?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
t4202: abstract away SHA-1-specific constants
[git/debian.git] / packfile.c
blob355066de17514e1a2cf8456e6d55d871bf15f09b
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 "streaming.h"
10 #include "sha1-lookup.h"
11 #include "commit.h"
12 #include "object.h"
13 #include "tag.h"
14 #include "tree-walk.h"
15 #include "tree.h"
16 #include "object-store.h"
17 #include "midx.h"
18 #include "commit-graph.h"
19 #include "promisor-remote.h"
20
21 char *odb_pack_name(struct strbuf *buf,
22 const unsigned char *hash,
23 const char *ext)
24 {
25 strbuf_reset(buf);
26 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
27 hash_to_hex(hash), ext);
28 return buf->buf;
29 }
30
31 char *sha1_pack_name(const unsigned char *sha1)
32 {
33 static struct strbuf buf = STRBUF_INIT;
34 return odb_pack_name(&buf, sha1, "pack");
35 }
36
37 char *sha1_pack_index_name(const unsigned char *sha1)
38 {
39 static struct strbuf buf = STRBUF_INIT;
40 return odb_pack_name(&buf, sha1, "idx");
41 }
42
43 static unsigned int pack_used_ctr;
44 static unsigned int pack_mmap_calls;
45 static unsigned int peak_pack_open_windows;
46 static unsigned int pack_open_windows;
47 static unsigned int pack_open_fds;
48 static unsigned int pack_max_fds;
49 static size_t peak_pack_mapped;
50 static size_t pack_mapped;
51
52 #define SZ_FMT PRIuMAX
53 static inline uintmax_t sz_fmt(size_t s) { return s; }
54
55 void pack_report(void)
56 {
57 fprintf(stderr,
58 "pack_report: getpagesize() = %10" SZ_FMT "\n"
59 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
61 sz_fmt(getpagesize()),
62 sz_fmt(packed_git_window_size),
63 sz_fmt(packed_git_limit));
64 fprintf(stderr,
65 "pack_report: pack_used_ctr = %10u\n"
66 "pack_report: pack_mmap_calls = %10u\n"
67 "pack_report: pack_open_windows = %10u / %10u\n"
68 "pack_report: pack_mapped = "
69 "%10" SZ_FMT " / %10" SZ_FMT "\n",
70 pack_used_ctr,
71 pack_mmap_calls,
72 pack_open_windows, peak_pack_open_windows,
73 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
74 }
75
76 /*
77 * Open and mmap the index file at path, perform a couple of
78 * consistency checks, then record its information to p. Return 0 on
79 * success.
80 */
81 static int check_packed_git_idx(const char *path, struct packed_git *p)
82 {
83 void *idx_map;
84 size_t idx_size;
85 int fd = git_open(path), ret;
86 struct stat st;
87 const unsigned int hashsz = the_hash_algo->rawsz;
88
89 if (fd < 0)
90 return -1;
91 if (fstat(fd, &st)) {
92 close(fd);
93 return -1;
94 }
95 idx_size = xsize_t(st.st_size);
96 if (idx_size < 4 * 256 + hashsz + hashsz) {
97 close(fd);
98 return error("index file %s is too small", path);
99 }
100 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
101 close(fd);
102
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
104
105 if (ret)
106 munmap(idx_map, idx_size);
107
108 return ret;
109 }
110
111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
112 size_t idx_size, struct packed_git *p)
113 {
114 struct pack_idx_header *hdr = idx_map;
115 uint32_t version, nr, i, *index;
116
117 if (idx_size < 4 * 256 + hashsz + hashsz)
118 return error("index file %s is too small", path);
119 if (idx_map == NULL)
120 return error("empty data");
121
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
123 version = ntohl(hdr->idx_version);
124 if (version < 2 || version > 2)
125 return error("index file %s is version %"PRIu32
126 " and is not supported by this binary"
127 " (try upgrading GIT to a newer version)",
128 path, version);
129 } else
130 version = 1;
131
132 nr = 0;
133 index = idx_map;
134 if (version > 1)
135 index += 2; /* skip index header */
136 for (i = 0; i < 256; i++) {
137 uint32_t n = ntohl(index[i]);
138 if (n < nr)
139 return error("non-monotonic index %s", path);
140 nr = n;
141 }
142
143 if (version == 1) {
144 /*
145 * Total size:
146 * - 256 index entries 4 bytes each
147 * - 24-byte entries * nr (object ID + 4-byte offset)
148 * - hash of the packfile
149 * - file checksum
150 */
151 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
152 return error("wrong index v1 file size in %s", path);
153 } else if (version == 2) {
154 /*
155 * Minimum size:
156 * - 8 bytes of header
157 * - 256 index entries 4 bytes each
158 * - object ID entry * nr
159 * - 4-byte crc entry * nr
160 * - 4-byte offset entry * nr
161 * - hash of the packfile
162 * - file checksum
163 * And after the 4-byte offset table might be a
164 * variable sized table containing 8-byte entries
165 * for offsets larger than 2^31.
166 */
167 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
168 unsigned long max_size = min_size;
169 if (nr)
170 max_size += (nr - 1)*8;
171 if (idx_size < min_size || idx_size > max_size)
172 return error("wrong index v2 file size in %s", path);
173 if (idx_size != min_size &&
174 /*
175 * make sure we can deal with large pack offsets.
176 * 31-bit signed offset won't be enough, neither
177 * 32-bit unsigned one will be.
178 */
179 (sizeof(off_t) <= 4))
180 return error("pack too large for current definition of off_t in %s", path);
181 }
182
183 p->index_version = version;
184 p->index_data = idx_map;
185 p->index_size = idx_size;
186 p->num_objects = nr;
187 return 0;
188 }
189
190 int open_pack_index(struct packed_git *p)
191 {
192 char *idx_name;
193 size_t len;
194 int ret;
195
196 if (p->index_data)
197 return 0;
198
199 if (!strip_suffix(p->pack_name, ".pack", &len))
200 BUG("pack_name does not end in .pack");
201 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
202 ret = check_packed_git_idx(idx_name, p);
203 free(idx_name);
204 return ret;
205 }
206
207 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
208 {
209 const uint32_t *level1_ofs = p->index_data;
210
211 if (!level1_ofs) {
212 if (open_pack_index(p))
213 return 0;
214 level1_ofs = p->index_data;
215 }
216
217 if (p->index_version > 1) {
218 level1_ofs += 2;
219 }
220
221 return ntohl(level1_ofs[value]);
222 }
223
224 static struct packed_git *alloc_packed_git(int extra)
225 {
226 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
227 memset(p, 0, sizeof(*p));
228 p->pack_fd = -1;
229 return p;
230 }
231
232 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
233 {
234 const char *path = sha1_pack_name(sha1);
235 size_t alloc = st_add(strlen(path), 1);
236 struct packed_git *p = alloc_packed_git(alloc);
237
238 memcpy(p->pack_name, path, alloc); /* includes NUL */
239 hashcpy(p->hash, sha1);
240 if (check_packed_git_idx(idx_path, p)) {
241 free(p);
242 return NULL;
243 }
244
245 return p;
246 }
247
248 static void scan_windows(struct packed_git *p,
249 struct packed_git **lru_p,
250 struct pack_window **lru_w,
251 struct pack_window **lru_l)
252 {
253 struct pack_window *w, *w_l;
254
255 for (w_l = NULL, w = p->windows; w; w = w->next) {
256 if (!w->inuse_cnt) {
257 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
258 *lru_p = p;
259 *lru_w = w;
260 *lru_l = w_l;
261 }
262 }
263 w_l = w;
264 }
265 }
266
267 static int unuse_one_window(struct packed_git *current)
268 {
269 struct packed_git *p, *lru_p = NULL;
270 struct pack_window *lru_w = NULL, *lru_l = NULL;
271
272 if (current)
273 scan_windows(current, &lru_p, &lru_w, &lru_l);
274 for (p = the_repository->objects->packed_git; p; p = p->next)
275 scan_windows(p, &lru_p, &lru_w, &lru_l);
276 if (lru_p) {
277 munmap(lru_w->base, lru_w->len);
278 pack_mapped -= lru_w->len;
279 if (lru_l)
280 lru_l->next = lru_w->next;
281 else
282 lru_p->windows = lru_w->next;
283 free(lru_w);
284 pack_open_windows--;
285 return 1;
286 }
287 return 0;
288 }
289
290 void 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 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_object_store(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 if (o->multi_pack_index) {
344 close_midx(o->multi_pack_index);
345 o->multi_pack_index = NULL;
346 }
347
348 close_commit_graph(o);
349 }
350
351 void unlink_pack_path(const char *pack_name, int force_delete)
352 {
353 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
354 int i;
355 struct strbuf buf = STRBUF_INIT;
356 size_t plen;
357
358 strbuf_addstr(&buf, pack_name);
359 strip_suffix_mem(buf.buf, &buf.len, ".pack");
360 plen = buf.len;
361
362 if (!force_delete) {
363 strbuf_addstr(&buf, ".keep");
364 if (!access(buf.buf, F_OK)) {
365 strbuf_release(&buf);
366 return;
367 }
368 }
369
370 for (i = 0; i < ARRAY_SIZE(exts); i++) {
371 strbuf_setlen(&buf, plen);
372 strbuf_addstr(&buf, exts[i]);
373 unlink(buf.buf);
374 }
375
376 strbuf_release(&buf);
377 }
378
379 /*
380 * The LRU pack is the one with the oldest MRU window, preferring packs
381 * with no used windows, or the oldest mtime if it has no windows allocated.
