midx: prevent duplicate packfile loads
[git/git.git] / packfile.c
1 #include "cache.h"
2 #include "list.h"
3 #include "pack.h"
4 #include "repository.h"
5 #include "dir.h"
6 #include "mergesort.h"
7 #include "packfile.h"
8 #include "delta.h"
9 #include "list.h"
10 #include "streaming.h"
11 #include "sha1-lookup.h"
12 #include "commit.h"
13 #include "object.h"
14 #include "tag.h"
15 #include "tree-walk.h"
16 #include "tree.h"
17 #include "object-store.h"
18 #include "midx.h"
19
20 char *odb_pack_name(struct strbuf *buf,
21 const unsigned char *sha1,
22 const char *ext)
23 {
24 strbuf_reset(buf);
25 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
26 sha1_to_hex(sha1), ext);
27 return buf->buf;
28 }
29
30 char *sha1_pack_name(const unsigned char *sha1)
31 {
32 static struct strbuf buf = STRBUF_INIT;
33 return odb_pack_name(&buf, sha1, "pack");
34 }
35
36 char *sha1_pack_index_name(const unsigned char *sha1)
37 {
38 static struct strbuf buf = STRBUF_INIT;
39 return odb_pack_name(&buf, sha1, "idx");
40 }
41
42 static unsigned int pack_used_ctr;
43 static unsigned int pack_mmap_calls;
44 static unsigned int peak_pack_open_windows;
45 static unsigned int pack_open_windows;
46 static unsigned int pack_open_fds;
47 static unsigned int pack_max_fds;
48 static size_t peak_pack_mapped;
49 static size_t pack_mapped;
50
51 #define SZ_FMT PRIuMAX
52 static inline uintmax_t sz_fmt(size_t s) { return s; }
53
54 void pack_report(void)
55 {
56 fprintf(stderr,
57 "pack_report: getpagesize() = %10" SZ_FMT "\n"
58 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
60 sz_fmt(getpagesize()),
61 sz_fmt(packed_git_window_size),
62 sz_fmt(packed_git_limit));
63 fprintf(stderr,
64 "pack_report: pack_used_ctr = %10u\n"
65 "pack_report: pack_mmap_calls = %10u\n"
66 "pack_report: pack_open_windows = %10u / %10u\n"
67 "pack_report: pack_mapped = "
68 "%10" SZ_FMT " / %10" SZ_FMT "\n",
69 pack_used_ctr,
70 pack_mmap_calls,
71 pack_open_windows, peak_pack_open_windows,
72 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
73 }
74
75 /*
76 * Open and mmap the index file at path, perform a couple of
77 * consistency checks, then record its information to p. Return 0 on
78 * success.
79 */
80 static int check_packed_git_idx(const char *path, struct packed_git *p)
81 {
82 void *idx_map;
83 struct pack_idx_header *hdr;
84 size_t idx_size;
85 uint32_t version, nr, i, *index;
86 int fd = git_open(path);
87 struct stat st;
88 const unsigned int hashsz = the_hash_algo->rawsz;
89
90 if (fd < 0)
91 return -1;
92 if (fstat(fd, &st)) {
93 close(fd);
94 return -1;
95 }
96 idx_size = xsize_t(st.st_size);
97 if (idx_size < 4 * 256 + hashsz + hashsz) {
98 close(fd);
99 return error("index file %s is too small", path);
100 }
101 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
102 close(fd);
103
104 hdr = idx_map;
105 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
106 version = ntohl(hdr->idx_version);
107 if (version < 2 || version > 2) {
108 munmap(idx_map, idx_size);
109 return error("index file %s is version %"PRIu32
110 " and is not supported by this binary"
111 " (try upgrading GIT to a newer version)",
112 path, version);
113 }
114 } else
115 version = 1;
116
117 nr = 0;
118 index = idx_map;
119 if (version > 1)
120 index += 2; /* skip index header */
121 for (i = 0; i < 256; i++) {
122 uint32_t n = ntohl(index[i]);
123 if (n < nr) {
124 munmap(idx_map, idx_size);
125 return error("non-monotonic index %s", path);
126 }
127 nr = n;
128 }
129
130 if (version == 1) {
131 /*
132 * Total size:
133 * - 256 index entries 4 bytes each
134 * - 24-byte entries * nr (object ID + 4-byte offset)
135 * - hash of the packfile
136 * - file checksum
137 */
138 if (idx_size != 4*256 + nr * (hashsz + 4) + hashsz + hashsz) {
139 munmap(idx_map, idx_size);
140 return error("wrong index v1 file size in %s", path);
141 }
142 } else if (version == 2) {
143 /*
144 * Minimum size:
145 * - 8 bytes of header
146 * - 256 index entries 4 bytes each
147 * - object ID entry * nr
148 * - 4-byte crc entry * nr
149 * - 4-byte offset entry * nr
150 * - hash of the packfile
151 * - file checksum
152 * And after the 4-byte offset table might be a
153 * variable sized table containing 8-byte entries
154 * for offsets larger than 2^31.
155 */
156 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
157 unsigned long max_size = min_size;
158 if (nr)
159 max_size += (nr - 1)*8;
160 if (idx_size < min_size || idx_size > max_size) {
161 munmap(idx_map, idx_size);
162 return error("wrong index v2 file size in %s", path);
163 }
164 if (idx_size != min_size &&
165 /*
166 * make sure we can deal with large pack offsets.
167 * 31-bit signed offset won't be enough, neither
168 * 32-bit unsigned one will be.
