c0eb5ac885f5ffe0ccab00b244619b7ce25af45d
[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 };
799
800 static void prepare_pack(const char *full_name, size_t full_name_len,
801 const char *file_name, void *_data)
802 {
803 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
804 struct packed_git *p;
805 size_t base_len = full_name_len;
806
807 if (strip_suffix_mem(full_name, &base_len, ".idx")) {
808 /* Don't reopen a pack we already have. */
809 for (p = data->r->objects->packed_git; p; p = p->next) {
810 size_t len;
811 if (strip_suffix(p->pack_name, ".pack", &len) &&
812 len == base_len &&
813 !memcmp(p->pack_name, full_name, len))
814 break;
815 }
816
817 if (!p) {
818 p = add_packed_git(full_name, full_name_len, data->local);
819 if (p)
820 install_packed_git(data->r, p);
821 }
822 }
823
824 if (!report_garbage)
825 return;
826
827 if (ends_with(file_name, ".idx") ||
828 ends_with(file_name, ".pack") ||
829 ends_with(file_name, ".bitmap") ||
830 ends_with(file_name, ".keep") ||
831 ends_with(file_name, ".promisor"))
832 string_list_append(data->garbage, full_name);
833 else
834 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
835 }
836
837 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
838 {
839 struct prepare_pack_data data;
840 struct string_list garbage = STRING_LIST_INIT_DUP;
841
842 data.r = r;
843 data.garbage = &garbage;
844 data.local = local;
845
846 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
847
848 report_pack_garbage(data.garbage);
849 string_list_clear(data.garbage, 0);
850 }
851
852 static void prepare_packed_git(struct repository *r);
853 /*
854 * Give a fast, rough count of the number of objects in the repository. This
855 * ignores loose objects completely. If you have a lot of them, then either
856 * you should repack because your performance will be awful, or they are
857 * all unreachable objects about to be pruned, in which case they're not really
858 * interesting as a measure of repo size in the first place.
859 */
860 unsigned long approximate_object_count(void)
861 {
862 if (!the_repository->objects->approximate_object_count_valid) {
863 unsigned long count;
864 struct packed_git *p;
865
866 prepare_packed_git(the_repository);
867 count = 0;
868 for (p = the_repository->objects->packed_git; p; p = p->next) {
869 if (open_pack_index(p))
870 continue;
871 count += p->num_objects;
872 }
873 the_repository->objects->approximate_object_count = count;
874 }
875 return the_repository->objects->approximate_object_count;
876 }
877
878 static void *get_next_packed_git(const void *p)
879 {
880 return ((const struct packed_git *)p)->next;
881 }
882
883 static void set_next_packed_git(void *p, void *next)
884 {
885 ((struct packed_git *)p)->next = next;
886 }
887
888 static int sort_pack(const void *a_, const void *b_)
889 {
890 const struct packed_git *a = a_;
891 const struct packed_git *b = b_;
892 int st;
893
894 /*
895 * Local packs tend to contain objects specific to our
896 * variant of the project than remote ones. In addition,
897 * remote ones could be on a network mounted filesystem.
898 * Favor local ones for these reasons.
899 */
900 st = a->pack_local - b->pack_local;
901 if (st)
902 return -st;
903
904 /*
905 * Younger packs tend to contain more recent objects,
906 * and more recent objects tend to get accessed more
907 * often.