382 */
383 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
384 {
385 struct pack_window *w, *this_mru_w;
386 int has_windows_inuse = 0;
387
388 /*
389 * Reject this pack if it has windows and the previously selected
390 * one does not. If this pack does not have windows, reject
391 * it if the pack file is newer than the previously selected one.
392 */
393 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
394 return;
395
396 for (w = this_mru_w = p->windows; w; w = w->next) {
397 /*
398 * Reject this pack if any of its windows are in use,
399 * but the previously selected pack did not have any
400 * inuse windows. Otherwise, record that this pack
401 * has windows in use.
402 */
403 if (w->inuse_cnt) {
404 if (*accept_windows_inuse)
405 has_windows_inuse = 1;
406 else
407 return;
408 }
409
410 if (w->last_used > this_mru_w->last_used)
411 this_mru_w = w;
412
413 /*
414 * Reject this pack if it has windows that have been
415 * used more recently than the previously selected pack.
416 * If the previously selected pack had windows inuse and
417 * we have not encountered a window in this pack that is
418 * inuse, skip this check since we prefer a pack with no
419 * inuse windows to one that has inuse windows.
420 */
421 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
422 this_mru_w->last_used > (*mru_w)->last_used)
423 return;
424 }
425
426 /*
427 * Select this pack.
428 */
429 *mru_w = this_mru_w;
430 *lru_p = p;
431 *accept_windows_inuse = has_windows_inuse;
432 }
433
434 static int close_one_pack(void)
435 {
436 struct packed_git *p, *lru_p = NULL;
437 struct pack_window *mru_w = NULL;
438 int accept_windows_inuse = 1;
439
440 for (p = the_repository->objects->packed_git; p; p = p->next) {
441 if (p->pack_fd == -1)
442 continue;
443 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
444 }
445
446 if (lru_p)
447 return close_pack_fd(lru_p);
448
449 return 0;
450 }
451
452 static unsigned int get_max_fd_limit(void)
453 {
454 #ifdef RLIMIT_NOFILE
455 {
456 struct rlimit lim;
457
458 if (!getrlimit(RLIMIT_NOFILE, &lim))
459 return lim.rlim_cur;
460 }
461 #endif
462
463 #ifdef _SC_OPEN_MAX
464 {
465 long open_max = sysconf(_SC_OPEN_MAX);
466 if (0 < open_max)
467 return open_max;
468 /*
469 * Otherwise, we got -1 for one of the two
470 * reasons:
471 *
472 * (1) sysconf() did not understand _SC_OPEN_MAX
473 * and signaled an error with -1; or
474 * (2) sysconf() said there is no limit.
475 *
476 * We _could_ clear errno before calling sysconf() to
477 * tell these two cases apart and return a huge number
478 * in the latter case to let the caller cap it to a
479 * value that is not so selfish, but letting the
480 * fallback OPEN_MAX codepath take care of these cases
481 * is a lot simpler.
482 */
483 }
484 #endif
485
486 #ifdef OPEN_MAX
487 return OPEN_MAX;
488 #else
489 return 1; /* see the caller ;-) */
490 #endif
491 }
492
493 const char *pack_basename(struct packed_git *p)
494 {
495 const char *ret = strrchr(p->pack_name, '/');
496 if (ret)
497 ret = ret + 1; /* skip past slash */
498 else
499 ret = p->pack_name; /* we only have a base */
500 return ret;
501 }
502
503 /*
504 * Do not call this directly as this leaks p->pack_fd on error return;
505 * call open_packed_git() instead.
506 */
507 static int open_packed_git_1(struct packed_git *p)
508 {
509 struct stat st;
510 struct pack_header hdr;
511 unsigned char hash[GIT_MAX_RAWSZ];
512 unsigned char *idx_hash;
513 long fd_flag;
514 ssize_t read_result;
515 const unsigned hashsz = the_hash_algo->rawsz;
516
517 if (!p->index_data) {
518 struct multi_pack_index *m;
519 const char *pack_name = pack_basename(p);
520
521 for (m = the_repository->objects->multi_pack_index;
522 m; m = m->next) {
523 if (midx_contains_pack(m, pack_name))
524 break;
525 }
526
527 if (!m && open_pack_index(p))
528 return error("packfile %s index unavailable", p->pack_name);
529 }
530
531 if (!pack_max_fds) {
532 unsigned int max_fds = get_max_fd_limit();
533
534 /* Save 3 for stdin/stdout/stderr, 22 for work */
535 if (25 < max_fds)
536 pack_max_fds = max_fds - 25;
537 else
538 pack_max_fds = 1;
539 }
540
541 while (pack_max_fds <= pack_open_fds && close_one_pack())
542 ; /* nothing */
543
544 p->pack_fd = git_open(p->pack_name);
545 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
546 return -1;
547 pack_open_fds++;
548
549 /* If we created the struct before we had the pack we lack size. */
550 if (!p->pack_size) {
551 if (!S_ISREG(st.st_mode))
552 return error("packfile %s not a regular file", p->pack_name);
553 p->pack_size = st.st_size;
554 } else if (p->pack_size != st.st_size)
555 return error("packfile %s size changed", p->pack_name);
556
557 /* We leave these file descriptors open with sliding mmap;
558 * there is no point keeping them open across exec(), though.