169 */
170 (sizeof(off_t) <= 4)) {
171 munmap(idx_map, idx_size);
172 return error("pack too large for current definition of off_t in %s", path);
173 }
174 }
175
176 p->index_version = version;
177 p->index_data = idx_map;
178 p->index_size = idx_size;
179 p->num_objects = nr;
180 return 0;
181 }
182
183 int open_pack_index(struct packed_git *p)
184 {
185 char *idx_name;
186 size_t len;
187 int ret;
188
189 if (p->index_data)
190 return 0;
191
192 if (!strip_suffix(p->pack_name, ".pack", &len))
193 BUG("pack_name does not end in .pack");
194 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
195 ret = check_packed_git_idx(idx_name, p);
196 free(idx_name);
197 return ret;
198 }
199
200 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
201 {
202 const uint32_t *level1_ofs = p->index_data;
203
204 if (!level1_ofs) {
205 if (open_pack_index(p))
206 return 0;
207 level1_ofs = p->index_data;
208 }
209
210 if (p->index_version > 1) {
211 level1_ofs += 2;
212 }
213
214 return ntohl(level1_ofs[value]);
215 }
216
217 static struct packed_git *alloc_packed_git(int extra)
218 {
219 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
220 memset(p, 0, sizeof(*p));
221 p->pack_fd = -1;
222 return p;
223 }
224
225 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
226 {
227 const char *path = sha1_pack_name(sha1);
228 size_t alloc = st_add(strlen(path), 1);
229 struct packed_git *p = alloc_packed_git(alloc);
230
231 memcpy(p->pack_name, path, alloc); /* includes NUL */
232 hashcpy(p->sha1, sha1);
233 if (check_packed_git_idx(idx_path, p)) {
234 free(p);
235 return NULL;
236 }
237
238 return p;
239 }
240
241 static void scan_windows(struct packed_git *p,
242 struct packed_git **lru_p,
243 struct pack_window **lru_w,
244 struct pack_window **lru_l)
245 {
246 struct pack_window *w, *w_l;
247
248 for (w_l = NULL, w = p->windows; w; w = w->next) {
249 if (!w->inuse_cnt) {
250 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
251 *lru_p = p;
252 *lru_w = w;
253 *lru_l = w_l;
254 }
255 }
256 w_l = w;
257 }
258 }
259
260 static int unuse_one_window(struct packed_git *current)
261 {
262 struct packed_git *p, *lru_p = NULL;
263 struct pack_window *lru_w = NULL, *lru_l = NULL;
264
265 if (current)
266 scan_windows(current, &lru_p, &lru_w, &lru_l);
267 for (p = the_repository->objects->packed_git; p; p = p->next)
268 scan_windows(p, &lru_p, &lru_w, &lru_l);
269 if (lru_p) {
270 munmap(lru_w->base, lru_w->len);
271 pack_mapped -= lru_w->len;
272 if (lru_l)
273 lru_l->next = lru_w->next;
274 else
275 lru_p->windows = lru_w->next;
276 free(lru_w);
277 pack_open_windows--;
278 return 1;
279 }
280 return 0;
281 }
282
283 void release_pack_memory(size_t need)
284 {
285 size_t cur = pack_mapped;
286 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
287 ; /* nothing */
288 }
289
290 void close_pack_windows(struct packed_git *p)
291 {
292 while (p->windows) {
293 struct pack_window *w = p->windows;
294
295 if (w->inuse_cnt)
296 die("pack '%s' still has open windows to it",
297 p->pack_name);
298 munmap(w->base, w->len);
299 pack_mapped -= w->len;
300 pack_open_windows--;
301 p->windows = w->next;
302 free(w);
303 }
304 }
305
306 static int close_pack_fd(struct packed_git *p)
307 {
308 if (p->pack_fd < 0)
309 return 0;
310
311 close(p->pack_fd);
312 pack_open_fds--;
313 p->pack_fd = -1;
314
315 return 1;
316 }
317
318 void close_pack_index(struct packed_git *p)
319 {
320 if (p->index_data) {
321 munmap((void *)p->index_data, p->index_size);
322 p->index_data = NULL;
323 }
324 }
325
326 void close_pack(struct packed_git *p)
327 {
328 close_pack_windows(p);
329 close_pack_fd(p);
330 close_pack_index(p);
331 }
332
333 void close_all_packs(struct raw_object_store *o)
334 {
335 struct packed_git *p;
336
337 for (p = o->packed_git; p; p = p->next)
338 if (p->do_not_close)
339 BUG("want to close pack marked 'do-not-close'");
340 else
341 close_pack(p);
342 }
343
344 /*
345 * The LRU pack is the one with the oldest MRU window, preferring packs
346 * with no used windows, or the oldest mtime if it has no windows allocated.
347 */
348 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
349 {
350 struct pack_window *w, *this_mru_w;
351 int has_windows_inuse = 0;
352
353 /*
354 * Reject this pack if it has windows and the previously selected
355 * one does not. If this pack does not have windows, reject
356 * it if the pack file is newer than the previously selected one.
357 */
358 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
359 return;
360
361 for (w = this_mru_w = p->windows; w; w = w->next) {
362 /*
363 * Reject this pack if any of its windows are in use,
364 * but the previously selected pack did not have any
365 * inuse windows. Otherwise, record that this pack
366 * has windows in use.
367 */
368 if (w->inuse_cnt) {
369 if (*accept_windows_inuse)
370 has_windows_inuse = 1;
371 else
372 return;
373 }
374
375 if (w->last_used > this_mru_w->last_used)
376 this_mru_w = w;
377
378 /*
379 * Reject this pack if it has windows that have been
380 * used more recently than the previously selected pack.
381 * If the previously selected pack had windows inuse and
382 * we have not encountered a window in this pack that is
383 * inuse, skip this check since we prefer a pack with no
384 * inuse windows to one that has inuse windows.
385 */
386 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
387 this_mru_w->last_used > (*mru_w)->last_used)
388 return;
389 }
390
391 /*
392 * Select this pack.
393 */
394 *mru_w = this_mru_w;
395 *lru_p = p;
396 *accept_windows_inuse = has_windows_inuse;
397 }
398
399 static int close_one_pack(void)
400 {
401 struct packed_git *p, *lru_p = NULL;
402 struct pack_window *mru_w = NULL;
403 int accept_windows_inuse = 1;
404
405 for (p = the_repository->objects->packed_git; p; p = p->next) {
406 if (p->pack_fd == -1)
407 continue;
408 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
409 }
410
411 if (lru_p)
412 return close_pack_fd(lru_p);
413
414 return 0;
415 }
416
417 static unsigned int get_max_fd_limit(void)
418 {
419 #ifdef RLIMIT_NOFILE
420 {
421 struct rlimit lim;
422
423 if (!getrlimit(RLIMIT_NOFILE, &lim))
424 return lim.rlim_cur;
425 }
426 #endif
427
428 #ifdef _SC_OPEN_MAX
429 {
430 long open_max = sysconf(_SC_OPEN_MAX);
431 if (0 < open_max)
432 return open_max;
433 /*
434 * Otherwise, we got -1 for one of the two
435 * reasons:
436 *
437 * (1) sysconf() did not understand _SC_OPEN_MAX
438 * and signaled an error with -1; or
439 * (2) sysconf() said there is no limit.
440 *
441 * We _could_ clear errno before calling sysconf() to
442 * tell these two cases apart and return a huge number
443 * in the latter case to let the caller cap it to a
444 * value that is not so selfish, but letting the
445 * fallback OPEN_MAX codepath take care of these cases
446 * is a lot simpler.
447 */
448 }
449 #endif
450
451 #ifdef OPEN_MAX
452 return OPEN_MAX;
453 #else
454 return 1; /* see the caller ;-) */
455 #endif
456 }
457
458 /*
459 * Do not call this directly as this leaks p->pack_fd on error return;
460 * call open_packed_git() instead.
461 */
462 static int open_packed_git_1(struct packed_git *p)
463 {
464 struct stat st;
465 struct pack_header hdr;
466 unsigned char hash[GIT_MAX_RAWSZ];
467 unsigned char *idx_hash;
468 long fd_flag;
469 ssize_t read_result;
470 const unsigned hashsz = the_hash_algo->rawsz;
471
472 if (!p->index_data && open_pack_index(p))
473 return error("packfile %s index unavailable", p->pack_name);
474
475 if (!pack_max_fds) {
476 unsigned int max_fds = get_max_fd_limit();
477
478 /* Save 3 for stdin/stdout/stderr, 22 for work */
479 if (25 < max_fds)
480 pack_max_fds = max_fds - 25;
481 else
482 pack_max_fds = 1;
483 }
484
485 while (pack_max_fds <= pack_open_fds && close_one_pack())
486 ; /* nothing */
487
488 p->pack_fd = git_open(p->pack_name);
489 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
490 return -1;
491 pack_open_fds++;
492
493 /* If we created the struct before we had the pack we lack size. */
494 if (!p->pack_size) {
495 if (!S_ISREG(st.st_mode))
496 return error("packfile %s not a regular file", p->pack_name);
497 p->pack_size = st.st_size;
498 } else if (p->pack_size != st.st_size)
499 return error("packfile %s size changed", p->pack_name);
500
501 /* We leave these file descriptors open with sliding mmap;
502 * there is no point keeping them open across exec(), though.
