repack: refactor pack deletion for future use
[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 size_t idx_size;
84 int fd = git_open(path), ret;
85 struct stat st;
86 const unsigned int hashsz = the_hash_algo->rawsz;
87
88 if (fd < 0)
89 return -1;
90 if (fstat(fd, &st)) {
91 close(fd);
92 return -1;
93 }
94 idx_size = xsize_t(st.st_size);
95 if (idx_size < 4 * 256 + hashsz + hashsz) {
96 close(fd);
97 return error("index file %s is too small", path);
98 }
99 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
100 close(fd);
101
102 ret = load_idx(path, hashsz, idx_map, idx_size, p);
103
104 if (ret)
105 munmap(idx_map, idx_size);
106
107 return ret;
108 }
109
110 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
111 size_t idx_size, struct packed_git *p)
112 {
113 struct pack_idx_header *hdr = idx_map;
114 uint32_t version, nr, i, *index;
115
116 if (idx_size < 4 * 256 + hashsz + hashsz)
117 return error("index file %s is too small", path);
118 if (idx_map == NULL)
119 return error("empty data");
120
121 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
122 version = ntohl(hdr->idx_version);
123 if (version < 2 || version > 2)
124 return error("index file %s is version %"PRIu32
125 " and is not supported by this binary"
126 " (try upgrading GIT to a newer version)",
127 path, version);
128 } else
129 version = 1;
130
131 nr = 0;
132 index = idx_map;
133 if (version > 1)
134 index += 2; /* skip index header */
135 for (i = 0; i < 256; i++) {
136 uint32_t n = ntohl(index[i]);
137 if (n < nr)
138 return error("non-monotonic index %s", path);
139 nr = n;
140 }
141
142 if (version == 1) {
143 /*
144 * Total size:
145 * - 256 index entries 4 bytes each
146 * - 24-byte entries * nr (object ID + 4-byte offset)
147 * - hash of the packfile
148 * - file checksum
149 */
150 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
151 return error("wrong index v1 file size in %s", path);
152 } else if (version == 2) {
153 /*
154 * Minimum size:
155 * - 8 bytes of header
156 * - 256 index entries 4 bytes each
157 * - object ID entry * nr
158 * - 4-byte crc entry * nr
159 * - 4-byte offset entry * nr
160 * - hash of the packfile
161 * - file checksum
162 * And after the 4-byte offset table might be a
163 * variable sized table containing 8-byte entries
164 * for offsets larger than 2^31.
165 */
166 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
167 unsigned long max_size = min_size;
168 if (nr)
169 max_size += (nr - 1)*8;
170 if (idx_size < min_size || idx_size > max_size)
171 return error("wrong index v2 file size in %s", path);
172 if (idx_size != min_size &&
173 /*
174 * make sure we can deal with large pack offsets.
175 * 31-bit signed offset won't be enough, neither
176 * 32-bit unsigned one will be.
177 */
178 (sizeof(off_t) <= 4))
179 return error("pack too large for current definition of off_t in %s", path);
180 }
181
182 p->index_version = version;
183 p->index_data = idx_map;
184 p->index_size = idx_size;
185 p->num_objects = nr;
186 return 0;
187 }
188
189 int open_pack_index(struct packed_git *p)
190 {
191 char *idx_name;
192 size_t len;
193 int ret;
194
195 if (p->index_data)
196 return 0;
197
198 if (!strip_suffix(p->pack_name, ".pack", &len))
199 BUG("pack_name does not end in .pack");
200 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
201 ret = check_packed_git_idx(idx_name, p);
202 free(idx_name);
203 return ret;
204 }
205
206 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
207 {
208 const uint32_t *level1_ofs = p->index_data;
209
210 if (!level1_ofs) {
211 if (open_pack_index(p))
212 return 0;
213 level1_ofs = p->index_data;
214 }
215
216 if (p->index_version > 1) {
217 level1_ofs += 2;
218 }
219
220 return ntohl(level1_ofs[value]);
221 }
222
223 static struct packed_git *alloc_packed_git(int extra)
224 {
225 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
226 memset(p, 0, sizeof(*p));
227 p->pack_fd = -1;
228 return p;
229 }
230
231 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
232 {
233 const char *path = sha1_pack_name(sha1);
234 size_t alloc = st_add(strlen(path), 1);
235 struct packed_git *p = alloc_packed_git(alloc);
236
237 memcpy(p->pack_name, path, alloc); /* includes NUL */
238 hashcpy(p->hash, sha1);
239 if (check_packed_git_idx(idx_path, p)) {
240 free(p);
241 return NULL;
242 }
243
244 return p;
245 }
246
247 static void scan_windows(struct packed_git *p,
248 struct packed_git **lru_p,
249 struct pack_window **lru_w,
250 struct pack_window **lru_l)
251 {
252 struct pack_window *w, *w_l;
253
254 for (w_l = NULL, w = p->windows; w; w = w->next) {
255 if (!w->inuse_cnt) {
256 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
257 *lru_p = p;
258 *lru_w = w;
259 *lru_l = w_l;
260 }
261 }
262 w_l = w;
263 }
264 }
265
266 static int unuse_one_window(struct packed_git *current)
267 {
268 struct packed_git *p, *lru_p = NULL;
269 struct pack_window *lru_w = NULL, *lru_l = NULL;
270
271 if (current)
272 scan_windows(current, &lru_p, &lru_w, &lru_l);
273 for (p = the_repository->objects->packed_git; p; p = p->next)
274 scan_windows(p, &lru_p, &lru_w, &lru_l);
275 if (lru_p) {
276 munmap(lru_w->base, lru_w->len);
277 pack_mapped -= lru_w->len;
278 if (lru_l)
279 lru_l->next = lru_w->next;
280 else
281 lru_p->windows = lru_w->next;
282 free(lru_w);
283 pack_open_windows--;
284 return 1;
285 }
286 return 0;
287 }
288
289 void release_pack_memory(size_t need)
290 {
291 size_t cur = pack_mapped;
292 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
293 ; /* nothing */
294 }
295
296 void close_pack_windows(struct packed_git *p)
297 {
298 while (p->windows) {
299 struct pack_window *w = p->windows;
300
301 if (w->inuse_cnt)
302 die("pack '%s' still has open windows to it",
303 p->pack_name);
304 munmap(w->base, w->len);
305 pack_mapped -= w->len;
306 pack_open_windows--;
307 p->windows = w->next;
308 free(w);
309 }
310 }
311
312 int close_pack_fd(struct packed_git *p)
313 {
314 if (p->pack_fd < 0)
315 return 0;
316
317 close(p->pack_fd);
318 pack_open_fds--;
319 p->pack_fd = -1;
320
321 return 1;
322 }
323
324 void close_pack_index(struct packed_git *p)
325 {
326 if (p->index_data) {
327 munmap((void *)p->index_data, p->index_size);
328 p->index_data = NULL;
329 }
330 }
331
332 void close_pack(struct packed_git *p)
333 {
334 close_pack_windows(p);
335 close_pack_fd(p);
336 close_pack_index(p);
337 }
338
339 void close_all_packs(struct raw_object_store *o)
340 {
341 struct packed_git *p;
342
343 for (p = o->packed_git; p; p = p->next)
344 if (p->do_not_close)
345 BUG("want to close pack marked 'do-not-close'");
346 else
347 close_pack(p);
348
349 if (o->multi_pack_index) {
350 close_midx(o->multi_pack_index);
351 o->multi_pack_index = NULL;
352 }
353 }
354
355 void unlink_pack_path(const char *pack_name, int force_delete)
356 {
357 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
358 int i;
359 struct strbuf buf = STRBUF_INIT;
360 size_t plen;
361
362 strbuf_addstr(&buf, pack_name);
363 strip_suffix_mem(buf.buf, &buf.len, ".pack");
364 plen = buf.len;
365
366 if (!force_delete) {
367 strbuf_addstr(&buf, ".keep");
368 if (!access(buf.buf, F_OK)) {
369 strbuf_release(&buf);
370 return;
371 }
372 }
373
374 for (i = 0; i < ARRAY_SIZE(exts); i++) {
375 strbuf_setlen(&buf, plen);
376 strbuf_addstr(&buf, exts[i]);
377 unlink(buf.buf);
378 }
379
380 strbuf_release(&buf);
381 }
382
383 /*
384 * The LRU pack is the one with the oldest MRU window, preferring packs
385 * with no used windows, or the oldest mtime if it has no windows allocated.