559 */
560 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
561 if (fd_flag < 0)
562 return error("cannot determine file descriptor flags");
563 fd_flag |= FD_CLOEXEC;
564 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
565 return error("cannot set FD_CLOEXEC");
566
567 /* Verify we recognize this pack file format. */
568 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
569 if (read_result < 0)
570 return error_errno("error reading from %s", p->pack_name);
571 if (read_result != sizeof(hdr))
572 return error("file %s is far too short to be a packfile", p->pack_name);
573 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
574 return error("file %s is not a GIT packfile", p->pack_name);
575 if (!pack_version_ok(hdr.hdr_version))
576 return error("packfile %s is version %"PRIu32" and not"
577 " supported (try upgrading GIT to a newer version)",
578 p->pack_name, ntohl(hdr.hdr_version));
579
580 /* Skip index checking if in multi-pack-index */
581 if (!p->index_data)
582 return 0;
583
584 /* Verify the pack matches its index. */
585 if (p->num_objects != ntohl(hdr.hdr_entries))
586 return error("packfile %s claims to have %"PRIu32" objects"
587 " while index indicates %"PRIu32" objects",
588 p->pack_name, ntohl(hdr.hdr_entries),
589 p->num_objects);
590 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
591 return error("end of packfile %s is unavailable", p->pack_name);
592 read_result = read_in_full(p->pack_fd, hash, hashsz);
593 if (read_result < 0)
594 return error_errno("error reading from %s", p->pack_name);
595 if (read_result != hashsz)
596 return error("packfile %s signature is unavailable", p->pack_name);
597 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
598 if (!hasheq(hash, idx_hash))
599 return error("packfile %s does not match index", p->pack_name);
600 return 0;
601 }
602
603 static int open_packed_git(struct packed_git *p)
604 {
605 if (!open_packed_git_1(p))
606 return 0;
607 close_pack_fd(p);
608 return -1;
609 }
610
611 static int in_window(struct pack_window *win, off_t offset)
612 {
613 /* We must promise at least one full hash after the
614 * offset is available from this window, otherwise the offset
615 * is not actually in this window and a different window (which
616 * has that one hash excess) must be used. This is to support
617 * the object header and delta base parsing routines below.
618 */
619 off_t win_off = win->offset;
620 return win_off <= offset
621 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
622 }
623
624 unsigned char *use_pack(struct packed_git *p,
625 struct pack_window **w_cursor,
626 off_t offset,
627 unsigned long *left)
628 {
629 struct pack_window *win = *w_cursor;
630
631 /* Since packfiles end in a hash of their content and it's
632 * pointless to ask for an offset into the middle of that
633 * hash, and the in_window function above wouldn't match
634 * don't allow an offset too close to the end of the file.
635 */
636 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
637 die("packfile %s cannot be accessed", p->pack_name);
638 if (offset > (p->pack_size - the_hash_algo->rawsz))
639 die("offset beyond end of packfile (truncated pack?)");
640 if (offset < 0)
641 die(_("offset before end of packfile (broken .idx?)"));
642
643 if (!win || !in_window(win, offset)) {
644 if (win)
645 win->inuse_cnt--;
646 for (win = p->windows; win; win = win->next) {
647 if (in_window(win, offset))
648 break;
649 }
650 if (!win) {
651 size_t window_align = packed_git_window_size / 2;
652 off_t len;
653
654 if (p->pack_fd == -1 && open_packed_git(p))
655 die("packfile %s cannot be accessed", p->pack_name);
656
657 win = xcalloc(1, sizeof(*win));
658 win->offset = (offset / window_align) * window_align;
659 len = p->pack_size - win->offset;
660 if (len > packed_git_window_size)
661 len = packed_git_window_size;
662 win->len = (size_t)len;
663 pack_mapped += win->len;
664 while (packed_git_limit < pack_mapped
665 && unuse_one_window(p))
666 ; /* nothing */
667 win->base = xmmap_gently(NULL, win->len,
668 PROT_READ, MAP_PRIVATE,
669 p->pack_fd, win->offset);
670 if (win->base == MAP_FAILED)
671 die_errno("packfile %s cannot be mapped",
672 p->pack_name);
673 if (!win->offset && win->len == p->pack_size
674 && !p->do_not_close)
675 close_pack_fd(p);
676 pack_mmap_calls++;
677 pack_open_windows++;
678 if (pack_mapped > peak_pack_mapped)
679 peak_pack_mapped = pack_mapped;
680 if (pack_open_windows > peak_pack_open_windows)
681 peak_pack_open_windows = pack_open_windows;
682 win->next = p->windows;
683 p->windows = win;
684 }
685 }
686 if (win != *w_cursor) {
687 win->last_used = pack_used_ctr++;
688 win->inuse_cnt++;
689 *w_cursor = win;
690 }
691 offset -= win->offset;
692 if (left)
693 *left = win->len - xsize_t(offset);
694 return win->base + offset;
695 }
696
697 void unuse_pack(struct pack_window **w_cursor)
698 {
699 struct pack_window *w = *w_cursor;
700 if (w) {
701 w->inuse_cnt--;
702 *w_cursor = NULL;
703 }
704 }
705
706 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
707 {
708 struct stat st;
709 size_t alloc;
710 struct packed_git *p;
711
712 /*
713 * Make sure a corresponding .pack file exists and that
714 * the index looks sane.
715 */
716 if (!strip_suffix_mem(path, &path_len, ".idx"))
717 return NULL;
718
719 /*
720 * ".promisor" is long enough to hold any suffix we're adding (and
721 * the use xsnprintf double-checks that)
722 */
723 alloc = st_add3(path_len, strlen(".promisor"), 1);
724 p = alloc_packed_git(alloc);
725 memcpy(p->pack_name, path, path_len);
726
727 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
728 if (!access(p->pack_name, F_OK))
729 p->pack_keep = 1;
730
731 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
732 if (!access(p->pack_name, F_OK))
733 p->pack_promisor = 1;
734
735 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
736 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
737 free(p);
738 return NULL;
739 }
740
741 /* ok, it looks sane as far as we can check without
742 * actually mapping the pack file.
743 */
744 p->pack_size = st.st_size;
745 p->pack_local = local;
746 p->mtime = st.st_mtime;
747 if (path_len < the_hash_algo->hexsz ||
748 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
749 hashclr(p->hash);
750 return p;
751 }
752
753 void install_packed_git(struct repository *r, struct packed_git *pack)
754 {
755 if (pack->pack_fd != -1)
756 pack_open_fds++;
757
758 pack->next = r->objects->packed_git;
759 r->objects->packed_git = pack;
760 }
761
762 void (*report_garbage)(unsigned seen_bits, const char *path);
763
764 static void report_helper(const struct string_list *list,
765 int seen_bits, int first, int last)
766 {
767 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
768 return;
769
770 for (; first < last; first++)
771 report_garbage(seen_bits, list->items[first].string);
772 }
773
774 static void report_pack_garbage(struct string_list *list)
775 {
776 int i, baselen = -1, first = 0, seen_bits = 0;
777
778 if (!report_garbage)
779 return;
780
781 string_list_sort(list);
782
783 for (i = 0; i < list->nr; i++) {
784 const char *path = list->items[i].string;
785 if (baselen != -1 &&
786 strncmp(path, list->items[first].string, baselen)) {
787 report_helper(list, seen_bits, first, i);
788 baselen = -1;
789 seen_bits = 0;
790 }
791 if (baselen == -1) {
792 const char *dot = strrchr(path, '.');
793 if (!dot) {
794 report_garbage(PACKDIR_FILE_GARBAGE, path);
795 continue;
796 }
797 baselen = dot - path + 1;
798 first = i;
799 }
800 if (!strcmp(path + baselen, "pack"))
801 seen_bits |= 1;
802 else if (!strcmp(path + baselen, "idx"))
803 seen_bits |= 2;
804 }
805 report_helper(list, seen_bits, first, list->nr);
806 }
807
808 void for_each_file_in_pack_dir(const char *objdir,
809 each_file_in_pack_dir_fn fn,
810 void *data)
811 {
812 struct strbuf path = STRBUF_INIT;
813 size_t dirnamelen;
814 DIR *dir;
815 struct dirent *de;
816
817 strbuf_addstr(&path, objdir);
818 strbuf_addstr(&path, "/pack");
819 dir = opendir(path.buf);
820 if (!dir) {