503 */
504 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
505 if (fd_flag < 0)
506 return error("cannot determine file descriptor flags");
507 fd_flag |= FD_CLOEXEC;
508 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
509 return error("cannot set FD_CLOEXEC");
510
511 /* Verify we recognize this pack file format. */
512 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
513 if (read_result < 0)
514 return error_errno("error reading from %s", p->pack_name);
515 if (read_result != sizeof(hdr))
516 return error("file %s is far too short to be a packfile", p->pack_name);
517 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
518 return error("file %s is not a GIT packfile", p->pack_name);
519 if (!pack_version_ok(hdr.hdr_version))
520 return error("packfile %s is version %"PRIu32" and not"
521 " supported (try upgrading GIT to a newer version)",
522 p->pack_name, ntohl(hdr.hdr_version));
523
524 /* Verify the pack matches its index. */
525 if (p->num_objects != ntohl(hdr.hdr_entries))
526 return error("packfile %s claims to have %"PRIu32" objects"
527 " while index indicates %"PRIu32" objects",
528 p->pack_name, ntohl(hdr.hdr_entries),
529 p->num_objects);
530 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
531 return error("end of packfile %s is unavailable", p->pack_name);
532 read_result = read_in_full(p->pack_fd, hash, hashsz);
533 if (read_result < 0)
534 return error_errno("error reading from %s", p->pack_name);
535 if (read_result != hashsz)
536 return error("packfile %s signature is unavailable", p->pack_name);
537 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
538 if (hashcmp(hash, idx_hash))
539 return error("packfile %s does not match index", p->pack_name);
540 return 0;
541 }
542
543 static int open_packed_git(struct packed_git *p)
544 {
545 if (!open_packed_git_1(p))
546 return 0;
547 close_pack_fd(p);
548 return -1;
549 }
550
551 static int in_window(struct pack_window *win, off_t offset)
552 {
553 /* We must promise at least one full hash after the
554 * offset is available from this window, otherwise the offset
555 * is not actually in this window and a different window (which
556 * has that one hash excess) must be used. This is to support
557 * the object header and delta base parsing routines below.
558 */
559 off_t win_off = win->offset;
560 return win_off <= offset
561 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
562 }
563
564 unsigned char *use_pack(struct packed_git *p,
565 struct pack_window **w_cursor,
566 off_t offset,
567 unsigned long *left)
568 {
569 struct pack_window *win = *w_cursor;
570
571 /* Since packfiles end in a hash of their content and it's
572 * pointless to ask for an offset into the middle of that
573 * hash, and the in_window function above wouldn't match
574 * don't allow an offset too close to the end of the file.
575 */
576 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
577 die("packfile %s cannot be accessed", p->pack_name);
578 if (offset > (p->pack_size - the_hash_algo->rawsz))
579 die("offset beyond end of packfile (truncated pack?)");
580 if (offset < 0)
581 die(_("offset before end of packfile (broken .idx?)"));
582
583 if (!win || !in_window(win, offset)) {
584 if (win)
585 win->inuse_cnt--;
586 for (win = p->windows; win; win = win->next) {
587 if (in_window(win, offset))
588 break;
589 }
590 if (!win) {
591 size_t window_align = packed_git_window_size / 2;
592 off_t len;
593
594 if (p->pack_fd == -1 && open_packed_git(p))
595 die("packfile %s cannot be accessed", p->pack_name);
596
597 win = xcalloc(1, sizeof(*win));
598 win->offset = (offset / window_align) * window_align;
599 len = p->pack_size - win->offset;
600 if (len > packed_git_window_size)
601 len = packed_git_window_size;
602 win->len = (size_t)len;
603 pack_mapped += win->len;
604 while (packed_git_limit < pack_mapped
605 && unuse_one_window(p))
606 ; /* nothing */
607 win->base = xmmap(NULL, win->len,
608 PROT_READ, MAP_PRIVATE,
609 p->pack_fd, win->offset);
610 if (win->base == MAP_FAILED)
611 die_errno("packfile %s cannot be mapped",
612 p->pack_name);
613 if (!win->offset && win->len == p->pack_size
614 && !p->do_not_close)
615 close_pack_fd(p);
616 pack_mmap_calls++;
617 pack_open_windows++;
618 if (pack_mapped > peak_pack_mapped)
619 peak_pack_mapped = pack_mapped;
620 if (pack_open_windows > peak_pack_open_windows)
621 peak_pack_open_windows = pack_open_windows;
622 win->next = p->windows;
623 p->windows = win;
624 }
625 }
626 if (win != *w_cursor) {
627 win->last_used = pack_used_ctr++;
628 win->inuse_cnt++;
629 *w_cursor = win;
630 }
631 offset -= win->offset;
632 if (left)
633 *left = win->len - xsize_t(offset);
634 return win->base + offset;
635 }
636
637 void unuse_pack(struct pack_window **w_cursor)
638 {
639 struct pack_window *w = *w_cursor;
640 if (w) {
641 w->inuse_cnt--;
642 *w_cursor = NULL;
643 }
644 }
645
646 static void try_to_free_pack_memory(size_t size)
647 {
648 release_pack_memory(size);
649 }
650
651 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
652 {
653 static int have_set_try_to_free_routine;
654 struct stat st;
655 size_t alloc;
656 struct packed_git *p;
657
658 if (!have_set_try_to_free_routine) {
659 have_set_try_to_free_routine = 1;
660 set_try_to_free_routine(try_to_free_pack_memory);
661 }
662
663 /*
664 * Make sure a corresponding .pack file exists and that
665 * the index looks sane.
666 */
667 if (!strip_suffix_mem(path, &path_len, ".idx"))
668 return NULL;
669
670 /*
671 * ".promisor" is long enough to hold any suffix we're adding (and
672 * the use xsnprintf double-checks that)
673 */
674 alloc = st_add3(path_len, strlen(".promisor"), 1);
675 p = alloc_packed_git(alloc);
676 memcpy(p->pack_name, path, path_len);
677
678 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
679 if (!access(p->pack_name, F_OK))
680 p->pack_keep = 1;
681
682 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
683 if (!access(p->pack_name, F_OK))
684 p->pack_promisor = 1;
685
686 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
687 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
688 free(p);
689 return NULL;
690 }
691
692 /* ok, it looks sane as far as we can check without
693 * actually mapping the pack file.