386 */
387 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
388 {
389 struct pack_window *w, *this_mru_w;
390 int has_windows_inuse = 0;
391
392 /*
393 * Reject this pack if it has windows and the previously selected
394 * one does not. If this pack does not have windows, reject
395 * it if the pack file is newer than the previously selected one.
396 */
397 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
398 return;
399
400 for (w = this_mru_w = p->windows; w; w = w->next) {
401 /*
402 * Reject this pack if any of its windows are in use,
403 * but the previously selected pack did not have any
404 * inuse windows. Otherwise, record that this pack
405 * has windows in use.
406 */
407 if (w->inuse_cnt) {
408 if (*accept_windows_inuse)
409 has_windows_inuse = 1;
410 else
411 return;
412 }
413
414 if (w->last_used > this_mru_w->last_used)
415 this_mru_w = w;
416
417 /*
418 * Reject this pack if it has windows that have been
419 * used more recently than the previously selected pack.
420 * If the previously selected pack had windows inuse and
421 * we have not encountered a window in this pack that is
422 * inuse, skip this check since we prefer a pack with no
423 * inuse windows to one that has inuse windows.
424 */
425 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
426 this_mru_w->last_used > (*mru_w)->last_used)
427 return;
428 }
429
430 /*
431 * Select this pack.
432 */
433 *mru_w = this_mru_w;
434 *lru_p = p;
435 *accept_windows_inuse = has_windows_inuse;
436 }
437
438 static int close_one_pack(void)
439 {
440 struct packed_git *p, *lru_p = NULL;
441 struct pack_window *mru_w = NULL;
442 int accept_windows_inuse = 1;
443
444 for (p = the_repository->objects->packed_git; p; p = p->next) {
445 if (p->pack_fd == -1)
446 continue;
447 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
448 }
449
450 if (lru_p)
451 return close_pack_fd(lru_p);
452
453 return 0;
454 }
455
456 static unsigned int get_max_fd_limit(void)
457 {
458 #ifdef RLIMIT_NOFILE
459 {
460 struct rlimit lim;
461
462 if (!getrlimit(RLIMIT_NOFILE, &lim))
463 return lim.rlim_cur;
464 }
465 #endif
466
467 #ifdef _SC_OPEN_MAX
468 {
469 long open_max = sysconf(_SC_OPEN_MAX);
470 if (0 < open_max)
471 return open_max;
472 /*
473 * Otherwise, we got -1 for one of the two
474 * reasons:
475 *
476 * (1) sysconf() did not understand _SC_OPEN_MAX
477 * and signaled an error with -1; or
478 * (2) sysconf() said there is no limit.
479 *
480 * We _could_ clear errno before calling sysconf() to
481 * tell these two cases apart and return a huge number
482 * in the latter case to let the caller cap it to a
483 * value that is not so selfish, but letting the
484 * fallback OPEN_MAX codepath take care of these cases
485 * is a lot simpler.
486 */
487 }
488 #endif
489
490 #ifdef OPEN_MAX
491 return OPEN_MAX;
492 #else
493 return 1; /* see the caller ;-) */
494 #endif
495 }
496
497 const char *pack_basename(struct packed_git *p)
498 {
499 const char *ret = strrchr(p->pack_name, '/');
500 if (ret)
501 ret = ret + 1; /* skip past slash */
502 else
503 ret = p->pack_name; /* we only have a base */
504 return ret;
505 }
506
507 /*
508 * Do not call this directly as this leaks p->pack_fd on error return;
509 * call open_packed_git() instead.
510 */
511 static int open_packed_git_1(struct packed_git *p)
512 {
513 struct stat st;
514 struct pack_header hdr;
515 unsigned char hash[GIT_MAX_RAWSZ];
516 unsigned char *idx_hash;
517 long fd_flag;
518 ssize_t read_result;
519 const unsigned hashsz = the_hash_algo->rawsz;
520
521 if (!p->index_data) {
522 struct multi_pack_index *m;
523 const char *pack_name = pack_basename(p);
524
525 for (m = the_repository->objects->multi_pack_index;
526 m; m = m->next) {
527 if (midx_contains_pack(m, pack_name))
528 break;
529 }
530
531 if (!m && open_pack_index(p))
532 return error("packfile %s index unavailable", p->pack_name);
533 }
534
535 if (!pack_max_fds) {
536 unsigned int max_fds = get_max_fd_limit();
537
538 /* Save 3 for stdin/stdout/stderr, 22 for work */
539 if (25 < max_fds)
540 pack_max_fds = max_fds - 25;
541 else
542 pack_max_fds = 1;
543 }
544
545 while (pack_max_fds <= pack_open_fds && close_one_pack())
546 ; /* nothing */
547
548 p->pack_fd = git_open(p->pack_name);
549 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
550 return -1;
551 pack_open_fds++;
552
553 /* If we created the struct before we had the pack we lack size. */
554 if (!p->pack_size) {
555 if (!S_ISREG(st.st_mode))
556 return error("packfile %s not a regular file", p->pack_name);
557 p->pack_size = st.st_size;
558 } else if (p->pack_size != st.st_size)
559 return error("packfile %s size changed", p->pack_name);
560
561 /* We leave these file descriptors open with sliding mmap;
562 * there is no point keeping them open across exec(), though.
563 */
564 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
565 if (fd_flag < 0)
566 return error("cannot determine file descriptor flags");
567 fd_flag |= FD_CLOEXEC;
568 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
569 return error("cannot set FD_CLOEXEC");
570
571 /* Verify we recognize this pack file format. */
572 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
573 if (read_result < 0)
574 return error_errno("error reading from %s", p->pack_name);
575 if (read_result != sizeof(hdr))
576 return error("file %s is far too short to be a packfile", p->pack_name);
577 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
578 return error("file %s is not a GIT packfile", p->pack_name);
579 if (!pack_version_ok(hdr.hdr_version))
580 return error("packfile %s is version %"PRIu32" and not"
581 " supported (try upgrading GIT to a newer version)",
582 p->pack_name, ntohl(hdr.hdr_version));
583
584 /* Skip index checking if in multi-pack-index */
585 if (!p->index_data)
586 return 0;
587
588 /* Verify the pack matches its index. */
589 if (p->num_objects != ntohl(hdr.hdr_entries))
590 return error("packfile %s claims to have %"PRIu32" objects"
591 " while index indicates %"PRIu32" objects",
592 p->pack_name, ntohl(hdr.hdr_entries),
593 p->num_objects);
594 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
595 return error("end of packfile %s is unavailable", p->pack_name);
596 read_result = read_in_full(p->pack_fd, hash, hashsz);
597 if (read_result < 0)
598 return error_errno("error reading from %s", p->pack_name);
599 if (read_result != hashsz)
600 return error("packfile %s signature is unavailable", p->pack_name);
601 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
602 if (!hasheq(hash, idx_hash))
603 return error("packfile %s does not match index", p->pack_name);
604 return 0;
605 }
606
607 static int open_packed_git(struct packed_git *p)
608 {
609 if (!open_packed_git_1(p))
610 return 0;
611 close_pack_fd(p);
612 return -1;
613 }
614
615 static int in_window(struct pack_window *win, off_t offset)
616 {
617 /* We must promise at least one full hash after the
618 * offset is available from this window, otherwise the offset
619 * is not actually in this window and a different window (which
620 * has that one hash excess) must be used. This is to support
621 * the object header and delta base parsing routines below.