821 if (errno != ENOENT)
822 error_errno("unable to open object pack directory: %s",
823 path.buf);
824 strbuf_release(&path);
825 return;
826 }
827 strbuf_addch(&path, '/');
828 dirnamelen = path.len;
829 while ((de = readdir(dir)) != NULL) {
830 if (is_dot_or_dotdot(de->d_name))
831 continue;
832
833 strbuf_setlen(&path, dirnamelen);
834 strbuf_addstr(&path, de->d_name);
835
836 fn(path.buf, path.len, de->d_name, data);
837 }
838
839 closedir(dir);
840 strbuf_release(&path);
841 }
842
843 struct prepare_pack_data {
844 struct repository *r;
845 struct string_list *garbage;
846 int local;
847 struct multi_pack_index *m;
848 };
849
850 static void prepare_pack(const char *full_name, size_t full_name_len,
851 const char *file_name, void *_data)
852 {
853 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
854 struct packed_git *p;
855 size_t base_len = full_name_len;
856
857 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
858 !(data->m && midx_contains_pack(data->m, file_name))) {
859 /* Don't reopen a pack we already have. */
860 for (p = data->r->objects->packed_git; p; p = p->next) {
861 size_t len;
862 if (strip_suffix(p->pack_name, ".pack", &len) &&
863 len == base_len &&
864 !memcmp(p->pack_name, full_name, len))
865 break;
866 }
867
868 if (!p) {
869 p = add_packed_git(full_name, full_name_len, data->local);
870 if (p)
871 install_packed_git(data->r, p);
872 }
873 }
874
875 if (!report_garbage)
876 return;
877
878 if (!strcmp(file_name, "multi-pack-index"))
879 return;
880 if (ends_with(file_name, ".idx") ||
881 ends_with(file_name, ".pack") ||
882 ends_with(file_name, ".bitmap") ||
883 ends_with(file_name, ".keep") ||
884 ends_with(file_name, ".promisor"))
885 string_list_append(data->garbage, full_name);
886 else
887 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
888 }
889
890 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
891 {
892 struct prepare_pack_data data;
893 struct string_list garbage = STRING_LIST_INIT_DUP;
894
895 data.m = r->objects->multi_pack_index;
896
897 /* look for the multi-pack-index for this object directory */
898 while (data.m && strcmp(data.m->object_dir, objdir))
899 data.m = data.m->next;
900
901 data.r = r;
902 data.garbage = &garbage;
903 data.local = local;
904
905 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
906
907 report_pack_garbage(data.garbage);
908 string_list_clear(data.garbage, 0);
909 }
910
911 static void prepare_packed_git(struct repository *r);
912 /*
913 * Give a fast, rough count of the number of objects in the repository. This
914 * ignores loose objects completely. If you have a lot of them, then either
915 * you should repack because your performance will be awful, or they are
916 * all unreachable objects about to be pruned, in which case they're not really
917 * interesting as a measure of repo size in the first place.
918 */
919 unsigned long repo_approximate_object_count(struct repository *r)
920 {
921 if (!r->objects->approximate_object_count_valid) {
922 unsigned long count;
923 struct multi_pack_index *m;
924 struct packed_git *p;
925
926 prepare_packed_git(r);
927 count = 0;
928 for (m = get_multi_pack_index(r); m; m = m->next)
929 count += m->num_objects;
930 for (p = r->objects->packed_git; p; p = p->next) {
931 if (open_pack_index(p))
932 continue;
933 count += p->num_objects;
934 }
935 r->objects->approximate_object_count = count;
936 }
937 return r->objects->approximate_object_count;
938 }
939
940 static void *get_next_packed_git(const void *p)
941 {
942 return ((const struct packed_git *)p)->next;
943 }
944
945 static void set_next_packed_git(void *p, void *next)
946 {
947 ((struct packed_git *)p)->next = next;
948 }
949
950 static int sort_pack(const void *a_, const void *b_)
951 {
952 const struct packed_git *a = a_;
953 const struct packed_git *b = b_;
954 int st;
955
956 /*
957 * Local packs tend to contain objects specific to our
958 * variant of the project than remote ones. In addition,
959 * remote ones could be on a network mounted filesystem.
960 * Favor local ones for these reasons.
961 */
962 st = a->pack_local - b->pack_local;
963 if (st)
964 return -st;
965
966 /*
967 * Younger packs tend to contain more recent objects,
968 * and more recent objects tend to get accessed more
969 * often.
970 */
971 if (a->mtime < b->mtime)
972 return 1;
973 else if (a->mtime == b->mtime)
974 return 0;
975 return -1;
976 }
977
978 static void rearrange_packed_git(struct repository *r)
979 {
980 r->objects->packed_git = llist_mergesort(
981 r->objects->packed_git, get_next_packed_git,
982 set_next_packed_git, sort_pack);
983 }
984
985 static void prepare_packed_git_mru(struct repository *r)
986 {
987 struct packed_git *p;
988
989 INIT_LIST_HEAD(&r->objects->packed_git_mru);
990
991 for (p = r->objects->packed_git; p; p = p->next)
992 list_add_tail(&p->mru, &r->objects->packed_git_mru);
993 }
994
995 static void prepare_packed_git(struct repository *r)
996 {
997 struct object_directory *odb;
998
999 if (r->objects->packed_git_initialized)
1000 return;
1001
1002 prepare_alt_odb(r);
1003 for (odb = r->objects->odb; odb; odb = odb->next) {
1004 int local = (odb == r->objects->odb);
1005 prepare_multi_pack_index_one(r, odb->path, local);
1006 prepare_packed_git_one(r, odb->path, local);
1007 }
1008 rearrange_packed_git(r);
1009
1010 prepare_packed_git_mru(r);
1011 r->objects->packed_git_initialized = 1;
1012 }
1013
1014 void reprepare_packed_git(struct repository *r)
1015 {
1016 struct object_directory *odb;
1017
1018 for (odb = r->objects->odb; odb; odb = odb->next)
1019 odb_clear_loose_cache(odb);
1020
1021 r->objects->approximate_object_count_valid = 0;
1022 r->objects->packed_git_initialized = 0;
1023 prepare_packed_git(r);
1024 }
1025
1026 struct packed_git *get_packed_git(struct repository *r)
1027 {
1028 prepare_packed_git(r);
1029 return r->objects->packed_git;
1030 }
1031
1032 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1033 {
1034 prepare_packed_git(r);
1035 return r->objects->multi_pack_index;
1036 }
1037
1038 struct packed_git *get_all_packs(struct repository *r)
1039 {
1040 struct multi_pack_index *m;
1041
1042 prepare_packed_git(r);
1043 for (m = r->objects->multi_pack_index; m; m = m->next) {
1044 uint32_t i;
1045 for (i = 0; i < m->num_packs; i++)
1046 prepare_midx_pack(r, m, i);
1047 }
1048
1049 return r->objects->packed_git;
1050 }
1051
1052 struct list_head *get_packed_git_mru(struct repository *r)
1053 {
1054 prepare_packed_git(r);
1055 return &r->objects->packed_git_mru;
1056 }
1057
1058 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1059 unsigned long len, enum object_type *type, unsigned long *sizep)
1060 {
1061 unsigned shift;
1062 unsigned long size, c;
1063 unsigned long used = 0;
1064
1065 c = buf[used++];
1066 *type = (c >> 4) & 7;
1067 size = c & 15;
1068 shift = 4;
1069 while (c & 0x80) {
1070 if (len <= used || bitsizeof(long) <= shift) {
1071 error("bad object header");
1072 size = used = 0;
1073 break;
1074 }
1075 c = buf[used++];
1076 size += (c & 0x7f) << shift;
1077 shift += 7;
1078 }
1079 *sizep = size;
1080 return used;
1081 }
1082
1083 unsigned long get_size_from_delta(struct packed_git *p,
1084 struct pack_window **w_curs,
1085 off_t curpos)
1086 {
1087 const unsigned char *data;
1088 unsigned char delta_head[20], *in;
1089 git_zstream stream;
1090 int st;
1091
1092 memset(&stream, 0, sizeof(stream));
1093 stream.next_out = delta_head;
1094 stream.avail_out = sizeof(delta_head);
1095
1096 git_inflate_init(&stream);
1097 do {
1098 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1099 stream.next_in = in;
1100 st = git_inflate(&stream, Z_FINISH);
1101 curpos += stream.next_in - in;
1102 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1103 stream.total_out < sizeof(delta_head));
1104 git_inflate_end(&stream);
1105 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1106 error("delta data unpack-initial failed");
1107 return 0;
1108 }
1109
1110 /* Examine the initial part of the delta to figure out
1111 * the result size.