694 */
695 p->pack_size = st.st_size;
696 p->pack_local = local;
697 p->mtime = st.st_mtime;
698 if (path_len < the_hash_algo->hexsz ||
699 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->sha1))
700 hashclr(p->sha1);
701 return p;
702 }
703
704 void install_packed_git(struct repository *r, struct packed_git *pack)
705 {
706 if (pack->pack_fd != -1)
707 pack_open_fds++;
708
709 pack->next = r->objects->packed_git;
710 r->objects->packed_git = pack;
711 }
712
713 void (*report_garbage)(unsigned seen_bits, const char *path);
714
715 static void report_helper(const struct string_list *list,
716 int seen_bits, int first, int last)
717 {
718 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
719 return;
720
721 for (; first < last; first++)
722 report_garbage(seen_bits, list->items[first].string);
723 }
724
725 static void report_pack_garbage(struct string_list *list)
726 {
727 int i, baselen = -1, first = 0, seen_bits = 0;
728
729 if (!report_garbage)
730 return;
731
732 string_list_sort(list);
733
734 for (i = 0; i < list->nr; i++) {
735 const char *path = list->items[i].string;
736 if (baselen != -1 &&
737 strncmp(path, list->items[first].string, baselen)) {
738 report_helper(list, seen_bits, first, i);
739 baselen = -1;
740 seen_bits = 0;
741 }
742 if (baselen == -1) {
743 const char *dot = strrchr(path, '.');
744 if (!dot) {
745 report_garbage(PACKDIR_FILE_GARBAGE, path);
746 continue;
747 }
748 baselen = dot - path + 1;
749 first = i;
750 }
751 if (!strcmp(path + baselen, "pack"))
752 seen_bits |= 1;
753 else if (!strcmp(path + baselen, "idx"))
754 seen_bits |= 2;
755 }
756 report_helper(list, seen_bits, first, list->nr);
757 }
758
759 void for_each_file_in_pack_dir(const char *objdir,
760 each_file_in_pack_dir_fn fn,
761 void *data)
762 {
763 struct strbuf path = STRBUF_INIT;
764 size_t dirnamelen;
765 DIR *dir;
766 struct dirent *de;
767
768 strbuf_addstr(&path, objdir);
769 strbuf_addstr(&path, "/pack");
770 dir = opendir(path.buf);
771 if (!dir) {
772 if (errno != ENOENT)
773 error_errno("unable to open object pack directory: %s",
774 path.buf);
775 strbuf_release(&path);
776 return;
777 }
778 strbuf_addch(&path, '/');
779 dirnamelen = path.len;
780 while ((de = readdir(dir)) != NULL) {
781 if (is_dot_or_dotdot(de->d_name))
782 continue;
783
784 strbuf_setlen(&path, dirnamelen);
785 strbuf_addstr(&path, de->d_name);
786
787 fn(path.buf, path.len, de->d_name, data);
788 }
789
790 closedir(dir);
791 strbuf_release(&path);
792 }
793
794 struct prepare_pack_data {
795 struct repository *r;
796 struct string_list *garbage;
797 int local;
798 struct multi_pack_index *m;
799 };
800
801 static void prepare_pack(const char *full_name, size_t full_name_len,
802 const char *file_name, void *_data)
803 {
804 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
805 struct packed_git *p;
806 size_t base_len = full_name_len;
807
808 if (strip_suffix_mem(full_name, &base_len, ".idx")) {
809 if (data->m && midx_contains_pack(data->m, file_name))
810 return;
811 /* Don't reopen a pack we already have. */
812 for (p = data->r->objects->packed_git; p; p = p->next) {
813 size_t len;
814 if (strip_suffix(p->pack_name, ".pack", &len) &&
815 len == base_len &&
816 !memcmp(p->pack_name, full_name, len))
817 break;
818 }
819
820 if (!p) {
821 p = add_packed_git(full_name, full_name_len, data->local);
822 if (p)
823 install_packed_git(data->r, p);
824 }
825 }
826
827 if (!report_garbage)
828 return;
829
830 if (ends_with(file_name, ".idx") ||
831 ends_with(file_name, ".pack") ||
832 ends_with(file_name, ".bitmap") ||
833 ends_with(file_name, ".keep") ||
834 ends_with(file_name, ".promisor"))
835 string_list_append(data->garbage, full_name);
836 else
837 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
838 }
839
840 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
841 {
842 struct prepare_pack_data data;
843 struct string_list garbage = STRING_LIST_INIT_DUP;
844
845 data.m = r->objects->multi_pack_index;
846
847 /* look for the multi-pack-index for this object directory */
848 while (data.m && strcmp(data.m->object_dir, objdir))
849 data.m = data.m->next;
850
851 data.r = r;
852 data.garbage = &garbage;
853 data.local = local;
854
855 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
856
857 report_pack_garbage(data.garbage);
858 string_list_clear(data.garbage, 0);
859 }
860
861 static void prepare_packed_git(struct repository *r);
862 /*
863 * Give a fast, rough count of the number of objects in the repository. This
864 * ignores loose objects completely. If you have a lot of them, then either
865 * you should repack because your performance will be awful, or they are
866 * all unreachable objects about to be pruned, in which case they're not really
867 * interesting as a measure of repo size in the first place.
868 */
869 unsigned long approximate_object_count(void)
870 {
871 if (!the_repository->objects->approximate_object_count_valid) {
872 unsigned long count;
873 struct multi_pack_index *m;
874 struct packed_git *p;
875
876 prepare_packed_git(the_repository);
877 count = 0;
878 for (m = get_multi_pack_index(the_repository); m; m = m->next)
879 count += m->num_objects;
880 for (p = the_repository->objects->packed_git; p; p = p->next) {
881 if (open_pack_index(p))
882 continue;
883 count += p->num_objects;
884 }
885 the_repository->objects->approximate_object_count = count;
886 }
887 return the_repository->objects->approximate_object_count;
888 }
889
890 static void *get_next_packed_git(const void *p)
891 {
892 return ((const struct packed_git *)p)->next;
893 }
894
895 static void set_next_packed_git(void *p, void *next)
896 {
897 ((struct packed_git *)p)->next = next;
898 }
899
900 static int sort_pack(const void *a_, const void *b_)
901 {
902 const struct packed_git *a = a_;
903 const struct packed_git *b = b_;
904 int st;
905
906 /*
907 * Local packs tend to contain objects specific to our
908 * variant of the project than remote ones. In addition,
909 * remote ones could be on a network mounted filesystem.
910 * Favor local ones for these reasons.
911 */
912 st = a->pack_local - b->pack_local;
913 if (st)
914 return -st;
915
916 /*
917 * Younger packs tend to contain more recent objects,
918 * and more recent objects tend to get accessed more
919 * often.
920 */
921 if (a->mtime < b->mtime)
922 return 1;
923 else if (a->mtime == b->mtime)
924 return 0;
925 return -1;
926 }
927
928 static void rearrange_packed_git(struct repository *r)
929 {
930 r->objects->packed_git = llist_mergesort(
931 r->objects->packed_git, get_next_packed_git,
932 set_next_packed_git, sort_pack);
933 }
934
935 static void prepare_packed_git_mru(struct repository *r)
936 {
937 struct packed_git *p;
938
939 INIT_LIST_HEAD(&r->objects->packed_git_mru);
940
941 for (p = r->objects->packed_git; p; p = p->next)
942 list_add_tail(&p->mru, &r->objects->packed_git_mru);
943 }
944
945 static void prepare_packed_git(struct repository *r)
946 {
947 struct alternate_object_database *alt;
948
949 if (r->objects->packed_git_initialized)
950 return;
951 prepare_multi_pack_index_one(r, r->objects->objectdir);
952 prepare_packed_git_one(r, r->objects->objectdir, 1);
953 prepare_alt_odb(r);
954 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
955 prepare_multi_pack_index_one(r, alt->path);
956 prepare_packed_git_one(r, alt->path, 0);
957 }
958 rearrange_packed_git(r);
959 prepare_packed_git_mru(r);
960 r->objects->packed_git_initialized = 1;
961 }
962
963 void reprepare_packed_git(struct repository *r)
964 {
965 r->objects->approximate_object_count_valid = 0;
966 r->objects->packed_git_initialized = 0;
967 prepare_packed_git(r);
968 }
969
970 struct packed_git *get_packed_git(struct repository *r)
971 {
972 prepare_packed_git(r);
973 return r->objects->packed_git;
974 }
975
976 struct multi_pack_index *get_multi_pack_index(struct repository *r)
977 {
978 prepare_packed_git(r);
979 return r->objects->multi_pack_index;
980 }
981
982 struct list_head *get_packed_git_mru(struct repository *r)
983 {
984 prepare_packed_git(r);
985 return &r->objects->packed_git_mru;
986 }
987
988 unsigned long unpack_object_header_buffer(const unsigned char *buf,
989 unsigned long len, enum object_type *type, unsigned long *sizep)
990 {
991 unsigned shift;
992 unsigned long size, c;
993 unsigned long used = 0;
994
995 c = buf[used++];
996 *type = (c >> 4) & 7;
997 size = c & 15;
998 shift = 4;
999 while (c & 0x80) {
1000 if (len <= used || bitsizeof(long) <= shift) {
1001 error("bad object header");
1002 size = used = 0;
1003 break;
1004 }
1005 c = buf[used++];
1006 size += (c & 0x7f) << shift;
1007 shift += 7;
1008 }
1009 *sizep = size;
1010 return used;
1011 }
1012
1013 unsigned long get_size_from_delta(struct packed_git *p,
1014 struct pack_window **w_curs,
1015 off_t curpos)
1016 {
1017 const unsigned char *data;
1018 unsigned char delta_head[20], *in;
1019 git_zstream stream;
1020 int st;
1021
1022 memset(&stream, 0, sizeof(stream));
1023 stream.next_out = delta_head;
1024 stream.avail_out = sizeof(delta_head);
1025
1026 git_inflate_init(&stream);
1027 do {
1028 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1029 stream.next_in = in;
1030 st = git_inflate(&stream, Z_FINISH);
1031 curpos += stream.next_in - in;
1032 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1033 stream.total_out < sizeof(delta_head));
1034 git_inflate_end(&stream);
1035 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1036 error("delta data unpack-initial failed");
1037 return 0;
1038 }
1039
1040 /* Examine the initial part of the delta to figure out
1041 * the result size.