908 */
909 if (a->mtime < b->mtime)
910 return 1;
911 else if (a->mtime == b->mtime)
912 return 0;
913 return -1;
914 }
915
916 static void rearrange_packed_git(struct repository *r)
917 {
918 r->objects->packed_git = llist_mergesort(
919 r->objects->packed_git, get_next_packed_git,
920 set_next_packed_git, sort_pack);
921 }
922
923 static void prepare_packed_git_mru(struct repository *r)
924 {
925 struct packed_git *p;
926
927 INIT_LIST_HEAD(&r->objects->packed_git_mru);
928
929 for (p = r->objects->packed_git; p; p = p->next)
930 list_add_tail(&p->mru, &r->objects->packed_git_mru);
931 }
932
933 static void prepare_packed_git(struct repository *r)
934 {
935 struct alternate_object_database *alt;
936
937 if (r->objects->packed_git_initialized)
938 return;
939 prepare_multi_pack_index_one(r, r->objects->objectdir);
940 prepare_packed_git_one(r, r->objects->objectdir, 1);
941 prepare_alt_odb(r);
942 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
943 prepare_multi_pack_index_one(r, alt->path);
944 prepare_packed_git_one(r, alt->path, 0);
945 }
946 rearrange_packed_git(r);
947 prepare_packed_git_mru(r);
948 r->objects->packed_git_initialized = 1;
949 }
950
951 void reprepare_packed_git(struct repository *r)
952 {
953 r->objects->approximate_object_count_valid = 0;
954 r->objects->packed_git_initialized = 0;
955 prepare_packed_git(r);
956 }
957
958 struct packed_git *get_packed_git(struct repository *r)
959 {
960 prepare_packed_git(r);
961 return r->objects->packed_git;
962 }
963
964 struct multi_pack_index *get_multi_pack_index(struct repository *r)
965 {
966 prepare_packed_git(r);
967 return r->objects->multi_pack_index;
968 }
969
970 struct list_head *get_packed_git_mru(struct repository *r)
971 {
972 prepare_packed_git(r);
973 return &r->objects->packed_git_mru;
974 }
975
976 unsigned long unpack_object_header_buffer(const unsigned char *buf,
977 unsigned long len, enum object_type *type, unsigned long *sizep)
978 {
979 unsigned shift;
980 unsigned long size, c;
981 unsigned long used = 0;
982
983 c = buf[used++];
984 *type = (c >> 4) & 7;
985 size = c & 15;
986 shift = 4;
987 while (c & 0x80) {
988 if (len <= used || bitsizeof(long) <= shift) {
989 error("bad object header");
990 size = used = 0;
991 break;
992 }
993 c = buf[used++];
994 size += (c & 0x7f) << shift;
995 shift += 7;
996 }
997 *sizep = size;
998 return used;
999 }
1000
1001 unsigned long get_size_from_delta(struct packed_git *p,
1002 struct pack_window **w_curs,
1003 off_t curpos)
1004 {
1005 const unsigned char *data;
1006 unsigned char delta_head[20], *in;
1007 git_zstream stream;
1008 int st;
1009
1010 memset(&stream, 0, sizeof(stream));
1011 stream.next_out = delta_head;
1012 stream.avail_out = sizeof(delta_head);
1013
1014 git_inflate_init(&stream);
1015 do {
1016 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1017 stream.next_in = in;
1018 st = git_inflate(&stream, Z_FINISH);
1019 curpos += stream.next_in - in;
1020 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1021 stream.total_out < sizeof(delta_head));
1022 git_inflate_end(&stream);
1023 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1024 error("delta data unpack-initial failed");
1025 return 0;
1026 }
1027
1028 /* Examine the initial part of the delta to figure out
1029 * the result size.
1030 */
1031 data = delta_head;
1032
1033 /* ignore base size */
1034 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1035
1036 /* Read the result size */
1037 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1038 }
1039
1040 int unpack_object_header(struct packed_git *p,
1041 struct pack_window **w_curs,
1042 off_t *curpos,
1043 unsigned long *sizep)
1044 {
1045 unsigned char *base;
1046 unsigned long left;
1047 unsigned long used;
1048 enum object_type type;
1049
1050 /* use_pack() assures us we have [base, base + 20) available
1051 * as a range that we can look at. (Its actually the hash
1052 * size that is assured.) With our object header encoding
1053 * the maximum deflated object size is 2^137, which is just
1054 * insane, so we know won't exceed what we have been given.
1055 */
1056 base = use_pack(p, w_curs, *curpos, &left);
1057 used = unpack_object_header_buffer(base, left, &type, sizep);
1058 if (!used) {
1059 type = OBJ_BAD;
1060 } else
1061 *curpos += used;
1062
1063 return type;
1064 }
1065
1066 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1067 {
1068 unsigned i;
1069 for (i = 0; i < p->num_bad_objects; i++)
1070 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1071 return;
1072 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1073 st_mult(GIT_MAX_RAWSZ,
1074 st_add(p->num_bad_objects, 1)));
1075 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1076 p->num_bad_objects++;
1077 }
1078
1079 const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1080 {
1081 struct packed_git *p;
1082 unsigned i;
1083
1084 for (p = the_repository->objects->packed_git; p; p = p->next)
1085 for (i = 0; i < p->num_bad_objects; i++)
1086 if (!hashcmp(sha1,
1087 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1088 return p;
1089 return NULL;
1090 }
1091
1092 static off_t get_delta_base(struct packed_git *p,
1093 struct pack_window **w_curs,
1094 off_t *curpos,
1095 enum object_type type,
1096 off_t delta_obj_offset)
1097 {
1098 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1099 off_t base_offset;
1100
1101 /* use_pack() assured us we have [base_info, base_info + 20)
1102 * as a range that we can look at without walking off the
1103 * end of the mapped window. Its actually the hash size
1104 * that is assured. An OFS_DELTA longer than the hash size
1105 * is stupid, as then a REF_DELTA would be smaller to store.
