha1), &st))
2004 return 0;
2005 return 1;
2006 }
2007
2008 int for_each_object_in_pack(struct packed_git *p,
2009 each_packed_object_fn cb, void *data,
2010 enum for_each_object_flags flags)
2011 {
2012 uint32_t i;
2013 int r = 0;
2014
2015 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2016 load_pack_revindex(p);
2017
2018 for (i = 0; i < p->num_objects; i++) {
2019 uint32_t pos;
2020 struct object_id oid;
2021
2022 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2023 pos = p->revindex[i].nr;
2024 else
2025 pos = i;
2026
2027 if (!nth_packed_object_oid(&oid, p, pos))
2028 return error("unable to get sha1 of object %u in %s",
2029 pos, p->pack_name);
2030
2031 r = cb(&oid, p, pos, data);
2032 if (r)
2033 break;
2034 }
2035 return r;
2036 }
2037
2038 int for_each_packed_object(each_packed_object_fn cb, void *data,
2039 enum for_each_object_flags flags)
2040 {
2041 struct packed_git *p;
2042 int r = 0;
2043 int pack_errors = 0;
2044
2045 prepare_packed_git(the_repository);
2046 for (p = get_all_packs(the_repository); p; p = p->next) {
2047 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2048 continue;
2049 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2050 !p->pack_promisor)
2051 continue;
2052 if (open_pack_index(p)) {
2053 pack_errors = 1;
2054 continue;
2055 }
2056 r = for_each_object_in_pack(p, cb, data, flags);
2057 if (r)
2058 break;
2059 }
2060 return r ? r : pack_errors;
2061 }
2062
2063 static int add_promisor_object(const struct object_id *oid,
2064 struct packed_git *pack,
2065 uint32_t pos,
2066 void *set_)
2067 {
2068 struct oidset *set = set_;
2069 struct object *obj = parse_object(the_repository, oid);
2070 if (!obj)
2071 return 1;
2072
2073 oidset_insert(set, oid);
2074
2075 /*
2076 * If this is a tree, commit, or tag, the objects it refers
2077 * to are also promisor objects. (Blobs refer to no objects->)
2078 */
2079 if (obj->type == OBJ_TREE) {
2080 struct tree *tree = (struct tree *)obj;
2081 struct tree_desc desc;
2082 struct name_entry entry;
2083 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2084 /*
2085 * Error messages are given when packs are
2086 * verified, so do not print any here.
2087 */
2088 return 0;
2089 while (tree_entry_gently(&desc, &entry))
2090 oidset_insert(set, entry.oid);
2091 } else if (obj->type == OBJ_COMMIT) {
2092 struct commit *commit = (struct commit *) obj;
2093 struct commit_list *parents = commit->parents;
2094
2095 oidset_insert(set, get_commit_tree_oid(commit));
2096 for (; parents; parents = parents->next)
2097 oidset_insert(set, &parents->item->object.oid);
2098 } else if (obj->type == OBJ_TAG) {
2099 struct tag *tag = (struct tag *) obj;
2100 oidset_insert(set, &tag->tagged->oid);
2101 }
2102 return 0;
2103 }
2104
2105 int is_promisor_object(const struct object_id *oid)
2106 {
2107 static struct oidset promisor_objects;
2108 static int promisor_objects_prepared;
2109
2110 if (!promisor_objects_prepared) {
2111 if (repository_format_partial_clone) {
2112 for_each_packed_object(add_promisor_object,
2113 &promisor_objects,
2114 FOR_EACH_OBJECT_PROMISOR_ONLY);
2115 }
2116 promisor_objects_prepared = 1;
2117 }
2118 return oidset_contains(&promisor_objects, oid);
2119 }
Eric Wong's git-svn branch