622 */
623 off_t win_off = win->offset;
624 return win_off <= offset
625 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
626 }
627
628 unsigned char *use_pack(struct packed_git *p,
629 struct pack_window **w_cursor,
630 off_t offset,
631 unsigned long *left)
632 {
633 struct pack_window *win = *w_cursor;
634
635 /* Since packfiles end in a hash of their content and it's
636 * pointless to ask for an offset into the middle of that
637 * hash, and the in_window function above wouldn't match
638 * don't allow an offset too close to the end of the file.
639 */
640 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
641 die("packfile %s cannot be accessed", p->pack_name);
642 if (offset > (p->pack_size - the_hash_algo->rawsz))
643 die("offset beyond end of packfile (truncated pack?)");
644 if (offset < 0)
645 die(_("offset before end of packfile (broken .idx?)"));
646
647 if (!win || !in_window(win, offset)) {
648 if (win)
649 win->inuse_cnt--;
650 for (win = p->windows; win; win = win->next) {
651 if (in_window(win, offset))
652 break;
653 }
654 if (!win) {
655 size_t window_align = packed_git_window_size / 2;
656 off_t len;
657
658 if (p->pack_fd == -1 && open_packed_git(p))
659 die("packfile %s cannot be accessed", p->pack_name);
660
661 win = xcalloc(1, sizeof(*win));
662 win->offset = (offset / window_align) * window_align;
663 len = p->pack_size - win->offset;
664 if (len > packed_git_window_size)
665 len = packed_git_window_size;
666 win->len = (size_t)len;
667 pack_mapped += win->len;
668 while (packed_git_limit < pack_mapped
669 && unuse_one_window(p))
670 ; /* nothing */
671 win->base = xmmap(NULL, win->len,
672 PROT_READ, MAP_PRIVATE,
673 p->pack_fd, win->offset);
674 if (win->base == MAP_FAILED)
675 die_errno("packfile %s cannot be mapped",
676 p->pack_name);
677 if (!win->offset && win->len == p->pack_size
678 && !p->do_not_close)
679 close_pack_fd(p);
680 pack_mmap_calls++;
681 pack_open_windows++;
682 if (pack_mapped > peak_pack_mapped)
683 peak_pack_mapped = pack_mapped;
684 if (pack_open_windows > peak_pack_open_windows)
685 peak_pack_open_windows = pack_open_windows;
686 win->next = p->windows;
687 p->windows = win;
688 }
689 }
690 if (win != *w_cursor) {
691 win->last_used = pack_used_ctr++;
692 win->inuse_cnt++;
693 *w_cursor = win;
694 }
695 offset -= win->offset;
696 if (left)
697 *left = win->len - xsize_t(offset);
698 return win->base + offset;
699 }
700
701 void unuse_pack(struct pack_window **w_cursor)
702 {
703 struct pack_window *w = *w_cursor;
704 if (w) {
705 w->inuse_cnt--;
706 *w_cursor = NULL;
707 }
708 }
709
710 static void try_to_free_pack_memory(size_t size)
711 {
712 release_pack_memory(size);
713 }
714
715 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
716 {
717 static int have_set_try_to_free_routine;
718 struct stat st;
719 size_t alloc;
720 struct packed_git *p;
721
722 if (!have_set_try_to_free_routine) {
723 have_set_try_to_free_routine = 1;
724 set_try_to_free_routine(try_to_free_pack_memory);
725 }
726
727 /*
728 * Make sure a corresponding .pack file exists and that
729 * the index looks sane.
730 */
731 if (!strip_suffix_mem(path, &path_len, ".idx"))
732 return NULL;
733
734 /*
735 * ".promisor" is long enough to hold any suffix we're adding (and
736 * the use xsnprintf double-checks that)
737 */
738 alloc = st_add3(path_len, strlen(".promisor"), 1);
739 p = alloc_packed_git(alloc);
740 memcpy(p->pack_name, path, path_len);
741
742 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
743 if (!access(p->pack_name, F_OK))
744 p->pack_keep = 1;
745
746 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
747 if (!access(p->pack_name, F_OK))
748 p->pack_promisor = 1;
749
750 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
751 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
752 free(p);
753 return NULL;
754 }
755
756 /* ok, it looks sane as far as we can check without
757 * actually mapping the pack file.
758 */
759 p->pack_size = st.st_size;
760 p->pack_local = local;
761 p->mtime = st.st_mtime;
762 if (path_len < the_hash_algo->hexsz ||
763 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
764 hashclr(p->hash);
765 return p;
766 }
767
768 void install_packed_git(struct repository *r, struct packed_git *pack)
769 {
770 if (pack->pack_fd != -1)
771 pack_open_fds++;
772
773 pack->next = r->objects->packed_git;
774 r->objects->packed_git = pack;
775 }
776
777 void (*report_garbage)(unsigned seen_bits, const char *path);
778
779 static void report_helper(const struct string_list *list,
780 int seen_bits, int first, int last)
781 {
782 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
783 return;
784
785 for (; first < last; first++)
786 report_garbage(seen_bits, list->items[first].string);
787 }
788
789 static void report_pack_garbage(struct string_list *list)
790 {
791 int i, baselen = -1, first = 0, seen_bits = 0;
792
793 if (!report_garbage)
794 return;
795
796 string_list_sort(list);
797
798 for (i = 0; i < list->nr; i++) {
799 const char *path = list->items[i].string;
800 if (baselen != -1 &&
801 strncmp(path, list->items[first].string, baselen)) {
802 report_helper(list, seen_bits, first, i);
803 baselen = -1;
804 seen_bits = 0;
805 }
806 if (baselen == -1) {
807 const char *dot = strrchr(path, '.');
808 if (!dot) {
809 report_garbage(PACKDIR_FILE_GARBAGE, path);
810 continue;
811 }
812 baselen = dot - path + 1;
813 first = i;
814 }
815 if (!strcmp(path + baselen, "pack"))
816 seen_bits |= 1;
817 else if (!strcmp(path + baselen, "idx"))
818 seen_bits |= 2;
819 }
820 report_helper(list, seen_bits, first, list->nr);
821 }
822
823 void for_each_file_in_pack_dir(const char *objdir,
824 each_file_in_pack_dir_fn fn,
825 void *data)
826 {
827 struct strbuf path = STRBUF_INIT;
828 size_t dirnamelen;
829 DIR *dir;
830 struct dirent *de;
831
832 strbuf_addstr(&path, objdir);
833 strbuf_addstr(&path, "/pack");
834 dir = opendir(path.buf);
835 if (!dir) {
836 if (errno != ENOENT)
837 error_errno("unable to open object pack directory: %s",
838 path.buf);
839 strbuf_release(&path);
840 return;
841 }
842 strbuf_addch(&path, '/');
843 dirnamelen = path.len;
844 while ((de = readdir(dir)) != NULL) {
845 if (is_dot_or_dotdot(de->d_name))
846 continue;
847
848 strbuf_setlen(&path, dirnamelen);
849 strbuf_addstr(&path, de->d_name);
850
851 fn(path.buf, path.len, de->d_name, data);
852 }
853
854 closedir(dir);
855 strbuf_release(&path);
856 }
857
858 struct prepare_pack_data {
859 struct repository *r;
860 struct string_list *garbage;
861 int local;
862 struct multi_pack_index *m;
863 };
864
865 static void prepare_pack(const char *full_name, size_t full_name_len,
866 const char *file_name, void *_data)
867 {
868 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
869 struct packed_git *p;
870 size_t base_len = full_name_len;
871
872 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
873 !(data->m && midx_contains_pack(data->m, file_name))) {
874 /* Don't reopen a pack we already have. */
875 for (p = data->r->objects->packed_git; p; p = p->next) {
876 size_t len;
877 if (strip_suffix(p->pack_name, ".pack", &len) &&
878 len == base_len &&
879 !memcmp(p->pack_name, full_name, len))
880 break;
881 }
882
883 if (!p) {
884 p = add_packed_git(full_name, full_name_len, data->local);
885 if (p)
886 install_packed_git(data->r, p);
887 }
888 }
889
890 if (!report_garbage)
891 return;
892
893 if (!strcmp(file_name, "multi-pack-index"))
894 return;
895 if (ends_with(file_name, ".idx") ||
896 ends_with(file_name, ".pack") ||
897 ends_with(file_name, ".bitmap") ||
898 ends_with(file_name, ".keep") ||
899 ends_with(file_name, ".promisor"))
900 string_list_append(data->garbage, full_name);
901 else
902 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
903 }
904
905 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
906 {
907 struct prepare_pack_data data;
908 struct string_list garbage = STRING_LIST_INIT_DUP;
909
910 data.m = r->objects->multi_pack_index;
911
912 /* look for the multi-pack-index for this object directory */
913 while (data.m && strcmp(data.m->object_dir, objdir))
914 data.m = data.m->next;
915
916 data.r = r;
917 data.garbage = &garbage;
918 data.local = local;
919
920 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
921
922 report_pack_garbage(data.garbage);
923 string_list_clear(data.garbage, 0);
924 }
925
926 static void prepare_packed_git(struct repository *r);
927 /*
928 * Give a fast, rough count of the number of objects in the repository. This
929 * ignores loose objects completely. If you have a lot of them, then either
930 * you should repack because your performance will be awful, or they are
931 * all unreachable objects about to be pruned, in which case they're not really
932 * interesting as a measure of repo size in the first place.