1112 */
1113 data = delta_head;
1114
1115 /* ignore base size */
1116 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1117
1118 /* Read the result size */
1119 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1120 }
1121
1122 int unpack_object_header(struct packed_git *p,
1123 struct pack_window **w_curs,
1124 off_t *curpos,
1125 unsigned long *sizep)
1126 {
1127 unsigned char *base;
1128 unsigned long left;
1129 unsigned long used;
1130 enum object_type type;
1131
1132 /* use_pack() assures us we have [base, base + 20) available
1133 * as a range that we can look at. (Its actually the hash
1134 * size that is assured.) With our object header encoding
1135 * the maximum deflated object size is 2^137, which is just
1136 * insane, so we know won't exceed what we have been given.
1137 */
1138 base = use_pack(p, w_curs, *curpos, &left);
1139 used = unpack_object_header_buffer(base, left, &type, sizep);
1140 if (!used) {
1141 type = OBJ_BAD;
1142 } else
1143 *curpos += used;
1144
1145 return type;
1146 }
1147
1148 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1149 {
1150 unsigned i;
1151 const unsigned hashsz = the_hash_algo->rawsz;
1152 for (i = 0; i < p->num_bad_objects; i++)
1153 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1154 return;
1155 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1156 st_mult(GIT_MAX_RAWSZ,
1157 st_add(p->num_bad_objects, 1)));
1158 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1159 p->num_bad_objects++;
1160 }
1161
1162 const struct packed_git *has_packed_and_bad(struct repository *r,
1163 const unsigned char *sha1)
1164 {
1165 struct packed_git *p;
1166 unsigned i;
1167
1168 for (p = r->objects->packed_git; p; p = p->next)
1169 for (i = 0; i < p->num_bad_objects; i++)
1170 if (hasheq(sha1,
1171 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1172 return p;
1173 return NULL;
1174 }
1175
1176 static off_t get_delta_base(struct packed_git *p,
1177 struct pack_window **w_curs,
1178 off_t *curpos,
1179 enum object_type type,
1180 off_t delta_obj_offset)
1181 {
1182 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1183 off_t base_offset;
1184
1185 /* use_pack() assured us we have [base_info, base_info + 20)
1186 * as a range that we can look at without walking off the
1187 * end of the mapped window. Its actually the hash size
1188 * that is assured. An OFS_DELTA longer than the hash size
1189 * is stupid, as then a REF_DELTA would be smaller to store.
1190 */
1191 if (type == OBJ_OFS_DELTA) {
1192 unsigned used = 0;
1193 unsigned char c = base_info[used++];
1194 base_offset = c & 127;
1195 while (c & 128) {
1196 base_offset += 1;
1197 if (!base_offset || MSB(base_offset, 7))
1198 return 0; /* overflow */
1199 c = base_info[used++];
1200 base_offset = (base_offset << 7) + (c & 127);
1201 }
1202 base_offset = delta_obj_offset - base_offset;
1203 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1204 return 0; /* out of bound */
1205 *curpos += used;
1206 } else if (type == OBJ_REF_DELTA) {
1207 /* The base entry _must_ be in the same pack */
1208 base_offset = find_pack_entry_one(base_info, p);
1209 *curpos += the_hash_algo->rawsz;
1210 } else
1211 die("I am totally screwed");
1212 return base_offset;
1213 }
1214
1215 /*
1216 * Like get_delta_base above, but we return the sha1 instead of the pack
1217 * offset. This means it is cheaper for REF deltas (we do not have to do
1218 * the final object lookup), but more expensive for OFS deltas (we
1219 * have to load the revidx to convert the offset back into a sha1).
1220 */
1221 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1222 struct pack_window **w_curs,
1223 off_t curpos,
1224 enum object_type type,
1225 off_t delta_obj_offset)
1226 {
1227 if (type == OBJ_REF_DELTA) {
1228 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1229 return base;
1230 } else if (type == OBJ_OFS_DELTA) {
1231 struct revindex_entry *revidx;
1232 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1233 type, delta_obj_offset);
1234
1235 if (!base_offset)
1236 return NULL;
1237
1238 revidx = find_pack_revindex(p, base_offset);
1239 if (!revidx)
1240 return NULL;
1241
1242 return nth_packed_object_sha1(p, revidx->nr);
1243 } else
1244 return NULL;
1245 }
1246
1247 static int retry_bad_packed_offset(struct repository *r,
1248 struct packed_git *p,
1249 off_t obj_offset)
1250 {
1251 int type;
1252 struct revindex_entry *revidx;
1253 struct object_id oid;
1254 revidx = find_pack_revindex(p, obj_offset);
1255 if (!revidx)
1256 return OBJ_BAD;
1257 nth_packed_object_oid(&oid, p, revidx->nr);
1258 mark_bad_packed_object(p, oid.hash);
1259 type = oid_object_info(r, &oid, NULL);
1260 if (type <= OBJ_NONE)
1261 return OBJ_BAD;
1262 return type;
1263 }
1264
1265 #define POI_STACK_PREALLOC 64
1266
1267 static enum object_type packed_to_object_type(struct repository *r,
1268 struct packed_git *p,
1269 off_t obj_offset,
1270 enum object_type type,
1271 struct pack_window **w_curs,
1272 off_t curpos)
1273 {
1274 off_t small_poi_stack[POI_STACK_PREALLOC];
1275 off_t *poi_stack = small_poi_stack;
1276 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1277
1278 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1279 off_t base_offset;
1280 unsigned long size;
1281 /* Push the object we're going to leave behind */
1282 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1283 poi_stack_alloc = alloc_nr(poi_stack_nr);
1284 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1285 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1286 } else {
1287 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1288 }
1289 poi_stack[poi_stack_nr++] = obj_offset;
1290 /* If parsing the base offset fails, just unwind */
1291 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1292 if (!base_offset)
1293 goto unwind;
1294 curpos = obj_offset = base_offset;
1295 type = unpack_object_header(p, w_curs, &curpos, &size);
1296 if (type <= OBJ_NONE) {
1297 /* If getting the base itself fails, we first
1298 * retry the base, otherwise unwind */
1299 type = retry_bad_packed_offset(r, p, base_offset);
1300 if (type > OBJ_NONE)
1301 goto out;
1302 goto unwind;
1303 }
1304 }
1305
1306 switch (type) {
1307 case OBJ_BAD:
1308 case OBJ_COMMIT:
1309 case OBJ_TREE:
1310 case OBJ_BLOB:
1311 case OBJ_TAG:
1312 break;
1313 default:
1314 error("unknown object type %i at offset %"PRIuMAX" in %s",
1315 type, (uintmax_t)obj_offset, p->pack_name);
1316 type = OBJ_BAD;
1317 }
1318
1319 out:
1320 if (poi_stack != small_poi_stack)
1321 free(poi_stack);
1322 return type;
1323
1324 unwind:
1325 while (poi_stack_nr) {
1326 obj_offset = poi_stack[--poi_stack_nr];
1327 type = retry_bad_packed_offset(r, p, obj_offset);
1328 if (type > OBJ_NONE)
1329 goto out;
1330 }
1331 type = OBJ_BAD;
1332 goto out;
1333 }
1334
1335 static struct hashmap delta_base_cache;
1336 static size_t delta_base_cached;
1337
1338 static LIST_HEAD(delta_base_cache_lru);
1339
1340 struct delta_base_cache_key {
1341 struct packed_git *p;
1342 off_t base_offset;
1343 };
1344
1345 struct delta_base_cache_entry {
1346 struct hashmap_entry ent;
1347 struct delta_base_cache_key key;
1348 struct list_head lru;
1349 void *data;
1350 unsigned long size;
1351 enum object_type type;
1352 };
1353
1354 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1355 {
1356 unsigned int hash;
1357
1358 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1359 hash += (hash >> 8) + (hash >> 16);
1360 return hash;
1361 }
1362
1363 static struct delta_base_cache_entry *
1364 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1365 {
1366 struct hashmap_entry entry, *e;
1367 struct delta_base_cache_key key;
1368
1369 if (!delta_base_cache.cmpfn)
1370 return NULL;
1371
1372 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1373 key.p = p;
1374 key.base_offset = base_offset;
1375 e = hashmap_get(&delta_base_cache, &entry, &key);
1376 return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1377 }
1378
1379 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1380 const struct delta_base_cache_key *b)
1381 {
1382 return a->p == b->p && a->base_offset == b->base_offset;
1383 }
1384
1385 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1386 const struct hashmap_entry *va,
1387 const struct hashmap_entry *vb,
1388 const void *vkey)
1389 {
1390 const struct delta_base_cache_entry *a, *b;
1391 const struct delta_base_cache_key *key = vkey;
1392
1393 a = container_of(va, const struct delta_base_cache_entry, ent);
1394 b = container_of(vb, const struct delta_base_cache_entry, ent);
1395
1396 if (key)
1397 return !delta_base_cache_key_eq(&a->key, key);
1398 else
1399 return !delta_base_cache_key_eq(&a->key, &b->key);
1400 }
1401
1402 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1403 {
1404 return !!get_delta_base_cache_entry(p, base_offset);
1405 }
1406
1407 /*
1408 * Remove the entry from the cache, but do _not_ free the associated
1409 * entry data. The caller takes ownership of the "data" buffer, and
1410 * should copy out any fields it wants before detaching.
