1#include "cache.h"
2#include "list.h"
3#include "pack.h"
4#include "dir.h"
5#include "mergesort.h"
6#include "packfile.h"
7#include "delta.h"
8#include "list.h"
9#include "streaming.h"
10#include "sha1-lookup.h"
11
12char *odb_pack_name(struct strbuf *buf,
13 const unsigned char *sha1,
14 const char *ext)
15{
16 strbuf_reset(buf);
17 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
18 sha1_to_hex(sha1), ext);
19 return buf->buf;
20}
21
22char *sha1_pack_name(const unsigned char *sha1)
23{
24 static struct strbuf buf = STRBUF_INIT;
25 return odb_pack_name(&buf, sha1, "pack");
26}
27
28char *sha1_pack_index_name(const unsigned char *sha1)
29{
30 static struct strbuf buf = STRBUF_INIT;
31 return odb_pack_name(&buf, sha1, "idx");
32}
33
34static unsigned int pack_used_ctr;
35static unsigned int pack_mmap_calls;
36static unsigned int peak_pack_open_windows;
37static unsigned int pack_open_windows;
38static unsigned int pack_open_fds;
39static unsigned int pack_max_fds;
40static size_t peak_pack_mapped;
41static size_t pack_mapped;
42struct packed_git *packed_git;
43LIST_HEAD(packed_git_mru);
44
45#define SZ_FMT PRIuMAX
46static inline uintmax_t sz_fmt(size_t s) { return s; }
47
48void pack_report(void)
49{
50 fprintf(stderr,
51 "pack_report: getpagesize() = %10" SZ_FMT "\n"
52 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
53 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
54 sz_fmt(getpagesize()),
55 sz_fmt(packed_git_window_size),
56 sz_fmt(packed_git_limit));
57 fprintf(stderr,
58 "pack_report: pack_used_ctr = %10u\n"
59 "pack_report: pack_mmap_calls = %10u\n"
60 "pack_report: pack_open_windows = %10u / %10u\n"
61 "pack_report: pack_mapped = "
62 "%10" SZ_FMT " / %10" SZ_FMT "\n",
63 pack_used_ctr,
64 pack_mmap_calls,
65 pack_open_windows, peak_pack_open_windows,
66 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
67}
68
69/*
70 * Open and mmap the index file at path, perform a couple of
71 * consistency checks, then record its information to p. Return 0 on
72 * success.
73 */
74static int check_packed_git_idx(const char *path, struct packed_git *p)
75{
76 void *idx_map;
77 struct pack_idx_header *hdr;
78 size_t idx_size;
79 uint32_t version, nr, i, *index;
80 int fd = git_open(path);
81 struct stat st;
82
83 if (fd < 0)
84 return -1;
85 if (fstat(fd, &st)) {
86 close(fd);
87 return -1;
88 }
89 idx_size = xsize_t(st.st_size);
90 if (idx_size < 4 * 256 + 20 + 20) {
91 close(fd);
92 return error("index file %s is too small", path);
93 }
94 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
95 close(fd);
96
97 hdr = idx_map;
98 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
99 version = ntohl(hdr->idx_version);
100 if (version < 2 || version > 2) {
101 munmap(idx_map, idx_size);
102 return error("index file %s is version %"PRIu32
103 " and is not supported by this binary"
104 " (try upgrading GIT to a newer version)",
105 path, version);
106 }
107 } else
108 version = 1;
109
110 nr = 0;
111 index = idx_map;
112 if (version > 1)
113 index += 2; /* skip index header */
114 for (i = 0; i < 256; i++) {
115 uint32_t n = ntohl(index[i]);
116 if (n < nr) {
117 munmap(idx_map, idx_size);
118 return error("non-monotonic index %s", path);
119 }
120 nr = n;
121 }
122
123 if (version == 1) {
124 /*
125 * Total size:
126 * - 256 index entries 4 bytes each
127 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
128 * - 20-byte SHA1 of the packfile
129 * - 20-byte SHA1 file checksum
130 */
131 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
132 munmap(idx_map, idx_size);
133 return error("wrong index v1 file size in %s", path);
134 }
135 } else if (version == 2) {
136 /*
137 * Minimum size:
138 * - 8 bytes of header
139 * - 256 index entries 4 bytes each
140 * - 20-byte sha1 entry * nr
141 * - 4-byte crc entry * nr
142 * - 4-byte offset entry * nr
143 * - 20-byte SHA1 of the packfile
144 * - 20-byte SHA1 file checksum
145 * And after the 4-byte offset table might be a
146 * variable sized table containing 8-byte entries
147 * for offsets larger than 2^31.