933 */
934 unsigned long repo_approximate_object_count(struct repository *r)
935 {
936 if (!r->objects->approximate_object_count_valid) {
937 unsigned long count;
938 struct multi_pack_index *m;
939 struct packed_git *p;
940
941 prepare_packed_git(r);
942 count = 0;
943 for (m = get_multi_pack_index(r); m; m = m->next)
944 count += m->num_objects;
945 for (p = r->objects->packed_git; p; p = p->next) {
946 if (open_pack_index(p))
947 continue;
948 count += p->num_objects;
949 }
950 r->objects->approximate_object_count = count;
951 }
952 return r->objects->approximate_object_count;
953 }
954
955 static void *get_next_packed_git(const void *p)
956 {
957 return ((const struct packed_git *)p)->next;
958 }
959
960 static void set_next_packed_git(void *p, void *next)
961 {
962 ((struct packed_git *)p)->next = next;
963 }
964
965 static int sort_pack(const void *a_, const void *b_)
966 {
967 const struct packed_git *a = a_;
968 const struct packed_git *b = b_;
969 int st;
970
971 /*
972 * Local packs tend to contain objects specific to our
973 * variant of the project than remote ones. In addition,
974 * remote ones could be on a network mounted filesystem.
975 * Favor local ones for these reasons.
976 */
977 st = a->pack_local - b->pack_local;
978 if (st)
979 return -st;
980
981 /*
982 * Younger packs tend to contain more recent objects,
983 * and more recent objects tend to get accessed more
984 * often.
985 */
986 if (a->mtime < b->mtime)
987 return 1;
988 else if (a->mtime == b->mtime)
989 return 0;
990 return -1;
991 }
992
993 static void rearrange_packed_git(struct repository *r)
994 {
995 r->objects->packed_git = llist_mergesort(
996 r->objects->packed_git, get_next_packed_git,
997 set_next_packed_git, sort_pack);
998 }
999
1000 static void prepare_packed_git_mru(struct repository *r)
1001 {
1002 struct packed_git *p;
1003
1004 INIT_LIST_HEAD(&r->objects->packed_git_mru);
1005
1006 for (p = r->objects->packed_git; p; p = p->next)
1007 list_add_tail(&p->mru, &r->objects->packed_git_mru);
1008 }
1009
1010 static void prepare_packed_git(struct repository *r)
1011 {
1012 struct object_directory *odb;
1013
1014 if (r->objects->packed_git_initialized)
1015 return;
1016
1017 prepare_alt_odb(r);
1018 for (odb = r->objects->odb; odb; odb = odb->next) {
1019 int local = (odb == r->objects->odb);
1020 prepare_multi_pack_index_one(r, odb->path, local);
1021 prepare_packed_git_one(r, odb->path, local);
1022 }
1023 rearrange_packed_git(r);
1024
1025 prepare_packed_git_mru(r);
1026 r->objects->packed_git_initialized = 1;
1027 }
1028
1029 void reprepare_packed_git(struct repository *r)
1030 {
1031 struct object_directory *odb;
1032
1033 for (odb = r->objects->odb; odb; odb = odb->next)
1034 odb_clear_loose_cache(odb);
1035
1036 r->objects->approximate_object_count_valid = 0;
1037 r->objects->packed_git_initialized = 0;
1038 prepare_packed_git(r);
1039 }
1040
1041 struct packed_git *get_packed_git(struct repository *r)
1042 {
1043 prepare_packed_git(r);
1044 return r->objects->packed_git;
1045 }
1046
1047 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1048 {
1049 prepare_packed_git(r);
1050 return r->objects->multi_pack_index;
1051 }
1052
1053 struct packed_git *get_all_packs(struct repository *r)
1054 {
1055 struct multi_pack_index *m;
1056
1057 prepare_packed_git(r);
1058 for (m = r->objects->multi_pack_index; m; m = m->next) {
1059 uint32_t i;
1060 for (i = 0; i < m->num_packs; i++)
1061 prepare_midx_pack(r, m, i);
1062 }
1063
1064 return r->objects->packed_git;
1065 }
1066
1067 struct list_head *get_packed_git_mru(struct repository *r)
1068 {
1069 prepare_packed_git(r);
1070 return &r->objects->packed_git_mru;
1071 }
1072
1073 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1074 unsigned long len, enum object_type *type, unsigned long *sizep)
1075 {
1076 unsigned shift;
1077 unsigned long size, c;
1078 unsigned long used = 0;
1079
1080 c = buf[used++];
1081 *type = (c >> 4) & 7;
1082 size = c & 15;
1083 shift = 4;
1084 while (c & 0x80) {
1085 if (len <= used || bitsizeof(long) <= shift) {
1086 error("bad object header");
1087 size = used = 0;
1088 break;
1089 }
1090 c = buf[used++];
1091 size += (c & 0x7f) << shift;
1092 shift += 7;
1093 }
1094 *sizep = size;
1095 return used;
1096 }
1097
1098 unsigned long get_size_from_delta(struct packed_git *p,
1099 struct pack_window **w_curs,
1100 off_t curpos)
1101 {
1102 const unsigned char *data;
1103 unsigned char delta_head[20], *in;
1104 git_zstream stream;
1105 int st;
1106
1107 memset(&stream, 0, sizeof(stream));
1108 stream.next_out = delta_head;
1109 stream.avail_out = sizeof(delta_head);
1110
1111 git_inflate_init(&stream);
1112 do {
1113 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1114 stream.next_in = in;
1115 st = git_inflate(&stream, Z_FINISH);
1116 curpos += stream.next_in - in;
1117 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1118 stream.total_out < sizeof(delta_head));
1119 git_inflate_end(&stream);
1120 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1121 error("delta data unpack-initial failed");
1122 return 0;
1123 }
1124
1125 /* Examine the initial part of the delta to figure out
1126 * the result size.