1042 */
1043 data = delta_head;
1044
1045 /* ignore base size */
1046 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1047
1048 /* Read the result size */
1049 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1050 }
1051
1052 int unpack_object_header(struct packed_git *p,
1053 struct pack_window **w_curs,
1054 off_t *curpos,
1055 unsigned long *sizep)
1056 {
1057 unsigned char *base;
1058 unsigned long left;
1059 unsigned long used;
1060 enum object_type type;
1061
1062 /* use_pack() assures us we have [base, base + 20) available
1063 * as a range that we can look at. (Its actually the hash
1064 * size that is assured.) With our object header encoding
1065 * the maximum deflated object size is 2^137, which is just
1066 * insane, so we know won't exceed what we have been given.
1067 */
1068 base = use_pack(p, w_curs, *curpos, &left);
1069 used = unpack_object_header_buffer(base, left, &type, sizep);
1070 if (!used) {
1071 type = OBJ_BAD;
1072 } else
1073 *curpos += used;
1074
1075 return type;
1076 }
1077
1078 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1079 {
1080 unsigned i;
1081 for (i = 0; i < p->num_bad_objects; i++)
1082 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1083 return;
1084 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1085 st_mult(GIT_MAX_RAWSZ,
1086 st_add(p->num_bad_objects, 1)));
1087 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1088 p->num_bad_objects++;
1089 }
1090
1091 const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1092 {
1093 struct packed_git *p;
1094 unsigned i;
1095
1096 for (p = the_repository->objects->packed_git; p; p = p->next)
1097 for (i = 0; i < p->num_bad_objects; i++)
1098 if (!hashcmp(sha1,
1099 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1100 return p;
1101 return NULL;
1102 }
1103
1104 static off_t get_delta_base(struct packed_git *p,
1105 struct pack_window **w_curs,
1106 off_t *curpos,
1107 enum object_type type,
1108 off_t delta_obj_offset)
1109 {
1110 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1111 off_t base_offset;
1112
1113 /* use_pack() assured us we have [base_info, base_info + 20)
1114 * as a range that we can look at without walking off the
1115 * end of the mapped window. Its actually the hash size
1116 * that is assured. An OFS_DELTA longer than the hash size
1117 * is stupid, as then a REF_DELTA would be smaller to store.
1118 */
1119 if (type == OBJ_OFS_DELTA) {
1120 unsigned used = 0;
1121 unsigned char c = base_info[used++];
1122 base_offset = c & 127;
1123 while (c & 128) {
1124 base_offset += 1;
1125 if (!base_offset || MSB(base_offset, 7))
1126 return 0; /* overflow */
1127 c = base_info[used++];
1128 base_offset = (base_offset << 7) + (c & 127);
1129 }
1130 base_offset = delta_obj_offset - base_offset;
1131 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1132 return 0; /* out of bound */
1133 *curpos += used;
1134 } else if (type == OBJ_REF_DELTA) {
1135 /* The base entry _must_ be in the same pack */
1136 base_offset = find_pack_entry_one(base_info, p);
1137 *curpos += the_hash_algo->rawsz;
1138 } else
1139 die("I am totally screwed");
1140 return base_offset;
1141 }
1142
1143 /*
1144 * Like get_delta_base above, but we return the sha1 instead of the pack
1145 * offset. This means it is cheaper for REF deltas (we do not have to do
1146 * the final object lookup), but more expensive for OFS deltas (we
1147 * have to load the revidx to convert the offset back into a sha1).
1148 */
1149 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1150 struct pack_window **w_curs,
1151 off_t curpos,
1152 enum object_type type,
1153 off_t delta_obj_offset)
1154 {
1155 if (type == OBJ_REF_DELTA) {
1156 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1157 return base;
1158 } else if (type == OBJ_OFS_DELTA) {
1159 struct revindex_entry *revidx;
1160 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1161 type, delta_obj_offset);
1162
1163 if (!base_offset)
1164 return NULL;
1165
1166 revidx = find_pack_revindex(p, base_offset);
1167 if (!revidx)
1168 return NULL;
1169
1170 return nth_packed_object_sha1(p, revidx->nr);
1171 } else
1172 return NULL;
1173 }
1174
1175 static int retry_bad_packed_offset(struct repository *r,
1176 struct packed_git *p,
1177 off_t obj_offset)
1178 {
1179 int type;
1180 struct revindex_entry *revidx;
1181 struct object_id oid;
1182 revidx = find_pack_revindex(p, obj_offset);
1183 if (!revidx)
1184 return OBJ_BAD;
1185 nth_packed_object_oid(&oid, p, revidx->nr);
1186 mark_bad_packed_object(p, oid.hash);
1187 type = oid_object_info(r, &oid, NULL);
1188 if (type <= OBJ_NONE)
1189 return OBJ_BAD;
1190 return type;
1191 }
1192
1193 #define POI_STACK_PREALLOC 64
1194
1195 static enum object_type packed_to_object_type(struct repository *r,
1196 struct packed_git *p,
1197 off_t obj_offset,
1198 enum object_type type,
1199 struct pack_window **w_curs,
1200 off_t curpos)
1201 {
1202 off_t small_poi_stack[POI_STACK_PREALLOC];
1203 off_t *poi_stack = small_poi_stack;
1204 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1205
1206 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1207 off_t base_offset;
1208 unsigned long size;
1209 /* Push the object we're going to leave behind */
1210 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1211 poi_stack_alloc = alloc_nr(poi_stack_nr);
1212 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1213 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1214 } else {
1215 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1216 }
1217 poi_stack[poi_stack_nr++] = obj_offset;
1218 /* If parsing the base offset fails, just unwind */
1219 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1220 if (!base_offset)
1221 goto unwind;
1222 curpos = obj_offset = base_offset;
1223 type = unpack_object_header(p, w_curs, &curpos, &size);
1224 if (type <= OBJ_NONE) {
1225 /* If getting the base itself fails, we first
1226 * retry the base, otherwise unwind */
1227 type = retry_bad_packed_offset(r, p, base_offset);
1228 if (type > OBJ_NONE)
1229 goto out;
1230 goto unwind;
1231 }
1232 }
1233
1234 switch (type) {
1235 case OBJ_BAD:
1236 case OBJ_COMMIT:
1237 case OBJ_TREE:
1238 case OBJ_BLOB:
1239 case OBJ_TAG:
1240 break;
1241 default:
1242 error("unknown object type %i at offset %"PRIuMAX" in %s",
1243 type, (uintmax_t)obj_offset, p->pack_name);
1244 type = OBJ_BAD;
1245 }
1246
1247 out:
1248 if (poi_stack != small_poi_stack)
1249 free(poi_stack);
1250 return type;
1251
1252 unwind:
1253 while (poi_stack_nr) {
1254 obj_offset = poi_stack[--poi_stack_nr];
1255 type = retry_bad_packed_offset(r, p, obj_offset);
1256 if (type > OBJ_NONE)
1257 goto out;
1258 }
1259 type = OBJ_BAD;
1260 goto out;
1261 }
1262
1263 static struct hashmap delta_base_cache;
1264 static size_t delta_base_cached;
1265
1266 static LIST_HEAD(delta_base_cache_lru);
1267
1268 struct delta_base_cache_key {
1269 struct packed_git *p;
1270 off_t base_offset;
1271 };
1272
1273 struct delta_base_cache_entry {
1274 struct hashmap hash;
1275 struct delta_base_cache_key key;
1276 struct list_head lru;
1277 void *data;
1278 unsigned long size;
1279 enum object_type type;
1280 };
1281
1282 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1283 {
1284 unsigned int hash;
1285
1286 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1287 hash += (hash >> 8) + (hash >> 16);
1288 return hash;
1289 }
1290
1291 static struct delta_base_cache_entry *
1292 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1293 {
1294 struct hashmap_entry entry;
1295 struct delta_base_cache_key key;
1296
1297 if (!delta_base_cache.cmpfn)
1298 return NULL;
1299
1300 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1301 key.p = p;
1302 key.base_offset = base_offset;
1303 return hashmap_get(&delta_base_cache, &entry, &key);
1304 }
1305
1306 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1307 const struct delta_base_cache_key *b)
1308 {
1309 return a->p == b->p && a->base_offset == b->base_offset;
1310 }
1311
1312 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1313 const void *va, const void *vb,
1314 const void *vkey)
1315 {
1316 const struct delta_base_cache_entry *a = va, *b = vb;
1317 const struct delta_base_cache_key *key = vkey;
1318 if (key)
1319 return !delta_base_cache_key_eq(&a->key, key);
1320 else
1321 return !delta_base_cache_key_eq(&a->key, &b->key);
1322 }
1323
1324 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1325 {
1326 return !!get_delta_base_cache_entry(p, base_offset);
1327 }
1328
1329 /*
1330 * Remove the entry from the cache, but do _not_ free the associated
1331 * entry data. The caller takes ownership of the "data" buffer, and
1332 * should copy out any fields it wants before detaching.