1411 */
1412 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1413 {
1414 hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1415 list_del(&ent->lru);
1416 delta_base_cached -= ent->size;
1417 free(ent);
1418 }
1419
1420 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1421 off_t base_offset, unsigned long *base_size,
1422 enum object_type *type)
1423 {
1424 struct delta_base_cache_entry *ent;
1425
1426 ent = get_delta_base_cache_entry(p, base_offset);
1427 if (!ent)
1428 return unpack_entry(r, p, base_offset, type, base_size);
1429
1430 if (type)
1431 *type = ent->type;
1432 if (base_size)
1433 *base_size = ent->size;
1434 return xmemdupz(ent->data, ent->size);
1435 }
1436
1437 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1438 {
1439 free(ent->data);
1440 detach_delta_base_cache_entry(ent);
1441 }
1442
1443 void clear_delta_base_cache(void)
1444 {
1445 struct list_head *lru, *tmp;
1446 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1447 struct delta_base_cache_entry *entry =
1448 list_entry(lru, struct delta_base_cache_entry, lru);
1449 release_delta_base_cache(entry);
1450 }
1451 }
1452
1453 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1454 void *base, unsigned long base_size, enum object_type type)
1455 {
1456 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1457 struct list_head *lru, *tmp;
1458
1459 delta_base_cached += base_size;
1460
1461 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1462 struct delta_base_cache_entry *f =
1463 list_entry(lru, struct delta_base_cache_entry, lru);
1464 if (delta_base_cached <= delta_base_cache_limit)
1465 break;
1466 release_delta_base_cache(f);
1467 }
1468
1469 ent->key.p = p;
1470 ent->key.base_offset = base_offset;
1471 ent->type = type;
1472 ent->data = base;
1473 ent->size = base_size;
1474 list_add_tail(&ent->lru, &delta_base_cache_lru);
1475
1476 if (!delta_base_cache.cmpfn)
1477 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1478 hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1479 hashmap_add(&delta_base_cache, &ent->ent);
1480 }
1481
1482 int packed_object_info(struct repository *r, struct packed_git *p,
1483 off_t obj_offset, struct object_info *oi)
1484 {
1485 struct pack_window *w_curs = NULL;
1486 unsigned long size;
1487 off_t curpos = obj_offset;
1488 enum object_type type;
1489
1490 /*
1491 * We always get the representation type, but only convert it to
1492 * a "real" type later if the caller is interested.
1493 */
1494 if (oi->contentp) {
1495 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1496 &type);
1497 if (!*oi->contentp)
1498 type = OBJ_BAD;
1499 } else {
1500 type = unpack_object_header(p, &w_curs, &curpos, &size);
1501 }
1502
1503 if (!oi->contentp && oi->sizep) {
1504 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1505 off_t tmp_pos = curpos;
1506 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1507 type, obj_offset);
1508 if (!base_offset) {
1509 type = OBJ_BAD;
1510 goto out;
1511 }
1512 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1513 if (*oi->sizep == 0) {
1514 type = OBJ_BAD;
1515 goto out;
1516 }
1517 } else {
1518 *oi->sizep = size;
1519 }
1520 }
1521
1522 if (oi->disk_sizep) {
1523 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1524 *oi->disk_sizep = revidx[1].offset - obj_offset;
1525 }
1526
1527 if (oi->typep || oi->type_name) {
1528 enum object_type ptot;
1529 ptot = packed_to_object_type(r, p, obj_offset,
1530 type, &w_curs, curpos);
1531 if (oi->typep)
1532 *oi->typep = ptot;
1533 if (oi->type_name) {
1534 const char *tn = type_name(ptot);
1535 if (tn)
1536 strbuf_addstr(oi->type_name, tn);
1537 }
1538 if (ptot < 0) {
1539 type = OBJ_BAD;
1540 goto out;
1541 }
1542 }
1543
1544 if (oi->delta_base_sha1) {
1545 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1546 const unsigned char *base;
1547
1548 base = get_delta_base_sha1(p, &w_curs, curpos,
1549 type, obj_offset);
1550 if (!base) {
1551 type = OBJ_BAD;
1552 goto out;
1553 }
1554
1555 hashcpy(oi->delta_base_sha1, base);
1556 } else
1557 hashclr(oi->delta_base_sha1);
1558 }
1559
1560 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1561 OI_PACKED;
1562
1563 out:
1564 unuse_pack(&w_curs);
1565 return type;
1566 }
1567
1568 static void *unpack_compressed_entry(struct packed_git *p,
1569 struct pack_window **w_curs,
1570 off_t curpos,
1571 unsigned long size)
1572 {
1573 int st;
1574 git_zstream stream;
1575 unsigned char *buffer, *in;
1576
1577 buffer = xmallocz_gently(size);
1578 if (!buffer)
1579 return NULL;
1580 memset(&stream, 0, sizeof(stream));
1581 stream.next_out = buffer;
1582 stream.avail_out = size + 1;
1583
1584 git_inflate_init(&stream);
1585 do {
1586 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1587 stream.next_in = in;
1588 st = git_inflate(&stream, Z_FINISH);
1589 if (!stream.avail_out)
1590 break; /* the payload is larger than it should be */
1591 curpos += stream.next_in - in;
1592 } while (st == Z_OK || st == Z_BUF_ERROR);
1593 git_inflate_end(&stream);
1594 if ((st != Z_STREAM_END) || stream.total_out != size) {
1595 free(buffer);
1596 return NULL;
1597 }
1598
1599 /* versions of zlib can clobber unconsumed portion of outbuf */
1600 buffer[size] = '\0';
1601
1602 return buffer;
1603 }
1604
1605 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1606 {
1607 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1608 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1609 p->pack_name, (uintmax_t)obj_offset);
1610 }
1611
1612 int do_check_packed_object_crc;
1613
1614 #define UNPACK_ENTRY_STACK_PREALLOC 64
1615 struct unpack_entry_stack_ent {
1616 off_t obj_offset;
1617 off_t curpos;
1618 unsigned long size;
1619 };
1620
1621 static void *read_object(struct repository *r,
1622 const struct object_id *oid,
1623 enum object_type *type,
1624 unsigned long *size)
1625 {
1626 struct object_info oi = OBJECT_INFO_INIT;
1627 void *content;
1628 oi.typep = type;
1629 oi.sizep = size;
1630 oi.contentp = &content;
1631
1632 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1633 return NULL;
1634 return content;
1635 }
1636
1637 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1638 enum object_type *final_type, unsigned long *final_size)
1639 {
1640 struct pack_window *w_curs = NULL;
1641 off_t curpos = obj_offset;
1642 void *data = NULL;
1643 unsigned long size;
1644 enum object_type type;
1645 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1646 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1647 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1648 int base_from_cache = 0;
1649
1650 write_pack_access_log(p, obj_offset);
1651
1652 /* PHASE 1: drill down to the innermost base object */
1653 for (;;) {
1654 off_t base_offset;
1655 int i;
1656 struct delta_base_cache_entry *ent;
1657
1658 ent = get_delta_base_cache_entry(p, curpos);
1659 if (ent) {
1660 type = ent->type;
1661 data = ent->data;
1662 size = ent->size;
1663 detach_delta_base_cache_entry(ent);
1664 base_from_cache = 1;
1665 break;
1666 }
1667
1668 if (do_check_packed_object_crc && p->index_version > 1) {
1669 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1670 off_t len = revidx[1].offset - obj_offset;
1671 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1672 struct object_id oid;
1673 nth_packed_object_oid(&oid, p, revidx->nr);
1674 error("bad packed object CRC for %s",
1675 oid_to_hex(&oid));
1676 mark_bad_packed_object(p, oid.hash);
1677 data = NULL;