1127 */
1128 data = delta_head;
1129
1130 /* ignore base size */
1131 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1132
1133 /* Read the result size */
1134 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1135 }
1136
1137 int unpack_object_header(struct packed_git *p,
1138 struct pack_window **w_curs,
1139 off_t *curpos,
1140 unsigned long *sizep)
1141 {
1142 unsigned char *base;
1143 unsigned long left;
1144 unsigned long used;
1145 enum object_type type;
1146
1147 /* use_pack() assures us we have [base, base + 20) available
1148 * as a range that we can look at. (Its actually the hash
1149 * size that is assured.) With our object header encoding
1150 * the maximum deflated object size is 2^137, which is just
1151 * insane, so we know won't exceed what we have been given.
1152 */
1153 base = use_pack(p, w_curs, *curpos, &left);
1154 used = unpack_object_header_buffer(base, left, &type, sizep);
1155 if (!used) {
1156 type = OBJ_BAD;
1157 } else
1158 *curpos += used;
1159
1160 return type;
1161 }
1162
1163 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1164 {
1165 unsigned i;
1166 const unsigned hashsz = the_hash_algo->rawsz;
1167 for (i = 0; i < p->num_bad_objects; i++)
1168 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1169 return;
1170 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1171 st_mult(GIT_MAX_RAWSZ,
1172 st_add(p->num_bad_objects, 1)));
1173 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1174 p->num_bad_objects++;
1175 }
1176
1177 const struct packed_git *has_packed_and_bad(struct repository *r,
1178 const unsigned char *sha1)
1179 {
1180 struct packed_git *p;
1181 unsigned i;
1182
1183 for (p = r->objects->packed_git; p; p = p->next)
1184 for (i = 0; i < p->num_bad_objects; i++)
1185 if (hasheq(sha1,
1186 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1187 return p;
1188 return NULL;
1189 }
1190
1191 static off_t get_delta_base(struct packed_git *p,
1192 struct pack_window **w_curs,
1193 off_t *curpos,
1194 enum object_type type,
1195 off_t delta_obj_offset)
1196 {
1197 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1198 off_t base_offset;
1199
1200 /* use_pack() assured us we have [base_info, base_info + 20)
1201 * as a range that we can look at without walking off the
1202 * end of the mapped window. Its actually the hash size
1203 * that is assured. An OFS_DELTA longer than the hash size
1204 * is stupid, as then a REF_DELTA would be smaller to store.
1205 */
1206 if (type == OBJ_OFS_DELTA) {
1207 unsigned used = 0;
1208 unsigned char c = base_info[used++];
1209 base_offset = c & 127;
1210 while (c & 128) {
1211 base_offset += 1;
1212 if (!base_offset || MSB(base_offset, 7))
1213 return 0; /* overflow */
1214 c = base_info[used++];
1215 base_offset = (base_offset << 7) + (c & 127);
1216 }
1217 base_offset = delta_obj_offset - base_offset;
1218 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1219 return 0; /* out of bound */
1220 *curpos += used;
1221 } else if (type == OBJ_REF_DELTA) {
1222 /* The base entry _must_ be in the same pack */
1223 base_offset = find_pack_entry_one(base_info, p);
1224 *curpos += the_hash_algo->rawsz;
1225 } else
1226 die("I am totally screwed");
1227 return base_offset;
1228 }
1229
1230 /*
1231 * Like get_delta_base above, but we return the sha1 instead of the pack
1232 * offset. This means it is cheaper for REF deltas (we do not have to do
1233 * the final object lookup), but more expensive for OFS deltas (we
1234 * have to load the revidx to convert the offset back into a sha1).
1235 */
1236 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1237 struct pack_window **w_curs,
1238 off_t curpos,
1239 enum object_type type,
1240 off_t delta_obj_offset)
1241 {
1242 if (type == OBJ_REF_DELTA) {
1243 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1244 return base;
1245 } else if (type == OBJ_OFS_DELTA) {
1246 struct revindex_entry *revidx;
1247 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1248 type, delta_obj_offset);
1249
1250 if (!base_offset)
1251 return NULL;
1252
1253 revidx = find_pack_revindex(p, base_offset);
1254 if (!revidx)
1255 return NULL;
1256
1257 return nth_packed_object_sha1(p, revidx->nr);
1258 } else
1259 return NULL;
1260 }
1261
1262 static int retry_bad_packed_offset(struct repository *r,
1263 struct packed_git *p,
1264 off_t obj_offset)
1265 {
1266 int type;
1267 struct revindex_entry *revidx;
1268 struct object_id oid;
1269 revidx = find_pack_revindex(p, obj_offset);
1270 if (!revidx)
1271 return OBJ_BAD;
1272 nth_packed_object_oid(&oid, p, revidx->nr);
1273 mark_bad_packed_object(p, oid.hash);
1274 type = oid_object_info(r, &oid, NULL);
1275 if (type <= OBJ_NONE)
1276 return OBJ_BAD;
1277 return type;
1278 }
1279
1280 #define POI_STACK_PREALLOC 64
1281
1282 static enum object_type packed_to_object_type(struct repository *r,
1283 struct packed_git *p,
1284 off_t obj_offset,
1285 enum object_type type,
1286 struct pack_window **w_curs,
1287 off_t curpos)
1288 {
1289 off_t small_poi_stack[POI_STACK_PREALLOC];
1290 off_t *poi_stack = small_poi_stack;
1291 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1292
1293 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1294 off_t base_offset;
1295 unsigned long size;
1296 /* Push the object we're going to leave behind */
1297 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1298 poi_stack_alloc = alloc_nr(poi_stack_nr);
1299 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1300 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1301 } else {
1302 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1303 }
1304 poi_stack[poi_stack_nr++] = obj_offset;
1305 /* If parsing the base offset fails, just unwind */
1306 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1307 if (!base_offset)
1308 goto unwind;
1309 curpos = obj_offset = base_offset;
1310 type = unpack_object_header(p, w_curs, &curpos, &size);
1311 if (type <= OBJ_NONE) {
1312 /* If getting the base itself fails, we first
1313 * retry the base, otherwise unwind */
1314 type = retry_bad_packed_offset(r, p, base_offset);
1315 if (type > OBJ_NONE)
1316 goto out;
1317 goto unwind;
1318 }
1319 }
1320
1321 switch (type) {
1322 case OBJ_BAD:
1323 case OBJ_COMMIT:
1324 case OBJ_TREE:
1325 case OBJ_BLOB:
1326 case OBJ_TAG:
1327 break;
1328 default:
1329 error("unknown object type %i at offset %"PRIuMAX" in %s",
1330 type, (uintmax_t)obj_offset, p->pack_name);
1331 type = OBJ_BAD;
1332 }
1333
1334 out:
1335 if (poi_stack != small_poi_stack)
1336 free(poi_stack);
1337 return type;
1338
1339 unwind:
1340 while (poi_stack_nr) {
1341 obj_offset = poi_stack[--poi_stack_nr];
1342 type = retry_bad_packed_offset(r, p, obj_offset);
1343 if (type > OBJ_NONE)
1344 goto out;
1345 }
1346 type = OBJ_BAD;
1347 goto out;
1348 }
1349
1350 static struct hashmap delta_base_cache;
1351 static size_t delta_base_cached;
1352
1353 static LIST_HEAD(delta_base_cache_lru);
1354
1355 struct delta_base_cache_key {
1356 struct packed_git *p;
1357 off_t base_offset;
1358 };
1359
1360 struct delta_base_cache_entry {
1361 struct hashmap hash;
1362 struct delta_base_cache_key key;
1363 struct list_head lru;
1364 void *data;
1365 unsigned long size;
1366 enum object_type type;
1367 };
1368
1369 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1370 {
1371 unsigned int hash;
1372
1373 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1374 hash += (hash >> 8) + (hash >> 16);
1375 return hash;
1376 }
1377
1378 static struct delta_base_cache_entry *
1379 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1380 {
1381 struct hashmap_entry entry;
1382 struct delta_base_cache_key key;
1383
1384 if (!delta_base_cache.cmpfn)
1385 return NULL;
1386
1387 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1388 key.p = p;
1389 key.base_offset = base_offset;
1390 return hashmap_get(&delta_base_cache, &entry, &key);
1391 }
1392
1393 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1394 const struct delta_base_cache_key *b)
1395 {
1396 return a->p == b->p && a->base_offset == b->base_offset;
1397 }
1398
1399 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1400 const void *va, const void *vb,
1401 const void *vkey)
1402 {
1403 const struct delta_base_cache_entry *a = va, *b = vb;
1404 const struct delta_base_cache_key *key = vkey;
1405 if (key)
1406 return !delta_base_cache_key_eq(&a->key, key);
1407 else
1408 return !delta_base_cache_key_eq(&a->key, &b->key);
1409 }
1410
1411 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1412 {
1413 return !!get_delta_base_cache_entry(p, base_offset);
1414 }
1415
1416 /*
1417 * Remove the entry from the cache, but do _not_ free the associated
1418 * entry data. The caller takes ownership of the "data" buffer, and
1419 * should copy out any fields it wants before detaching.