1333 */
1334 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1335 {
1336 hashmap_remove(&delta_base_cache, ent, &ent->key);
1337 list_del(&ent->lru);
1338 delta_base_cached -= ent->size;
1339 free(ent);
1340 }
1341
1342 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1343 off_t base_offset, unsigned long *base_size,
1344 enum object_type *type)
1345 {
1346 struct delta_base_cache_entry *ent;
1347
1348 ent = get_delta_base_cache_entry(p, base_offset);
1349 if (!ent)
1350 return unpack_entry(r, p, base_offset, type, base_size);
1351
1352 if (type)
1353 *type = ent->type;
1354 if (base_size)
1355 *base_size = ent->size;
1356 return xmemdupz(ent->data, ent->size);
1357 }
1358
1359 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1360 {
1361 free(ent->data);
1362 detach_delta_base_cache_entry(ent);
1363 }
1364
1365 void clear_delta_base_cache(void)
1366 {
1367 struct list_head *lru, *tmp;
1368 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1369 struct delta_base_cache_entry *entry =
1370 list_entry(lru, struct delta_base_cache_entry, lru);
1371 release_delta_base_cache(entry);
1372 }
1373 }
1374
1375 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1376 void *base, unsigned long base_size, enum object_type type)
1377 {
1378 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1379 struct list_head *lru, *tmp;
1380
1381 delta_base_cached += base_size;
1382
1383 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1384 struct delta_base_cache_entry *f =
1385 list_entry(lru, struct delta_base_cache_entry, lru);
1386 if (delta_base_cached <= delta_base_cache_limit)
1387 break;
1388 release_delta_base_cache(f);
1389 }
1390
1391 ent->key.p = p;
1392 ent->key.base_offset = base_offset;
1393 ent->type = type;
1394 ent->data = base;
1395 ent->size = base_size;
1396 list_add_tail(&ent->lru, &delta_base_cache_lru);
1397
1398 if (!delta_base_cache.cmpfn)
1399 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1400 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1401 hashmap_add(&delta_base_cache, ent);
1402 }
1403
1404 int packed_object_info(struct repository *r, struct packed_git *p,
1405 off_t obj_offset, struct object_info *oi)
1406 {
1407 struct pack_window *w_curs = NULL;
1408 unsigned long size;
1409 off_t curpos = obj_offset;
1410 enum object_type type;
1411
1412 /*
1413 * We always get the representation type, but only convert it to
1414 * a "real" type later if the caller is interested.
1415 */
1416 if (oi->contentp) {
1417 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1418 &type);
1419 if (!*oi->contentp)
1420 type = OBJ_BAD;
1421 } else {
1422 type = unpack_object_header(p, &w_curs, &curpos, &size);
1423 }
1424
1425 if (!oi->contentp && oi->sizep) {
1426 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1427 off_t tmp_pos = curpos;
1428 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1429 type, obj_offset);
1430 if (!base_offset) {
1431 type = OBJ_BAD;
1432 goto out;
1433 }
1434 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1435 if (*oi->sizep == 0) {
1436 type = OBJ_BAD;
1437 goto out;
1438 }
1439 } else {
1440 *oi->sizep = size;
1441 }
1442 }
1443
1444 if (oi->disk_sizep) {
1445 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1446 *oi->disk_sizep = revidx[1].offset - obj_offset;
1447 }
1448
1449 if (oi->typep || oi->type_name) {
1450 enum object_type ptot;
1451 ptot = packed_to_object_type(r, p, obj_offset,
1452 type, &w_curs, curpos);
1453 if (oi->typep)
1454 *oi->typep = ptot;
1455 if (oi->type_name) {
1456 const char *tn = type_name(ptot);
1457 if (tn)
1458 strbuf_addstr(oi->type_name, tn);
1459 }
1460 if (ptot < 0) {
1461 type = OBJ_BAD;
1462 goto out;
1463 }
1464 }
1465
1466 if (oi->delta_base_sha1) {
1467 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1468 const unsigned char *base;
1469
1470 base = get_delta_base_sha1(p, &w_curs, curpos,
1471 type, obj_offset);
1472 if (!base) {
1473 type = OBJ_BAD;
1474 goto out;
1475 }
1476
1477 hashcpy(oi->delta_base_sha1, base);
1478 } else
1479 hashclr(oi->delta_base_sha1);
1480 }
1481
1482 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1483 OI_PACKED;
1484
1485 out:
1486 unuse_pack(&w_curs);
1487 return type;
1488 }
1489
1490 static void *unpack_compressed_entry(struct packed_git *p,
1491 struct pack_window **w_curs,
1492 off_t curpos,
1493 unsigned long size)
1494 {
1495 int st;
1496 git_zstream stream;
1497 unsigned char *buffer, *in;
1498
1499 buffer = xmallocz_gently(size);
1500 if (!buffer)
1501 return NULL;
1502 memset(&stream, 0, sizeof(stream));
1503 stream.next_out = buffer;
1504 stream.avail_out = size + 1;
1505
1506 git_inflate_init(&stream);
1507 do {
1508 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1509 stream.next_in = in;
1510 st = git_inflate(&stream, Z_FINISH);
1511 if (!stream.avail_out)
1512 break; /* the payload is larger than it should be */
1513 curpos += stream.next_in - in;
1514 } while (st == Z_OK || st == Z_BUF_ERROR);
1515 git_inflate_end(&stream);
1516 if ((st != Z_STREAM_END) || stream.total_out != size) {
1517 free(buffer);
1518 return NULL;
1519 }
1520
1521 /* versions of zlib can clobber unconsumed portion of outbuf */
1522 buffer[size] = '\0';
1523
1524 return buffer;
1525 }
1526
1527 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1528 {
1529 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1530 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1531 p->pack_name, (uintmax_t)obj_offset);
1532 }
1533
1534 int do_check_packed_object_crc;
1535
1536 #define UNPACK_ENTRY_STACK_PREALLOC 64
1537 struct unpack_entry_stack_ent {