1106 */
1107 if (type == OBJ_OFS_DELTA) {
1108 unsigned used = 0;
1109 unsigned char c = base_info[used++];
1110 base_offset = c & 127;
1111 while (c & 128) {
1112 base_offset += 1;
1113 if (!base_offset || MSB(base_offset, 7))
1114 return 0; /* overflow */
1115 c = base_info[used++];
1116 base_offset = (base_offset << 7) + (c & 127);
1117 }
1118 base_offset = delta_obj_offset - base_offset;
1119 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1120 return 0; /* out of bound */
1121 *curpos += used;
1122 } else if (type == OBJ_REF_DELTA) {
1123 /* The base entry _must_ be in the same pack */
1124 base_offset = find_pack_entry_one(base_info, p);
1125 *curpos += the_hash_algo->rawsz;
1126 } else
1127 die("I am totally screwed");
1128 return base_offset;
1129 }
1130
1131 /*
1132 * Like get_delta_base above, but we return the sha1 instead of the pack
1133 * offset. This means it is cheaper for REF deltas (we do not have to do
1134 * the final object lookup), but more expensive for OFS deltas (we
1135 * have to load the revidx to convert the offset back into a sha1).
1136 */
1137 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1138 struct pack_window **w_curs,
1139 off_t curpos,
1140 enum object_type type,
1141 off_t delta_obj_offset)
1142 {
1143 if (type == OBJ_REF_DELTA) {
1144 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1145 return base;
1146 } else if (type == OBJ_OFS_DELTA) {
1147 struct revindex_entry *revidx;
1148 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1149 type, delta_obj_offset);
1150
1151 if (!base_offset)
1152 return NULL;
1153
1154 revidx = find_pack_revindex(p, base_offset);
1155 if (!revidx)
1156 return NULL;
1157
1158 return nth_packed_object_sha1(p, revidx->nr);
1159 } else
1160 return NULL;
1161 }
1162
1163 static int retry_bad_packed_offset(struct repository *r,
1164 struct packed_git *p,
1165 off_t obj_offset)
1166 {
1167 int type;
1168 struct revindex_entry *revidx;
1169 struct object_id oid;
1170 revidx = find_pack_revindex(p, obj_offset);
1171 if (!revidx)
1172 return OBJ_BAD;
1173 nth_packed_object_oid(&oid, p, revidx->nr);
1174 mark_bad_packed_object(p, oid.hash);
1175 type = oid_object_info(r, &oid, NULL);
1176 if (type <= OBJ_NONE)
1177 return OBJ_BAD;
1178 return type;
1179 }
1180
1181 #define POI_STACK_PREALLOC 64
1182
1183 static enum object_type packed_to_object_type(struct repository *r,
1184 struct packed_git *p,
1185 off_t obj_offset,
1186 enum object_type type,
1187 struct pack_window **w_curs,
1188 off_t curpos)
1189 {
1190 off_t small_poi_stack[POI_STACK_PREALLOC];
1191 off_t *poi_stack = small_poi_stack;
1192 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1193
1194 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1195 off_t base_offset;
1196 unsigned long size;
1197 /* Push the object we're going to leave behind */
1198 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1199 poi_stack_alloc = alloc_nr(poi_stack_nr);
1200 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1201 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1202 } else {
1203 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1204 }
1205 poi_stack[poi_stack_nr++] = obj_offset;
1206 /* If parsing the base offset fails, just unwind */
1207 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1208 if (!base_offset)
1209 goto unwind;
1210 curpos = obj_offset = base_offset;
1211 type = unpack_object_header(p, w_curs, &curpos, &size);
1212 if (type <= OBJ_NONE) {
1213 /* If getting the base itself fails, we first
1214 * retry the base, otherwise unwind */
1215 type = retry_bad_packed_offset(r, p, base_offset);
1216 if (type > OBJ_NONE)
1217 goto out;
1218 goto unwind;
1219 }
1220 }
1221
1222 switch (type) {
1223 case OBJ_BAD:
1224 case OBJ_COMMIT:
1225 case OBJ_TREE:
1226 case OBJ_BLOB:
1227 case OBJ_TAG:
1228 break;
1229 default:
1230 error("unknown object type %i at offset %"PRIuMAX" in %s",
1231 type, (uintmax_t)obj_offset, p->pack_name);
1232 type = OBJ_BAD;
1233 }
1234
1235 out:
1236 if (poi_stack != small_poi_stack)
1237 free(poi_stack);
1238 return type;
1239
1240 unwind:
1241 while (poi_stack_nr) {
1242 obj_offset = poi_stack[--poi_stack_nr];
1243 type = retry_bad_packed_offset(r, p, obj_offset);
1244 if (type > OBJ_NONE)
1245 goto out;
1246 }
1247 type = OBJ_BAD;
1248 goto out;
1249 }
1250
1251 static struct hashmap delta_base_cache;
1252 static size_t delta_base_cached;