1420 */
1421 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1422 {
1423 hashmap_remove(&delta_base_cache, ent, &ent->key);
1424 list_del(&ent->lru);
1425 delta_base_cached -= ent->size;
1426 free(ent);
1427 }
1428
1429 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1430 off_t base_offset, unsigned long *base_size,
1431 enum object_type *type)
1432 {
1433 struct delta_base_cache_entry *ent;
1434
1435 ent = get_delta_base_cache_entry(p, base_offset);
1436 if (!ent)
1437 return unpack_entry(r, p, base_offset, type, base_size);
1438
1439 if (type)
1440 *type = ent->type;
1441 if (base_size)
1442 *base_size = ent->size;
1443 return xmemdupz(ent->data, ent->size);
1444 }
1445
1446 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1447 {
1448 free(ent->data);
1449 detach_delta_base_cache_entry(ent);
1450 }
1451
1452 void clear_delta_base_cache(void)
1453 {
1454 struct list_head *lru, *tmp;
1455 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1456 struct delta_base_cache_entry *entry =
1457 list_entry(lru, struct delta_base_cache_entry, lru);
1458 release_delta_base_cache(entry);
1459 }
1460 }
1461
1462 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1463 void *base, unsigned long base_size, enum object_type type)
1464 {
1465 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1466 struct list_head *lru, *tmp;
1467
1468 delta_base_cached += base_size;
1469
1470 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1471 struct delta_base_cache_entry *f =
1472 list_entry(lru, struct delta_base_cache_entry, lru);
1473 if (delta_base_cached <= delta_base_cache_limit)
1474 break;
1475 release_delta_base_cache(f);
1476 }
1477
1478 ent->key.p = p;
1479 ent->key.base_offset = base_offset;
1480 ent->type = type;
1481 ent->data = base;
1482 ent->size = base_size;
1483 list_add_tail(&ent->lru, &delta_base_cache_lru);
1484
1485 if (!delta_base_cache.cmpfn)
1486 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1487 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1488 hashmap_add(&delta_base_cache, ent);
1489 }
1490
1491 int packed_object_info(struct repository *r, struct packed_git *p,
1492 off_t obj_offset, struct object_info *oi)
1493 {
1494 struct pack_window *w_curs = NULL;
1495 unsigned long size;
1496 off_t curpos = obj_offset;
1497 enum object_type type;
1498
1499 /*
1500 * We always get the representation type, but only convert it to
1501 * a "real" type later if the caller is interested.
1502 */
1503 if (oi->contentp) {
1504 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1505 &type);
1506 if (!*oi->contentp)
1507 type = OBJ_BAD;
1508 } else {
1509 type = unpack_object_header(p, &w_curs, &curpos, &size);
1510 }
1511
1512 if (!oi->contentp && oi->sizep) {
1513 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1514 off_t tmp_pos = curpos;
1515 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1516 type, obj_offset);
1517 if (!base_offset) {
1518 type = OBJ_BAD;
1519 goto out;
1520 }
1521 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1522 if (*oi->sizep == 0) {
1523 type = OBJ_BAD;
1524 goto out;
1525 }
1526 } else {
1527 *oi->sizep = size;
1528 }
1529 }
1530
1531 if (oi->disk_sizep) {
1532 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1533 *oi->disk_sizep = revidx[1].offset - obj_offset;
1534 }
1535
1536 if (oi->typep || oi->type_name) {
1537 enum object_type ptot;
1538 ptot = packed_to_object_type(r, p, obj_offset,
1539 type, &w_curs, curpos);
1540 if (oi->typep)
1541 *oi->typep = ptot;
1542 if (oi->type_name) {
1543 const char *tn = type_name(ptot);
1544 if (tn)
1545 strbuf_addstr(oi->type_name, tn);
1546 }
1547 if (ptot < 0) {
1548 type = OBJ_BAD;
1549 goto out;
1550 }
1551 }
1552
1553 if (oi->delta_base_sha1) {
1554 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1555 const unsigned char *base;
1556
1557 base = get_delta_base_sha1(p, &w_curs, curpos,
1558 type, obj_offset);
1559 if (!base) {
1560 type = OBJ_BAD;
1561 goto out;
1562 }
1563
1564 hashcpy(oi->delta_base_sha1, base);
1565 } else
1566 hashclr(oi->delta_base_sha1);
1567 }
1568
1569 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1570 OI_PACKED;
1571
1572 out:
1573 unuse_pack(&w_curs);
1574 return type;
1575 }
1576
1577 static void *unpack_compressed_entry(struct packed_git *p,
1578 struct pack_window **w_curs,
1579 off_t curpos,
1580 unsigned long size)
1581 {
1582 int st;
1583 git_zstream stream;
1584 unsigned char *buffer, *in;
1585
1586 buffer = xmallocz_gently(size);
1587 if (!buffer)
1588 return NULL;
1589 memset(&stream, 0, sizeof(stream));
1590 stream.next_out = buffer;
1591 stream.avail_out = size + 1;
1592
1593 git_inflate_init(&stream);
1594 do {
1595 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1596 stream.next_in = in;
1597 st = git_inflate(&stream, Z_FINISH);
1598 if (!stream.avail_out)
1599 break; /* the payload is larger than it should be */
1600 curpos += stream.next_in - in;
1601 } while (st == Z_OK || st == Z_BUF_ERROR);
1602 git_inflate_end(&stream);
1603 if ((st != Z_STREAM_END) || stream.total_out != size) {
1604 free(buffer);
1605 return NULL;
1606 }
1607
1608 /* versions of zlib can clobber unconsumed portion of outbuf */
1609 buffer[size] = '\0';
1610
1611 return buffer;
1612 }
1613
1614 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1615 {
1616 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1617 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1618 p->pack_name, (uintmax_t)obj_offset);
1619 }
1620
1621 int do_check_packed_object_crc;
1622
1623 #define UNPACK_ENTRY_STACK_PREALLOC 64
1624 struct unpack_entry_stack_ent {
1625 off_t obj_offset;
1626 off_t curpos;
1627 unsigned long size;
1628 };
1629
1630 static void *read_object(struct repository *r,
1631 const struct object_id *oid,
1632 enum object_type *type,
1633 unsigned long *size)
1634 {
1635 struct object_info oi = OBJECT_INFO_INIT;
1636 void *content;
1637 oi.typep = type;
1638 oi.sizep = size;
1639 oi.contentp = &content;
1640
1641 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1642 return NULL;
1643 return content;
1644 }
1645
1646 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1647 enum object_type *final_type, unsigned long *final_size)
1648 {
1649 struct pack_window *w_curs = NULL;
1650 off_t curpos = obj_offset;
1651 void *data = NULL;
1652 unsigned long size;
1653 enum object_type type;
1654 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1655 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1656 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1657 int base_from_cache = 0;
1658
1659 write_pack_access_log(p, obj_offset);
1660
1661 /* PHASE 1: drill down to the innermost base object */
1662 for (;;) {
1663 off_t base_offset;
1664 int i;
1665 struct delta_base_cache_entry *ent;
1666
1667 ent = get_delta_base_cache_entry(p, curpos);
1668 if (ent) {
1669 type = ent->type;
1670 data = ent->data;
1671 size = ent->size;
1672 detach_delta_base_cache_entry(ent);
1673 base_from_cache = 1;
1674 break;
1675 }
1676
1677 if (do_check_packed_object_crc && p->index_version > 1) {
1678 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1679 off_t len = revidx[1].offset - obj_offset;
1680 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1681 struct object_id oid;
1682 nth_packed_object_oid(&oid, p, revidx->nr);
1683 error("bad packed object CRC for %s",
1684 oid_to_hex(&oid));
1685 mark_bad_packed_object(p, oid.hash);
1686 data = NULL;
1687 goto out;