1538 off_t obj_offset;
1539 off_t curpos;
1540 unsigned long size;
1541 };
1542
1543 static void *read_object(struct repository *r,
1544 const struct object_id *oid,
1545 enum object_type *type,
1546 unsigned long *size)
1547 {
1548 struct object_info oi = OBJECT_INFO_INIT;
1549 void *content;
1550 oi.typep = type;
1551 oi.sizep = size;
1552 oi.contentp = &content;
1553
1554 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1555 return NULL;
1556 return content;
1557 }
1558
1559 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1560 enum object_type *final_type, unsigned long *final_size)
1561 {
1562 struct pack_window *w_curs = NULL;
1563 off_t curpos = obj_offset;
1564 void *data = NULL;
1565 unsigned long size;
1566 enum object_type type;
1567 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1568 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1569 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1570 int base_from_cache = 0;
1571
1572 write_pack_access_log(p, obj_offset);
1573
1574 /* PHASE 1: drill down to the innermost base object */
1575 for (;;) {
1576 off_t base_offset;
1577 int i;
1578 struct delta_base_cache_entry *ent;
1579
1580 ent = get_delta_base_cache_entry(p, curpos);
1581 if (ent) {
1582 type = ent->type;
1583 data = ent->data;
1584 size = ent->size;
1585 detach_delta_base_cache_entry(ent);
1586 base_from_cache = 1;
1587 break;
1588 }
1589
1590 if (do_check_packed_object_crc && p->index_version > 1) {
1591 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1592 off_t len = revidx[1].offset - obj_offset;
1593 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1594 struct object_id oid;
1595 nth_packed_object_oid(&oid, p, revidx->nr);
1596 error("bad packed object CRC for %s",
1597 oid_to_hex(&oid));
1598 mark_bad_packed_object(p, oid.hash);
1599 data = NULL;
1600 goto out;
1601 }
1602 }
1603
1604 type = unpack_object_header(p, &w_curs, &curpos, &size);
1605 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1606 break;
1607
1608 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1609 if (!base_offset) {
1610 error("failed to validate delta base reference "
1611 "at offset %"PRIuMAX" from %s",
1612 (uintmax_t)curpos, p->pack_name);
1613 /* bail to phase 2, in hopes of recovery */
1614 data = NULL;
1615 break;
1616 }
1617
1618 /* push object, proceed to base */
1619 if (delta_stack_nr >= delta_stack_alloc
1620 && delta_stack == small_delta_stack) {
1621 delta_stack_alloc = alloc_nr(delta_stack_nr);
1622 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1623 memcpy(delta_stack, small_delta_stack,
1624 sizeof(*delta_stack)*delta_stack_nr);
1625 } else {
1626 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1627 }
1628 i = delta_stack_nr++;
1629 delta_stack[i].obj_offset = obj_offset;
1630 delta_stack[i].curpos = curpos;
1631 delta_stack[i].size = size;
1632
1633 curpos = obj_offset = base_offset;
1634 }
1635
1636 /* PHASE 2: handle the base */
1637 switch (type) {
1638 case OBJ_OFS_DELTA:
1639 case OBJ_REF_DELTA:
1640 if (data)
1641 BUG("unpack_entry: left loop at a valid delta");
1642 break;
1643 case OBJ_COMMIT:
1644 case OBJ_TREE:
1645 case OBJ_BLOB:
1646 case OBJ_TAG:
1647 if (!base_from_cache)
1648 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1649 break;
1650 default:
1651 data = NULL;
1652 error("unknown object type %i at offset %"PRIuMAX" in %s",
1653 type, (uintmax_t)obj_offset, p->pack_name);
1654 }
1655
1656 /* PHASE 3: apply deltas in order */
1657
1658 /* invariants:
1659 * 'data' holds the base data, or NULL if there was corruption
1660 */
1661 while (delta_stack_nr) {
1662 void *delta_data;
1663 void *base = data;
1664 void *external_base = NULL;
1665 unsigned long delta_size, base_size = size;
1666 int i;
1667
1668 data = NULL;
1669
1670 if (base)
1671 add_delta_base_cache(p, obj_offset, base, base_size, type);
1672
1673 if (!base) {
1674 /*
1675 * We're probably in deep shit, but let's try to fetch
1676 * the required base anyway from another pack or loose.
1677 * This is costly but should happen only in the presence
1678 * of a corrupted pack, and is better than failing outright.
1679 */
1680 struct revindex_entry *revidx;
1681 struct object_id base_oid;
1682 revidx = find_pack_revindex(p, obj_offset);
1683 if (revidx) {
1684 nth_packed_object_oid(&base_oid, p, revidx->nr);
1685 error("failed to read delta base object %s"
1686 " at offset %"PRIuMAX" from %s",
1687 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1688 p->pack_name);
1689 mark_bad_packed_object(p, base_oid.hash);
1690 base = read_object(r, &base_oid, &type, &base_size);
1691 external_base = base;
1692 }
1693 }
1694
1695 i = --delta_stack_nr;
1696 obj_offset = delta_stack[i].obj_offset;
1697 curpos = delta_stack[i].curpos;
1698 delta_size = delta_stack[i].size;
1699
1700 if (!base)
1701 continue;
1702
1703 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1704
1705 if (!delta_data) {
1706 error("failed to unpack compressed delta "
1707 "at offset %"PRIuMAX" from %s",
1708 (uintmax_t)curpos, p->pack_name);
1709 data = NULL;
1710 free(external_base);
1711 continue;
1712 }
1713
1714 data = patch_delta(base, base_size,
1715 delta_data, delta_size,
1716 &size);
1717
1718 /*
1719 * We could not apply the delta; warn the user, but keep going.
1720 * Our failure will be noticed either in the next iteration of
1721 * the loop, or if this is the final delta, in the caller when
1722 * we return NULL. Those code paths will take care of making
1723 * a more explicit warning and retrying with another copy of
1724 * the object.