1253
1254 static LIST_HEAD(delta_base_cache_lru);
1255
1256 struct delta_base_cache_key {
1257 struct packed_git *p;
1258 off_t base_offset;
1259 };
1260
1261 struct delta_base_cache_entry {
1262 struct hashmap hash;
1263 struct delta_base_cache_key key;
1264 struct list_head lru;
1265 void *data;
1266 unsigned long size;
1267 enum object_type type;
1268 };
1269
1270 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1271 {
1272 unsigned int hash;
1273
1274 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1275 hash += (hash >> 8) + (hash >> 16);
1276 return hash;
1277 }
1278
1279 static struct delta_base_cache_entry *
1280 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1281 {
1282 struct hashmap_entry entry;
1283 struct delta_base_cache_key key;
1284
1285 if (!delta_base_cache.cmpfn)
1286 return NULL;
1287
1288 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1289 key.p = p;
1290 key.base_offset = base_offset;
1291 return hashmap_get(&delta_base_cache, &entry, &key);
1292 }
1293
1294 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1295 const struct delta_base_cache_key *b)
1296 {
1297 return a->p == b->p && a->base_offset == b->base_offset;
1298 }
1299
1300 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1301 const void *va, const void *vb,
1302 const void *vkey)
1303 {
1304 const struct delta_base_cache_entry *a = va, *b = vb;
1305 const struct delta_base_cache_key *key = vkey;
1306 if (key)
1307 return !delta_base_cache_key_eq(&a->key, key);
1308 else
1309 return !delta_base_cache_key_eq(&a->key, &b->key);
1310 }
1311
1312 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1313 {
1314 return !!get_delta_base_cache_entry(p, base_offset);
1315 }
1316
1317 /*
1318 * Remove the entry from the cache, but do _not_ free the associated
1319 * entry data. The caller takes ownership of the "data" buffer, and
1320 * should copy out any fields it wants before detaching.
1321 */
1322 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1323 {
1324 hashmap_remove(&delta_base_cache, ent, &ent->key);
1325 list_del(&ent->lru);
1326 delta_base_cached -= ent->size;
1327 free(ent);
1328 }
1329
1330 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1331 off_t base_offset, unsigned long *base_size,
1332 enum object_type *type)
1333 {
1334 struct delta_base_cache_entry *ent;
1335
1336 ent = get_delta_base_cache_entry(p, base_offset);
1337 if (!ent)
1338 return unpack_entry(r, p, base_offset, type, base_size);
1339
1340 if (type)
1341 *type = ent->type;
1342 if (base_size)
1343 *base_size = ent->size;
1344 return xmemdupz(ent->data, ent->size);
1345 }
1346
1347 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1348 {
1349 free(ent->data);
1350 detach_delta_base_cache_entry(ent);
1351 }
1352
1353 void clear_delta_base_cache(void)
1354 {
1355 struct list_head *lru, *tmp;
1356 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1357 struct delta_base_cache_entry *entry =
1358 list_entry(lru, struct delta_base_cache_entry, lru);
1359 release_delta_base_cache(entry);
1360 }
1361 }
1362
1363 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1364 void *base, unsigned long base_size, enum object_type type)
1365 {
1366 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1367 struct list_head *lru, *tmp;
1368
1369 delta_base_cached += base_size;
1370
1371 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1372 struct delta_base_cache_entry *f =
1373 list_entry(lru, struct delta_base_cache_entry, lru);
1374 if (delta_base_cached <= delta_base_cache_limit)
1375 break;
1376 release_delta_base_cache(f);
1377 }
1378
1379 ent->key.p = p;
1380 ent->key.base_offset = base_offset;
1381 ent->type = type;
1382 ent->data = base;
1383 ent->size = base_size;
1384 list_add_tail(&ent->lru, &delta_base_cache_lru);
1385
1386 if (!delta_base_cache.cmpfn)
1387 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1388 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1389 hashmap_add(&delta_base_cache, ent);
1390 }
1391
1392 int packed_object_info(struct repository *r, struct packed_git *p,
1393 off_t obj_offset, struct object_info *oi)
1394 {
1395 struct pack_window *w_curs = NULL;
1396 unsigned long size;
1397 off_t curpos = obj_offset;
1398 enum object_type type;
1399
1400 /*
1401 * We always get the representation type, but only convert it to
1402 * a "real" type later if the caller is interested.