1678 goto out;
1679 }
1680 }
1681
1682 type = unpack_object_header(p, &w_curs, &curpos, &size);
1683 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1684 break;
1685
1686 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1687 if (!base_offset) {
1688 error("failed to validate delta base reference "
1689 "at offset %"PRIuMAX" from %s",
1690 (uintmax_t)curpos, p->pack_name);
1691 /* bail to phase 2, in hopes of recovery */
1692 data = NULL;
1693 break;
1694 }
1695
1696 /* push object, proceed to base */
1697 if (delta_stack_nr >= delta_stack_alloc
1698 && delta_stack == small_delta_stack) {
1699 delta_stack_alloc = alloc_nr(delta_stack_nr);
1700 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1701 COPY_ARRAY(delta_stack, small_delta_stack,
1702 delta_stack_nr);
1703 } else {
1704 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1705 }
1706 i = delta_stack_nr++;
1707 delta_stack[i].obj_offset = obj_offset;
1708 delta_stack[i].curpos = curpos;
1709 delta_stack[i].size = size;
1710
1711 curpos = obj_offset = base_offset;
1712 }
1713
1714 /* PHASE 2: handle the base */
1715 switch (type) {
1716 case OBJ_OFS_DELTA:
1717 case OBJ_REF_DELTA:
1718 if (data)
1719 BUG("unpack_entry: left loop at a valid delta");
1720 break;
1721 case OBJ_COMMIT:
1722 case OBJ_TREE:
1723 case OBJ_BLOB:
1724 case OBJ_TAG:
1725 if (!base_from_cache)
1726 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1727 break;
1728 default:
1729 data = NULL;
1730 error("unknown object type %i at offset %"PRIuMAX" in %s",
1731 type, (uintmax_t)obj_offset, p->pack_name);
1732 }
1733
1734 /* PHASE 3: apply deltas in order */
1735
1736 /* invariants:
1737 * 'data' holds the base data, or NULL if there was corruption
1738 */
1739 while (delta_stack_nr) {
1740 void *delta_data;
1741 void *base = data;
1742 void *external_base = NULL;
1743 unsigned long delta_size, base_size = size;
1744 int i;
1745
1746 data = NULL;
1747
1748 if (base)
1749 add_delta_base_cache(p, obj_offset, base, base_size, type);
1750
1751 if (!base) {
1752 /*
1753 * We're probably in deep shit, but let's try to fetch
1754 * the required base anyway from another pack or loose.
1755 * This is costly but should happen only in the presence
1756 * of a corrupted pack, and is better than failing outright.
1757 */
1758 struct revindex_entry *revidx;
1759 struct object_id base_oid;
1760 revidx = find_pack_revindex(p, obj_offset);
1761 if (revidx) {
1762 nth_packed_object_oid(&base_oid, p, revidx->nr);
1763 error("failed to read delta base object %s"
1764 " at offset %"PRIuMAX" from %s",
1765 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1766 p->pack_name);
1767 mark_bad_packed_object(p, base_oid.hash);
1768 base = read_object(r, &base_oid, &type, &base_size);
1769 external_base = base;
1770 }
1771 }
1772
1773 i = --delta_stack_nr;
1774 obj_offset = delta_stack[i].obj_offset;
1775 curpos = delta_stack[i].curpos;
1776 delta_size = delta_stack[i].size;
1777
1778 if (!base)
1779 continue;
1780
1781 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1782
1783 if (!delta_data) {
1784 error("failed to unpack compressed delta "
1785 "at offset %"PRIuMAX" from %s",
1786 (uintmax_t)curpos, p->pack_name);
1787 data = NULL;
1788 free(external_base);
1789 continue;
1790 }
1791
1792 data = patch_delta(base, base_size,
1793 delta_data, delta_size,
1794 &size);
1795
1796 /*
1797 * We could not apply the delta; warn the user, but keep going.
1798 * Our failure will be noticed either in the next iteration of
1799 * the loop, or if this is the final delta, in the caller when
1800 * we return NULL. Those code paths will take care of making
1801 * a more explicit warning and retrying with another copy of
1802 * the object.
1803 */
1804 if (!data)
1805 error("failed to apply delta");
1806
1807 free(delta_data);
1808 free(external_base);
1809 }
1810
1811 if (final_type)
1812 *final_type = type;
1813 if (final_size)
1814 *final_size = size;
1815
1816 out:
1817 unuse_pack(&w_curs);
1818
1819 if (delta_stack != small_delta_stack)
1820 free(delta_stack);
1821
1822 return data;
1823 }
1824
1825 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1826 {
1827 const unsigned char *index_fanout = p->index_data;
1828 const unsigned char *index_lookup;
1829 const unsigned int hashsz = the_hash_algo->rawsz;
1830 int index_lookup_width;
1831
1832 if (!index_fanout)
1833 BUG("bsearch_pack called without a valid pack-index");
1834
1835 index_lookup = index_fanout + 4 * 256;
1836 if (p->index_version == 1) {
1837 index_lookup_width = hashsz + 4;
1838 index_lookup += 4;
1839 } else {
1840 index_lookup_width = hashsz;
1841 index_fanout += 8;
1842 index_lookup += 8;
1843 }
1844
1845 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1846 index_lookup, index_lookup_width, result);
1847 }
1848
1849 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1850 uint32_t n)
1851 {
1852 const unsigned char *index = p->index_data;
1853 const unsigned int hashsz = the_hash_algo->rawsz;
1854 if (!index) {
1855 if (open_pack_index(p))
1856 return NULL;
1857 index = p->index_data;
1858 }
1859 if (n >= p->num_objects)
1860 return NULL;
1861 index += 4 * 256;
1862 if (p->index_version == 1) {
1863 return index + (hashsz + 4) * n + 4;
1864 } else {
1865 index += 8;
1866 return index + hashsz * n;
1867 }
1868 }
1869
1870 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1871 struct packed_git *p,
1872 uint32_t n)
1873 {
1874 const unsigned char *hash = nth_packed_object_sha1(p, n);
1875 if (!hash)
1876 return NULL;
1877 hashcpy(oid->hash, hash);
1878 return oid;
1879 }
1880
1881 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1882 {
1883 const unsigned char *ptr = vptr;
1884 const unsigned char *start = p->index_data;
1885 const unsigned char *end = start + p->index_size;
1886 if (ptr < start)
1887 die(_("offset before start of pack index for %s (corrupt index?)"),
1888 p->pack_name);
1889 /* No need to check for underflow; .idx files must be at least 8 bytes */
1890 if (ptr >= end - 8)
1891 die(_("offset beyond end of pack index for %s (truncated index?)"),
1892 p->pack_name);
1893 }
1894
1895 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1896 {
1897 const unsigned char *index = p->index_data;
1898 const unsigned int hashsz = the_hash_algo->rawsz;
1899 index += 4 * 256;
1900 if (p->index_version == 1) {
1901 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1902 } else {
1903 uint32_t off;
1904 index += 8 + p->num_objects * (hashsz + 4);
1905 off = ntohl(*((uint32_t *)(index + 4 * n)));
1906 if (!(off & 0x80000000))
1907 return off;
1908 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1909 check_pack_index_ptr(p, index);
1910 return get_be64(index);
1911 }
1912 }
1913
1914 off_t find_pack_entry_one(const unsigned char *sha1,
1915 struct packed_git *p)
1916 {
1917 const unsigned char *index = p->index_data;
1918 struct object_id oid;
1919 uint32_t result;
1920
1921 if (!index) {
1922 if (open_pack_index(p))
1923 return 0;
1924 }
1925
1926 hashcpy(oid.hash, sha1);
1927 if (bsearch_pack(&oid, p, &result))
1928 return nth_packed_object_offset(p, result);
1929 return 0;
1930 }
1931
1932 int is_pack_valid(struct packed_git *p)
1933 {
1934 /* An already open pack is known to be valid. */
1935 if (p->pack_fd != -1)
1936 return 1;
1937
1938 /* If the pack has one window completely covering the
1939 * file size, the pack is known to be valid even if
1940 * the descriptor is not currently open.