1725 */
1726 if (!data)
1727 error("failed to apply delta");
1728
1729 free(delta_data);
1730 free(external_base);
1731 }
1732
1733 if (final_type)
1734 *final_type = type;
1735 if (final_size)
1736 *final_size = size;
1737
1738 out:
1739 unuse_pack(&w_curs);
1740
1741 if (delta_stack != small_delta_stack)
1742 free(delta_stack);
1743
1744 return data;
1745 }
1746
1747 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1748 {
1749 const unsigned char *index_fanout = p->index_data;
1750 const unsigned char *index_lookup;
1751 const unsigned int hashsz = the_hash_algo->rawsz;
1752 int index_lookup_width;
1753
1754 if (!index_fanout)
1755 BUG("bsearch_pack called without a valid pack-index");
1756
1757 index_lookup = index_fanout + 4 * 256;
1758 if (p->index_version == 1) {
1759 index_lookup_width = hashsz + 4;
1760 index_lookup += 4;
1761 } else {
1762 index_lookup_width = hashsz;
1763 index_fanout += 8;
1764 index_lookup += 8;
1765 }
1766
1767 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1768 index_lookup, index_lookup_width, result);
1769 }
1770
1771 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1772 uint32_t n)
1773 {
1774 const unsigned char *index = p->index_data;
1775 const unsigned int hashsz = the_hash_algo->rawsz;
1776 if (!index) {
1777 if (open_pack_index(p))
1778 return NULL;
1779 index = p->index_data;
1780 }
1781 if (n >= p->num_objects)
1782 return NULL;
1783 index += 4 * 256;
1784 if (p->index_version == 1) {
1785 return index + (hashsz + 4) * n + 4;
1786 } else {
1787 index += 8;
1788 return index + hashsz * n;
1789 }
1790 }
1791
1792 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1793 struct packed_git *p,
1794 uint32_t n)
1795 {
1796 const unsigned char *hash = nth_packed_object_sha1(p, n);
1797 if (!hash)
1798 return NULL;
1799 hashcpy(oid->hash, hash);
1800 return oid;
1801 }
1802
1803 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1804 {
1805 const unsigned char *ptr = vptr;
1806 const unsigned char *start = p->index_data;
1807 const unsigned char *end = start + p->index_size;
1808 if (ptr < start)
1809 die(_("offset before start of pack index for %s (corrupt index?)"),
1810 p->pack_name);
1811 /* No need to check for underflow; .idx files must be at least 8 bytes */
1812 if (ptr >= end - 8)
1813 die(_("offset beyond end of pack index for %s (truncated index?)"),
1814 p->pack_name);
1815 }
1816
1817 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1818 {
1819 const unsigned char *index = p->index_data;
1820 const unsigned int hashsz = the_hash_algo->rawsz;
1821 index += 4 * 256;
1822 if (p->index_version == 1) {
1823 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1824 } else {
1825 uint32_t off;
1826 index += 8 + p->num_objects * (hashsz + 4);
1827 off = ntohl(*((uint32_t *)(index + 4 * n)));
1828 if (!(off & 0x80000000))
1829 return off;
1830 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1831 check_pack_index_ptr(p, index);
1832 return get_be64(index);
1833 }
1834 }
1835
1836 off_t find_pack_entry_one(const unsigned char *sha1,
1837 struct packed_git *p)
1838 {
1839 const unsigned char *index = p->index_data;
1840 struct object_id oid;
1841 uint32_t result;
1842
1843 if (!index) {
1844 if (open_pack_index(p))
1845 return 0;
1846 }
1847
1848 hashcpy(oid.hash, sha1);
1849 if (bsearch_pack(&oid, p, &result))
1850 return nth_packed_object_offset(p, result);
1851 return 0;
1852 }
1853
1854 int is_pack_valid(struct packed_git *p)
1855 {
1856 /* An already open pack is known to be valid. */
1857 if (p->pack_fd != -1)
1858 return 1;
1859
1860 /* If the pack has one window completely covering the
1861 * file size, the pack is known to be valid even if
1862 * the descriptor is not currently open.
1863 */
1864 if (p->windows) {
1865 struct pack_window *w = p->windows;
1866
1867 if (!w->offset && w->len == p->pack_size)
1868 return 1;
1869 }
1870
1871 /* Force the pack to open to prove its valid. */
1872 return !open_packed_git(p);
1873 }
1874
1875 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1876 struct packed_git *packs)
1877 {
1878 struct packed_git *p;
1879
1880 for (p = packs; p; p = p->next) {
1881 if (find_pack_entry_one(sha1, p))
1882 return p;
1883 }
1884 return NULL;
1885
1886 }
1887
1888 static int fill_pack_entry(const struct object_id *oid,
1889 struct pack_entry *e,
1890 struct packed_git *p)
1891 {
1892 off_t offset;
1893
1894 if (p->num_bad_objects) {
1895 unsigned i;
1896 for (i = 0; i < p->num_bad_objects; i++)
1897 if (!hashcmp(oid->hash,
1898 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1899 return 0;
1900 }
1901
1902 offset = find_pack_entry_one(oid->hash, p);
1903 if (!offset)
1904 return 0;
1905
1906 /*
1907 * We are about to tell the caller where they can locate the
1908 * requested object. We better make sure the packfile is
1909 * still here and can be accessed before supplying that
1910 * answer, as it may have been deleted since the index was
1911 * loaded!
1912 */
1913 if (!is_pack_valid(p))
1914 return 0;
1915 e->offset = offset;
1916 e->p = p;
1917 return 1;
1918 }
1919
1920 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1921 {
1922 struct list_head *pos;
1923 struct multi_pack_index *m;
1924
1925 prepare_packed_git(r);
1926 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1927 return 0;
1928
1929 for (m = r->objects->multi_pack_index; m; m = m->next) {
1930 if (fill_midx_entry(oid, e, m))
1931 return 1;
1932 }
1933
1934 list_for_each(pos, &r->objects->packed_git_mru) {
1935 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1936 if (fill_pack_entry(oid, e, p)) {
1937 list_move(&p->mru, &r->objects->packed_git_mru);
1938 return 1;
1939 }
1940 }
1941 return 0;
1942 }
1943
1944 int has_object_pack(const struct object_id *oid)
1945 {
1946 struct pack_entry e;
1947 return find_pack_entry(the_repository, oid, &e);
1948 }
1949
1950 int has_pack_index(const unsigned char *sha1)
1951 {
1952 struct stat st;
1953 if (stat(sha1_pack_index_name(sha1), &st))
1954 return 0;
1955 return 1;
1956 }
1957
1958 int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1959 {
1960 uint32_t i;
1961 int r = 0;
1962
1963 for (i = 0; i < p->num_objects; i++) {
1964 struct object_id oid;
1965
1966 if (!nth_packed_object_oid(&oid, p, i))
1967 return error("unable to get sha1 of object %u in %s",
1968 i, p->pack_name);
1969
1970 r = cb(&oid, p, i, data);
1971 if (r)
1972 break;
1973 }
1974 return r;
1975 }
1976
1977 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1978 {
1979 struct packed_git *p;
1980 int r = 0;
1981 int pack_errors = 0;
1982
1983 prepare_packed_git(the_repository);
1984 for (p = the_repository->objects->packed_git; p; p = p->next) {
1985 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1986 continue;
1987 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
1988 !p->pack_promisor)
1989 continue;
1990 if (open_pack_index(p)) {
1991 pack_errors = 1;
1992 continue;
1993 }
1994 r = for_each_object_in_pack(p, cb, data);
1995 if (r)
1996 break;
1997 }
1998 return r ? r : pack_errors;
1999 }
2000
2001 static int add_promisor_object(const struct object_id *oid,
2002 struct packed_git *pack,
2003 uint32_t pos,
2004 void *set_)
2005 {
2006 struct oidset *set = set_;
2007 struct object *obj = parse_object(oid);
2008 if (!obj)
2009 return 1;
2010
2011 oidset_insert(set, oid);
2012
2013 /*
2014 * If this is a tree, commit, or tag, the objects it refers
2015 * to are also promisor objects. (Blobs refer to no objects->)
2016 */
2017 if (obj->type == OBJ_TREE) {
2018 struct tree *tree = (struct tree *)obj;
2019 struct tree_desc desc;
2020 struct name_entry entry;
2021 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2022 /*
2023 * Error messages are given when packs are
2024 * verified, so do not print any here.
2025 */
2026 return 0;
2027 while (tree_entry_gently(&desc, &entry))
2028 oidset_insert(set, entry.oid);
2029 } else if (obj->type == OBJ_COMMIT) {
2030 struct commit *commit = (struct commit *) obj;
2031 struct commit_list *parents = commit->parents;
2032
2033 oidset_insert(set, get_commit_tree_oid(commit));
2034 for (; parents; parents = parents->next)
2035 oidset_insert(set, &parents->item->object.oid);
2036 } else if (obj->type == OBJ_TAG) {
2037 struct tag *tag = (struct tag *) obj;
2038 oidset_insert(set, &tag->tagged->oid);
2039 }
2040 return 0;
2041 }
2042
2043 int is_promisor_object(const struct object_id *oid)
2044 {
2045 static struct oidset promisor_objects;
2046 static int promisor_objects_prepared;
2047
2048 if (!promisor_objects_prepared) {
2049 if (repository_format_partial_clone) {
2050 for_each_packed_object(add_promisor_object,
2051 &promisor_objects,
2052 FOR_EACH_OBJECT_PROMISOR_ONLY);
2053 }
2054 promisor_objects_prepared = 1;
2055 }
2056 return oidset_contains(&promisor_objects, oid);
2057 }