1403 */
1404 if (oi->contentp) {
1405 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1406 &type);
1407 if (!*oi->contentp)
1408 type = OBJ_BAD;
1409 } else {
1410 type = unpack_object_header(p, &w_curs, &curpos, &size);
1411 }
1412
1413 if (!oi->contentp && oi->sizep) {
1414 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1415 off_t tmp_pos = curpos;
1416 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1417 type, obj_offset);
1418 if (!base_offset) {
1419 type = OBJ_BAD;
1420 goto out;
1421 }
1422 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1423 if (*oi->sizep == 0) {
1424 type = OBJ_BAD;
1425 goto out;
1426 }
1427 } else {
1428 *oi->sizep = size;
1429 }
1430 }
1431
1432 if (oi->disk_sizep) {
1433 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1434 *oi->disk_sizep = revidx[1].offset - obj_offset;
1435 }
1436
1437 if (oi->typep || oi->type_name) {
1438 enum object_type ptot;
1439 ptot = packed_to_object_type(r, p, obj_offset,
1440 type, &w_curs, curpos);
1441 if (oi->typep)
1442 *oi->typep = ptot;
1443 if (oi->type_name) {
1444 const char *tn = type_name(ptot);
1445 if (tn)
1446 strbuf_addstr(oi->type_name, tn);
1447 }
1448 if (ptot < 0) {
1449 type = OBJ_BAD;
1450 goto out;
1451 }
1452 }
1453
1454 if (oi->delta_base_sha1) {
1455 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1456 const unsigned char *base;
1457
1458 base = get_delta_base_sha1(p, &w_curs, curpos,
1459 type, obj_offset);
1460 if (!base) {
1461 type = OBJ_BAD;
1462 goto out;
1463 }
1464
1465 hashcpy(oi->delta_base_sha1, base);
1466 } else
1467 hashclr(oi->delta_base_sha1);
1468 }
1469
1470 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1471 OI_PACKED;
1472
1473 out:
1474 unuse_pack(&w_curs);
1475 return type;
1476 }
1477
1478 static void *unpack_compressed_entry(struct packed_git *p,
1479 struct pack_window **w_curs,
1480 off_t curpos,
1481 unsigned long size)
1482 {
1483 int st;
1484 git_zstream stream;
1485 unsigned char *buffer, *in;
1486
1487 buffer = xmallocz_gently(size);
1488 if (!buffer)
1489 return NULL;
1490 memset(&stream, 0, sizeof(stream));
1491 stream.next_out = buffer;
1492 stream.avail_out = size + 1;
1493
1494 git_inflate_init(&stream);
1495 do {
1496 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1497 stream.next_in = in;
1498 st = git_inflate(&stream, Z_FINISH);
1499 if (!stream.avail_out)
1500 break; /* the payload is larger than it should be */
1501 curpos += stream.next_in - in;
1502 } while (st == Z_OK || st == Z_BUF_ERROR);
1503 git_inflate_end(&stream);
1504 if ((st != Z_STREAM_END) || stream.total_out != size) {
1505 free(buffer);
1506 return NULL;
1507 }
1508
1509 /* versions of zlib can clobber unconsumed portion of outbuf */
1510 buffer[size] = '\0';
1511
1512 return buffer;
1513 }
1514
1515 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1516 {
1517 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1518 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1519 p->pack_name, (uintmax_t)obj_offset);
1520 }
1521
1522 int do_check_packed_object_crc;
1523
1524 #define UNPACK_ENTRY_STACK_PREALLOC 64
1525 struct unpack_entry_stack_ent {
1526 off_t obj_offset;
1527 off_t curpos;
1528 unsigned long size;
1529 };
1530
1531 static void *read_object(struct repository *r,
1532 const struct object_id *oid,
1533 enum object_type *type,
1534 unsigned long *size)
1535 {
1536 struct object_info oi = OBJECT_INFO_INIT;
1537 void *content;
1538 oi.typep = type;
1539 oi.sizep = size;
1540 oi.contentp = &content;
1541
1542 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1543 return NULL;
1544 return content;
1545 }
1546
1547 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1548 enum object_type *final_type, unsigned long *final_size)
1549 {
1550 struct pack_window *w_curs = NULL;
1551 off_t curpos = obj_offset;
1552 void *data = NULL;
1553 unsigned long size;
1554 enum object_type type;
1555 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1556 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1557 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1558 int base_from_cache = 0;
1559
1560 write_pack_access_log(p, obj_offset);
1561
1562 /* PHASE 1: drill down to the innermost base object */
1563 for (;;) {
1564 off_t base_offset;
1565 int i;
1566 struct delta_base_cache_entry *ent;
1567
1568 ent = get_delta_base_cache_entry(p, curpos);