1688 }
1689 }
1690
1691 type = unpack_object_header(p, &w_curs, &curpos, &size);
1692 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1693 break;
1694
1695 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1696 if (!base_offset) {
1697 error("failed to validate delta base reference "
1698 "at offset %"PRIuMAX" from %s",
1699 (uintmax_t)curpos, p->pack_name);
1700 /* bail to phase 2, in hopes of recovery */
1701 data = NULL;
1702 break;
1703 }
1704
1705 /* push object, proceed to base */
1706 if (delta_stack_nr >= delta_stack_alloc
1707 && delta_stack == small_delta_stack) {
1708 delta_stack_alloc = alloc_nr(delta_stack_nr);
1709 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1710 memcpy(delta_stack, small_delta_stack,
1711 sizeof(*delta_stack)*delta_stack_nr);
1712 } else {
1713 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1714 }
1715 i = delta_stack_nr++;
1716 delta_stack[i].obj_offset = obj_offset;
1717 delta_stack[i].curpos = curpos;
1718 delta_stack[i].size = size;
1719
1720 curpos = obj_offset = base_offset;
1721 }
1722
1723 /* PHASE 2: handle the base */
1724 switch (type) {
1725 case OBJ_OFS_DELTA:
1726 case OBJ_REF_DELTA:
1727 if (data)
1728 BUG("unpack_entry: left loop at a valid delta");
1729 break;
1730 case OBJ_COMMIT:
1731 case OBJ_TREE:
1732 case OBJ_BLOB:
1733 case OBJ_TAG:
1734 if (!base_from_cache)
1735 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1736 break;
1737 default:
1738 data = NULL;
1739 error("unknown object type %i at offset %"PRIuMAX" in %s",
1740 type, (uintmax_t)obj_offset, p->pack_name);
1741 }
1742
1743 /* PHASE 3: apply deltas in order */
1744
1745 /* invariants:
1746 * 'data' holds the base data, or NULL if there was corruption
1747 */
1748 while (delta_stack_nr) {
1749 void *delta_data;
1750 void *base = data;
1751 void *external_base = NULL;
1752 unsigned long delta_size, base_size = size;
1753 int i;
1754
1755 data = NULL;
1756
1757 if (base)
1758 add_delta_base_cache(p, obj_offset, base, base_size, type);
1759
1760 if (!base) {
1761 /*
1762 * We're probably in deep shit, but let's try to fetch
1763 * the required base anyway from another pack or loose.
1764 * This is costly but should happen only in the presence
1765 * of a corrupted pack, and is better than failing outright.
1766 */
1767 struct revindex_entry *revidx;
1768 struct object_id base_oid;
1769 revidx = find_pack_revindex(p, obj_offset);
1770 if (revidx) {
1771 nth_packed_object_oid(&base_oid, p, revidx->nr);
1772 error("failed to read delta base object %s"
1773 " at offset %"PRIuMAX" from %s",
1774 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1775 p->pack_name);
1776 mark_bad_packed_object(p, base_oid.hash);
1777 base = read_object(r, &base_oid, &type, &base_size);
1778 external_base = base;
1779 }
1780 }
1781
1782 i = --delta_stack_nr;
1783 obj_offset = delta_stack[i].obj_offset;
1784 curpos = delta_stack[i].curpos;
1785 delta_size = delta_stack[i].size;
1786
1787 if (!base)
1788 continue;
1789
1790 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1791
1792 if (!delta_data) {
1793 error("failed to unpack compressed delta "
1794 "at offset %"PRIuMAX" from %s",
1795 (uintmax_t)curpos, p->pack_name);
1796 data = NULL;
1797 free(external_base);
1798 continue;
1799 }
1800
1801 data = patch_delta(base, base_size,
1802 delta_data, delta_size,
1803 &size);
1804
1805 /*
1806 * We could not apply the delta; warn the user, but keep going.
1807 * Our failure will be noticed either in the next iteration of
1808 * the loop, or if this is the final delta, in the caller when
1809 * we return NULL. Those code paths will take care of making
1810 * a more explicit warning and retrying with another copy of
1811 * the object.
1812 */
1813 if (!data)
1814 error("failed to apply delta");
1815
1816 free(delta_data);
1817 free(external_base);
1818 }
1819
1820 if (final_type)
1821 *final_type = type;
1822 if (final_size)
1823 *final_size = size;
1824
1825 out:
1826 unuse_pack(&w_curs);
1827
1828 if (delta_stack != small_delta_stack)
1829 free(delta_stack);
1830
1831 return data;
1832 }
1833
1834 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1835 {
1836 const unsigned char *index_fanout = p->index_data;
1837 const unsigned char *index_lookup;
1838 const unsigned int hashsz = the_hash_algo->rawsz;
1839 int index_lookup_width;
1840
1841 if (!index_fanout)
1842 BUG("bsearch_pack called without a valid pack-index");
1843
1844 index_lookup = index_fanout + 4 * 256;
1845 if (p->index_version == 1) {
1846 index_lookup_width = hashsz + 4;
1847 index_lookup += 4;
1848 } else {
1849 index_lookup_width = hashsz;
1850 index_fanout += 8;
1851 index_lookup += 8;
1852 }
1853
1854 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1855 index_lookup, index_lookup_width, result);
1856 }
1857
1858 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1859 uint32_t n)
1860 {
1861 const unsigned char *index = p->index_data;
1862 const unsigned int hashsz = the_hash_algo->rawsz;
1863 if (!index) {
1864 if (open_pack_index(p))
1865 return NULL;
1866 index = p->index_data;
1867 }
1868 if (n >= p->num_objects)
1869 return NULL;
1870 index += 4 * 256;
1871 if (p->index_version == 1) {
1872 return index + (hashsz + 4) * n + 4;
1873 } else {
1874 index += 8;
1875 return index + hashsz * n;
1876 }
1877 }
1878
1879 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1880 struct packed_git *p,
1881 uint32_t n)
1882 {
1883 const unsigned char *hash = nth_packed_object_sha1(p, n);
1884 if (!hash)
1885 return NULL;
1886 hashcpy(oid->hash, hash);
1887 return oid;
1888 }
1889
1890 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1891 {
1892 const unsigned char *ptr = vptr;
1893 const unsigned char *start = p->index_data;
1894 const unsigned char *end = start + p->index_size;
1895 if (ptr < start)
1896 die(_("offset before start of pack index for %s (corrupt index?)"),
1897 p->pack_name);
1898 /* No need to check for underflow; .idx files must be at least 8 bytes */
1899 if (ptr >= end - 8)
1900 die(_("offset beyond end of pack index for %s (truncated index?)"),
1901 p->pack_name);
1902 }
1903
1904 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1905 {
1906 const unsigned char *index = p->index_data;
1907 const unsigned int hashsz = the_hash_algo->rawsz;
1908 index += 4 * 256;
1909 if (p->index_version == 1) {
1910 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1911 } else {
1912 uint32_t off;
1913 index += 8 + p->num_objects * (hashsz + 4);
1914 off = ntohl(*((uint32_t *)(index + 4 * n)));
1915 if (!(off & 0x80000000))
1916 return off;
1917 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1918 check_pack_index_ptr(p, index);
1919 return get_be64(index);
1920 }
1921 }
1922
1923 off_t find_pack_entry_one(const unsigned char *sha1,
1924 struct packed_git *p)
1925 {
1926 const unsigned char *index = p->index_data;
1927 struct object_id oid;
1928 uint32_t result;
1929
1930 if (!index) {
1931 if (open_pack_index(p))
1932 return 0;
1933 }
1934
1935 hashcpy(oid.hash, sha1);
1936 if (bsearch_pack(&oid, p, &result))
1937 return nth_packed_object_offset(p, result);
1938 return 0;
1939 }
1940
1941 int is_pack_valid(struct packed_git *p)
1942 {
1943 /* An already open pack is known to be valid. */
1944 if (p->pack_fd != -1)
1945 return 1;
1946
1947 /* If the pack has one window completely covering the
1948 * file size, the pack is known to be valid even if
1949 * the descriptor is not currently open.