1941 */
1942 if (p->windows) {
1943 struct pack_window *w = p->windows;
1944
1945 if (!w->offset && w->len == p->pack_size)
1946 return 1;
1947 }
1948
1949 /* Force the pack to open to prove its valid. */
1950 return !open_packed_git(p);
1951 }
1952
1953 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1954 struct packed_git *packs)
1955 {
1956 struct packed_git *p;
1957
1958 for (p = packs; p; p = p->next) {
1959 if (find_pack_entry_one(sha1, p))
1960 return p;
1961 }
1962 return NULL;
1963
1964 }
1965
1966 static int fill_pack_entry(const struct object_id *oid,
1967 struct pack_entry *e,
1968 struct packed_git *p)
1969 {
1970 off_t offset;
1971
1972 if (p->num_bad_objects) {
1973 unsigned i;
1974 for (i = 0; i < p->num_bad_objects; i++)
1975 if (hasheq(oid->hash,
1976 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1977 return 0;
1978 }
1979
1980 offset = find_pack_entry_one(oid->hash, p);
1981 if (!offset)
1982 return 0;
1983
1984 /*
1985 * We are about to tell the caller where they can locate the
1986 * requested object. We better make sure the packfile is
1987 * still here and can be accessed before supplying that
1988 * answer, as it may have been deleted since the index was
1989 * loaded!
1990 */
1991 if (!is_pack_valid(p))
1992 return 0;
1993 e->offset = offset;
1994 e->p = p;
1995 return 1;
1996 }
1997
1998 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1999 {
2000 struct list_head *pos;
2001 struct multi_pack_index *m;
2002
2003 prepare_packed_git(r);
2004 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2005 return 0;
2006
2007 for (m = r->objects->multi_pack_index; m; m = m->next) {
2008 if (fill_midx_entry(r, oid, e, m))
2009 return 1;
2010 }
2011
2012 list_for_each(pos, &r->objects->packed_git_mru) {
2013 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2014 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2015 list_move(&p->mru, &r->objects->packed_git_mru);
2016 return 1;
2017 }
2018 }
2019 return 0;
2020 }
2021
2022 int has_object_pack(const struct object_id *oid)
2023 {
2024 struct pack_entry e;
2025 return find_pack_entry(the_repository, oid, &e);
2026 }
2027
2028 int has_pack_index(const unsigned char *sha1)
2029 {
2030 struct stat st;
2031 if (stat(sha1_pack_index_name(sha1), &st))
2032 return 0;
2033 return 1;
2034 }
2035
2036 int for_each_object_in_pack(struct packed_git *p,
2037 each_packed_object_fn cb, void *data,
2038 enum for_each_object_flags flags)
2039 {
2040 uint32_t i;
2041 int r = 0;
2042
2043 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2044 if (load_pack_revindex(p))
2045 return -1;
2046 }
2047
2048 for (i = 0; i < p->num_objects; i++) {
2049 uint32_t pos;
2050 struct object_id oid;
2051
2052 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2053 pos = p->revindex[i].nr;
2054 else
2055 pos = i;
2056
2057 if (!nth_packed_object_oid(&oid, p, pos))
2058 return error("unable to get sha1 of object %u in %s",
2059 pos, p->pack_name);
2060
2061 r = cb(&oid, p, pos, data);
2062 if (r)
2063 break;
2064 }
2065 return r;
2066 }
2067
2068 int for_each_packed_object(each_packed_object_fn cb, void *data,
2069 enum for_each_object_flags flags)
2070 {
2071 struct packed_git *p;
2072 int r = 0;
2073 int pack_errors = 0;
2074
2075 prepare_packed_git(the_repository);
2076 for (p = get_all_packs(the_repository); p; p = p->next) {
2077 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2078 continue;
2079 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2080 !p->pack_promisor)
2081 continue;
2082 if (open_pack_index(p)) {
2083 pack_errors = 1;
2084 continue;
2085 }
2086 r = for_each_object_in_pack(p, cb, data, flags);
2087 if (r)
2088 break;
2089 }
2090 return r ? r : pack_errors;
2091 }
2092
2093 static int add_promisor_object(const struct object_id *oid,
2094 struct packed_git *pack,
2095 uint32_t pos,
2096 void *set_)
2097 {
2098 struct oidset *set = set_;
2099 struct object *obj = parse_object(the_repository, oid);
2100 if (!obj)
2101 return 1;
2102
2103 oidset_insert(set, oid);
2104
2105 /*
2106 * If this is a tree, commit, or tag, the objects it refers
2107 * to are also promisor objects. (Blobs refer to no objects->)
2108 */
2109 if (obj->type == OBJ_TREE) {
2110 struct tree *tree = (struct tree *)obj;
2111 struct tree_desc desc;
2112 struct name_entry entry;
2113 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2114 /*
2115 * Error messages are given when packs are
2116 * verified, so do not print any here.
2117 */
2118 return 0;
2119 while (tree_entry_gently(&desc, &entry))
2120 oidset_insert(set, &entry.oid);
2121 } else if (obj->type == OBJ_COMMIT) {
2122 struct commit *commit = (struct commit *) obj;
2123 struct commit_list *parents = commit->parents;
2124
2125 oidset_insert(set, get_commit_tree_oid(commit));
2126 for (; parents; parents = parents->next)
2127 oidset_insert(set, &parents->item->object.oid);
2128 } else if (obj->type == OBJ_TAG) {
2129 struct tag *tag = (struct tag *) obj;
2130 oidset_insert(set, get_tagged_oid(tag));
2131 }
2132 return 0;
2133 }
2134
2135 int is_promisor_object(const struct object_id *oid)
2136 {
2137 static struct oidset promisor_objects;
2138 static int promisor_objects_prepared;
2139
2140 if (!promisor_objects_prepared) {
2141 if (has_promisor_remote()) {
2142 for_each_packed_object(add_promisor_object,
2143 &promisor_objects,
2144 FOR_EACH_OBJECT_PROMISOR_ONLY);
2145 }
2146 promisor_objects_prepared = 1;
2147 }
2148 return oidset_contains(&promisor_objects, oid);
2149 }
git for Debian