1569 if (ent) {
1570 type = ent->type;
1571 data = ent->data;
1572 size = ent->size;
1573 detach_delta_base_cache_entry(ent);
1574 base_from_cache = 1;
1575 break;
1576 }
1577
1578 if (do_check_packed_object_crc && p->index_version > 1) {
1579 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1580 off_t len = revidx[1].offset - obj_offset;
1581 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1582 struct object_id oid;
1583 nth_packed_object_oid(&oid, p, revidx->nr);
1584 error("bad packed object CRC for %s",
1585 oid_to_hex(&oid));
1586 mark_bad_packed_object(p, oid.hash);
1587 data = NULL;
1588 goto out;
1589 }
1590 }
1591
1592 type = unpack_object_header(p, &w_curs, &curpos, &size);
1593 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1594 break;
1595
1596 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1597 if (!base_offset) {
1598 error("failed to validate delta base reference "
1599 "at offset %"PRIuMAX" from %s",
1600 (uintmax_t)curpos, p->pack_name);
1601 /* bail to phase 2, in hopes of recovery */
1602 data = NULL;
1603 break;
1604 }
1605
1606 /* push object, proceed to base */
1607 if (delta_stack_nr >= delta_stack_alloc
1608 && delta_stack == small_delta_stack) {
1609 delta_stack_alloc = alloc_nr(delta_stack_nr);
1610 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1611 memcpy(delta_stack, small_delta_stack,
1612 sizeof(*delta_stack)*delta_stack_nr);
1613 } else {
1614 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1615 }
1616 i = delta_stack_nr++;
1617 delta_stack[i].obj_offset = obj_offset;
1618 delta_stack[i].curpos = curpos;
1619 delta_stack[i].size = size;
1620
1621 curpos = obj_offset = base_offset;
1622 }
1623
1624 /* PHASE 2: handle the base */
1625 switch (type) {
1626 case OBJ_OFS_DELTA:
1627 case OBJ_REF_DELTA:
1628 if (data)
1629 BUG("unpack_entry: left loop at a valid delta");
1630 break;
1631 case OBJ_COMMIT:
1632 case OBJ_TREE:
1633 case OBJ_BLOB:
1634 case OBJ_TAG:
1635 if (!base_from_cache)
1636 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1637 break;
1638 default:
1639 data = NULL;
1640 error("unknown object type %i at offset %"PRIuMAX" in %s",
1641 type, (uintmax_t)obj_offset, p->pack_name);
1642 }
1643
1644 /* PHASE 3: apply deltas in order */
1645
1646 /* invariants:
1647 * 'data' holds the base data, or NULL if there was corruption
1648 */
1649 while (delta_stack_nr) {
1650 void *delta_data;
1651 void *base = data;
1652 void *external_base = NULL;
1653 unsigned long delta_size, base_size = size;
1654 int i;
1655
1656 data = NULL;
1657
1658 if (base)
1659 add_delta_base_cache(p, obj_offset, base, base_size, type);
1660
1661 if (!base) {
1662 /*
1663 * We're probably in deep shit, but let's try to fetch
1664 * the required base anyway from another pack or loose.
1665 * This is costly but should happen only in the presence
1666 * of a corrupted pack, and is better than failing outright.
1667 */
1668 struct revindex_entry *revidx;
1669 struct object_id base_oid;
1670 revidx = find_pack_revindex(p, obj_offset);
1671 if (revidx) {
1672 nth_packed_object_oid(&base_oid, p, revidx->nr);
1673 error("failed to read delta base object %s"
1674 " at offset %"PRIuMAX" from %s",
1675 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1676 p->pack_name);
1677 mark_bad_packed_object(p, base_oid.hash);
1678 base = read_object(r, &base_oid, &type, &base_size);
1679 external_base = base;
1680 }
1681 }
1682
1683 i = --delta_stack_nr;
1684 obj_offset = delta_stack[i].obj_offset;
1685 curpos = delta_stack[i].curpos;
1686 delta_size = delta_stack[i].size;
1687
1688 if (!base)
1689 continue;
1690
1691 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1692
1693 if (!delta_data) {
1694 error("failed to unpack compressed delta "
1695 "at offset %"PRIuMAX" from %s",
1696 (uintmax_t)curpos, p->pack_name);
1697 data = NULL;
1698 free(external_base);
1699 continue;
1700 }
1701
1702 data = patch_delta(base, base_size,
1703 delta_data, delta_size,
1704 &size);
1705
1706 /*
1707 * We could not apply the delta; warn the user, but keep going.
1708 * Our failure will be noticed either in the next iteration of
1709 * the loop, or if this is the final delta, in the caller when
1710 * we return NULL. Those code paths will take care of making
1711 * a more explicit warning and retrying with another copy of
1712 * the object.