1950 */
1951 if (p->windows) {
1952 struct pack_window *w = p->windows;
1953
1954 if (!w->offset && w->len == p->pack_size)
1955 return 1;
1956 }
1957
1958 /* Force the pack to open to prove its valid. */
1959 return !open_packed_git(p);
1960 }
1961
1962 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1963 struct packed_git *packs)
1964 {
1965 struct packed_git *p;
1966
1967 for (p = packs; p; p = p->next) {
1968 if (find_pack_entry_one(sha1, p))
1969 return p;
1970 }
1971 return NULL;
1972
1973 }
1974
1975 static int fill_pack_entry(const struct object_id *oid,
1976 struct pack_entry *e,
1977 struct packed_git *p)
1978 {
1979 off_t offset;
1980
1981 if (p->num_bad_objects) {
1982 unsigned i;
1983 for (i = 0; i < p->num_bad_objects; i++)
1984 if (hasheq(oid->hash,
1985 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1986 return 0;
1987 }
1988
1989 offset = find_pack_entry_one(oid->hash, p);
1990 if (!offset)
1991 return 0;
1992
1993 /*
1994 * We are about to tell the caller where they can locate the
1995 * requested object. We better make sure the packfile is
1996 * still here and can be accessed before supplying that
1997 * answer, as it may have been deleted since the index was
1998 * loaded!
1999 */
2000 if (!is_pack_valid(p))
2001 return 0;
2002 e->offset = offset;
2003 e->p = p;
2004 return 1;
2005 }
2006
2007 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2008 {
2009 struct list_head *pos;
2010 struct multi_pack_index *m;
2011
2012 prepare_packed_git(r);
2013 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2014 return 0;
2015
2016 for (m = r->objects->multi_pack_index; m; m = m->next) {
2017 if (fill_midx_entry(r, oid, e, m))
2018 return 1;
2019 }
2020
2021 list_for_each(pos, &r->objects->packed_git_mru) {
2022 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2023 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2024 list_move(&p->mru, &r->objects->packed_git_mru);
2025 return 1;
2026 }
2027 }
2028 return 0;
2029 }
2030
2031 int has_object_pack(const struct object_id *oid)
2032 {
2033 struct pack_entry e;
2034 return find_pack_entry(the_repository, oid, &e);
2035 }
2036
2037 int has_pack_index(const unsigned char *sha1)
2038 {
2039 struct stat st;
2040 if (stat(sha1_pack_index_name(sha1), &st))
2041 return 0;
2042 return 1;
2043 }
2044
2045 int for_each_object_in_pack(struct packed_git *p,
2046 each_packed_object_fn cb, void *data,
2047 enum for_each_object_flags flags)
2048 {
2049 uint32_t i;
2050 int r = 0;
2051
2052 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2053 if (load_pack_revindex(p))
2054 return -1;
2055 }
2056
2057 for (i = 0; i < p->num_objects; i++) {
2058 uint32_t pos;
2059 struct object_id oid;
2060
2061 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2062 pos = p->revindex[i].nr;
2063 else
2064 pos = i;
2065
2066 if (!nth_packed_object_oid(&oid, p, pos))
2067 return error("unable to get sha1 of object %u in %s",
2068 pos, p->pack_name);
2069
2070 r = cb(&oid, p, pos, data);
2071 if (r)
2072 break;
2073 }
2074 return r;
2075 }
2076
2077 int for_each_packed_object(each_packed_object_fn cb, void *data,
2078 enum for_each_object_flags flags)
2079 {
2080 struct packed_git *p;
2081 int r = 0;
2082 int pack_errors = 0;
2083
2084 prepare_packed_git(the_repository);
2085 for (p = get_all_packs(the_repository); p; p = p->next) {
2086 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2087 continue;
2088 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2089 !p->pack_promisor)
2090 continue;
2091 if (open_pack_index(p)) {
2092 pack_errors = 1;
2093 continue;
2094 }
2095 r = for_each_object_in_pack(p, cb, data, flags);
2096 if (r)
2097 break;
2098 }
2099 return r ? r : pack_errors;
2100 }
2101
2102 static int add_promisor_object(const struct object_id *oid,
2103 struct packed_git *pack,
2104 uint32_t pos,
2105 void *set_)
2106 {
2107 struct oidset *set = set_;
2108 struct object *obj = parse_object(the_repository, oid);
2109 if (!obj)
2110 return 1;
2111
2112 oidset_insert(set, oid);
2113
2114 /*
2115 * If this is a tree, commit, or tag, the objects it refers
2116 * to are also promisor objects. (Blobs refer to no objects->)
2117 */
2118 if (obj->type == OBJ_TREE) {
2119 struct tree *tree = (struct tree *)obj;
2120 struct tree_desc desc;
2121 struct name_entry entry;
2122 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2123 /*
2124 * Error messages are given when packs are
2125 * verified, so do not print any here.
2126 */
2127 return 0;
2128 while (tree_entry_gently(&desc, &entry))
2129 oidset_insert(set, &entry.oid);
2130 } else if (obj->type == OBJ_COMMIT) {
2131 struct commit *commit = (struct commit *) obj;
2132 struct commit_list *parents = commit->parents;
2133
2134 oidset_insert(set, get_commit_tree_oid(commit));
2135 for (; parents; parents = parents->next)
2136 oidset_insert(set, &parents->item->object.oid);
2137 } else if (obj->type == OBJ_TAG) {
2138 struct tag *tag = (struct tag *) obj;
2139 oidset_insert(set, &tag->tagged->oid);
2140 }
2141 return 0;
2142 }
2143
2144 int is_promisor_object(const struct object_id *oid)
2145 {
2146 static struct oidset promisor_objects;
2147 static int promisor_objects_prepared;
2148
2149 if (!promisor_objects_prepared) {
2150 if (repository_format_partial_clone) {
2151 for_each_packed_object(add_promisor_object,
2152 &promisor_objects,
2153 FOR_EACH_OBJECT_PROMISOR_ONLY);
2154 }
2155 promisor_objects_prepared = 1;
2156 }
2157 return oidset_contains(&promisor_objects, oid);
2158 }