1713 */
1714 if (!data)
1715 error("failed to apply delta");
1716
1717 free(delta_data);
1718 free(external_base);
1719 }
1720
1721 if (final_type)
1722 *final_type = type;
1723 if (final_size)
1724 *final_size = size;
1725
1726 out:
1727 unuse_pack(&w_curs);
1728
1729 if (delta_stack != small_delta_stack)
1730 free(delta_stack);
1731
1732 return data;
1733 }
1734
1735 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1736 {
1737 const unsigned char *index_fanout = p->index_data;
1738 const unsigned char *index_lookup;
1739 const unsigned int hashsz = the_hash_algo->rawsz;
1740 int index_lookup_width;
1741
1742 if (!index_fanout)
1743 BUG("bsearch_pack called without a valid pack-index");
1744
1745 index_lookup = index_fanout + 4 * 256;
1746 if (p->index_version == 1) {
1747 index_lookup_width = hashsz + 4;
1748 index_lookup += 4;
1749 } else {
1750 index_lookup_width = hashsz;
1751 index_fanout += 8;
1752 index_lookup += 8;
1753 }
1754
1755 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1756 index_lookup, index_lookup_width, result);
1757 }
1758
1759 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1760 uint32_t n)
1761 {
1762 const unsigned char *index = p->index_data;
1763 const unsigned int hashsz = the_hash_algo->rawsz;
1764 if (!index) {
1765 if (open_pack_index(p))
1766 return NULL;
1767 index = p->index_data;
1768 }
1769 if (n >= p->num_objects)
1770 return NULL;
1771 index += 4 * 256;
1772 if (p->index_version == 1) {
1773 return index + (hashsz + 4) * n + 4;
1774 } else {
1775 index += 8;
1776 return index + hashsz * n;
1777 }
1778 }
1779
1780 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1781 struct packed_git *p,
1782 uint32_t n)
1783 {
1784 const unsigned char *hash = nth_packed_object_sha1(p, n);
1785 if (!hash)
1786 return NULL;
1787 hashcpy(oid->hash, hash);
1788 return oid;
1789 }
1790
1791 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1792 {
1793 const unsigned char *ptr = vptr;
1794 const unsigned char *start = p->index_data;
1795 const unsigned char *end = start + p->index_size;
1796 if (ptr < start)
1797 die(_("offset before start of pack index for %s (corrupt index?)"),
1798 p->pack_name);
1799 /* No need to check for underflow; .idx files must be at least 8 bytes */
1800 if (ptr >= end - 8)
1801 die(_("offset beyond end of pack index for %s (truncated index?)"),
1802 p->pack_name);
1803 }
1804
1805 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1806 {
1807 const unsigned char *index = p->index_data;
1808 const unsigned int hashsz = the_hash_algo->rawsz;
1809 index += 4 * 256;
1810 if (p->index_version == 1) {
1811 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1812 } else {
1813 uint32_t off;
1814 index += 8 + p->num_objects * (hashsz + 4);
1815 off = ntohl(*((uint32_t *)(index + 4 * n)));
1816 if (!(off & 0x80000000))
1817 return off;
1818 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1819 check_pack_index_ptr(p, index);
1820 return get_be64(index);
1821 }
1822 }
1823
1824 off_t find_pack_entry_one(const unsigned char *sha1,
1825 struct packed_git *p)
1826 {
1827 const unsigned char *index = p->index_data;
1828 struct object_id oid;
1829 uint32_t result;
1830
1831 if (!index) {
1832 if (open_pack_index(p))
1833 return 0;
1834 }
1835
1836 hashcpy(oid.hash, sha1);
1837 if (bsearch_pack(&oid, p, &result))
1838 return nth_packed_object_offset(p, result);
1839 return 0;
1840 }
1841
1842 int is_pack_valid(struct packed_git *p)
1843 {
1844 /* An already open pack is known to be valid. */
1845 if (p->pack_fd != -1)
1846 return 1;
1847
1848 /* If the pack has one window completely covering the
1849 * file size, the pack is known to be valid even if
1850 * the descriptor is not currently open.
1851 */
1852 if (p->windows) {
1853 struct pack_window *w = p->windows;
1854
1855 if (!w->offset && w->len == p->pack_size)
1856 return 1;
1857 }
1858
1859 /* Force the pack to open to prove its valid. */
1860 return !open_packed_git(p);
1861 }
1862
1863 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1864 struct packed_git *packs)
1865 {
1866 struct packed_git *p;
1867
1868 for (p = packs; p; p = p->next) {
1869 if (find_pack_entry_one(sha1, p))
1870 return p;
1871 }
1872 return NULL;
1873
1874 }
1875
1876 static int fill_pack_entry(const struct object_id *oid,
1877 struct pack_entry *e,
1878 struct packed_git *p)
1879 {
1880 off_t offset;
1881
1882 if (p->num_bad_objects) {
1883 unsigned i;
1884 for (i = 0; i < p->num_bad_objects; i++)
1885 if (!hashcmp(oid->hash,
1886 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1887 return 0;
1888 }
1889
1890 offset = find_pack_entry_one(oid->hash, p);
1891 if (!offset)
1892 return 0;
1893
1894 /*
1895 * We are about to tell the caller where they can locate the
1896 * requested object. We better make sure the packfile is
1897 * still here and can be accessed before supplying that
1898 * answer, as it may have been deleted since the index was
1899 * loaded!
1900 */
1901 if (!is_pack_valid(p))
1902 return 0;
1903 e->offset = offset;
1904 e->p = p;
1905 return 1;
1906 }
1907
1908 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1909 {
1910 struct list_head *pos;
1911 struct multi_pack_index *m;
1912
1913 prepare_packed_git(r);
1914 if (!r->objects->packed_git && !r->objects->multi_pack_index)
1915 return 0;
1916
1917 for (m = r->objects->multi_pack_index; m; m = m->next) {
1918 if (fill_midx_entry(oid, e, m))
1919 return 1;
1920 }
1921
1922 list_for_each(pos, &r->objects->packed_git_mru) {
1923 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1924 if (fill_pack_entry(oid, e, p)) {
1925 list_move(&p->mru, &r->objects->packed_git_mru);
1926 return 1;
1927 }
1928 }
1929 return 0;
1930 }
1931
1932 int has_object_pack(const struct object_id *oid)
1933 {
1934 struct pack_entry e;
1935 return find_pack_entry(the_repository, oid, &e);
1936 }
1937
1938 int has_pack_index(const unsigned char *sha1)
1939 {
1940 struct stat st;
1941 if (stat(sha1_pack_index_name(sha1), &st))
1942 return 0;
1943 return 1;
1944 }
1945
1946 int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1947 {
1948 uint32_t i;
1949 int r = 0;
1950
1951 for (i = 0; i < p->num_objects; i++) {
1952 struct object_id oid;
1953
1954 if (!nth_packed_object_oid(&oid, p, i))
1955 return error("unable to get sha1 of object %u in %s",
1956 i, p->pack_name);
1957
1958 r = cb(&oid, p, i, data);
1959 if (r)
1960 break;
1961 }
1962 return r;
1963 }
1964
1965 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1966 {
1967 struct packed_git *p;
1968 int r = 0;
1969 int pack_errors = 0;
1970
1971 prepare_packed_git(the_repository);
1972 for (p = the_repository->objects->packed_git; p; p = p->next) {
1973 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1974 continue;
1975 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
1976 !p->pack_promisor)
1977 continue;
1978 if (open_pack_index(p)) {
1979 pack_errors = 1;
1980 continue;
1981 }
1982 r = for_each_object_in_pack(p, cb, data);
1983 if (r)
1984 break;
1985 }
1986 return r ? r : pack_errors;
1987 }
1988
1989 static int add_promisor_object(const struct object_id *oid,
1990 struct packed_git *pack,
1991 uint32_t pos,
1992 void *set_)
1993 {
1994 struct oidset *set = set_;
1995 struct object *obj = parse_object(oid);
1996 if (!obj)
1997 return 1;
1998
1999 oidset_insert(set, oid);
2000
2001 /*
2002 * If this is a tree, commit, or tag, the objects it refers
2003 * to are also promisor objects. (Blobs refer to no objects->)
2004 */
2005 if (obj->type == OBJ_TREE) {
2006 struct tree *tree = (struct tree *)obj;
2007 struct tree_desc desc;
2008 struct name_entry entry;
2009 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2010 /*
2011 * Error messages are given when packs are
2012 * verified, so do not print any here.
2013 */
2014 return 0;
2015 while (tree_entry_gently(&desc, &entry))
2016 oidset_insert(set, entry.oid);
2017 } else if (obj->type == OBJ_COMMIT) {
2018 struct commit *commit = (struct commit *) obj;
2019 struct commit_list *parents = commit->parents;
2020
2021 oidset_insert(set, get_commit_tree_oid(commit));
2022 for (; parents; parents = parents->next)
2023 oidset_insert(set, &parents->item->object.oid);
2024 } else if (obj->type == OBJ_TAG) {
2025 struct tag *tag = (struct tag *) obj;
2026 oidset_insert(set, &tag->tagged->oid);
2027 }
2028 return 0;
2029 }
2030
2031 int is_promisor_object(const struct object_id *oid)
2032 {
2033 static struct oidset promisor_objects;
2034 static int promisor_objects_prepared;
2035
2036 if (!promisor_objects_prepared) {
2037 if (repository_format_partial_clone) {
2038 for_each_packed_object(add_promisor_object,
2039 &promisor_objects,
2040 FOR_EACH_OBJECT_PROMISOR_ONLY);
2041 }
2042 promisor_objects_prepared = 1;
2043 }
2044 return oidset_contains(&promisor_objects, oid);
2045 }