148 */
149 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
150 unsigned long max_size = min_size;
151 if (nr)
152 max_size += (nr - 1)*8;
153 if (idx_size < min_size || idx_size > max_size) {
154 munmap(idx_map, idx_size);
155 return error("wrong index v2 file size in %s", path);
156 }
157 if (idx_size != min_size &&
158 /*
159 * make sure we can deal with large pack offsets.
160 * 31-bit signed offset won't be enough, neither
161 * 32-bit unsigned one will be.
162 */
163 (sizeof(off_t) <= 4)) {
164 munmap(idx_map, idx_size);
165 return error("pack too large for current definition of off_t in %s", path);
166 }
167 }
168
169 p->index_version = version;
170 p->index_data = idx_map;
171 p->index_size = idx_size;
172 p->num_objects = nr;
173 return 0;
174}
175
176int open_pack_index(struct packed_git *p)
177{
178 char *idx_name;
179 size_t len;
180 int ret;
181
182 if (p->index_data)
183 return 0;
184
185 if (!strip_suffix(p->pack_name, ".pack", &len))
186 die("BUG: pack_name does not end in .pack");
187 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
188 ret = check_packed_git_idx(idx_name, p);
189 free(idx_name);
190 return ret;
191}
192
193static struct packed_git *alloc_packed_git(int extra)
194{
195 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
196 memset(p, 0, sizeof(*p));
197 p->pack_fd = -1;
198 return p;
199}
200
201struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
202{
203 const char *path = sha1_pack_name(sha1);
204 size_t alloc = st_add(strlen(path), 1);
205 struct packed_git *p = alloc_packed_git(alloc);
206
207 memcpy(p->pack_name, path, alloc); /* includes NUL */
208 hashcpy(p->sha1, sha1);
209 if (check_packed_git_idx(idx_path, p)) {
210 free(p);
211 return NULL;
212 }
213
214 return p;
215}
216
217static void scan_windows(struct packed_git *p,
218 struct packed_git **lru_p,
219 struct pack_window **lru_w,
220 struct pack_window **lru_l)
221{
222 struct pack_window *w, *w_l;
223
224 for (w_l = NULL, w = p->windows; w; w = w->next) {
225 if (!w->inuse_cnt) {
226 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
227 *lru_p = p;
228 *lru_w = w;
229 *lru_l = w_l;
230 }
231 }
232 w_l = w;
233 }
234}
235
236static int unuse_one_window(struct packed_git *current)
237{
238 struct packed_git *p, *lru_p = NULL;
239 struct pack_window *lru_w = NULL, *lru_l = NULL;
240
241 if (current)
242 scan_windows(current, &lru_p, &lru_w, &lru_l);
243 for (p = packed_git; p; p = p->next)
244 scan_windows(p, &lru_p, &lru_w, &lru_l);
245 if (lru_p) {
246 munmap(lru_w->base, lru_w->len);
247 pack_mapped -= lru_w->len;
248 if (lru_l)
249 lru_l->next = lru_w->next;
250 else
251 lru_p->windows = lru_w->next;
252 free(lru_w);
253 pack_open_windows--;
254 return 1;
255 }
256 return 0;
257}
258
259void release_pack_memory(size_t need)
260{
261 size_t cur = pack_mapped;
262 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
263 ; /* nothing */
264}
265
266void close_pack_windows(struct packed_git *p)
267{
268 while (p->windows) {
269 struct pack_window *w = p->windows;
270
271 if (w->inuse_cnt)
272 die("pack '%s' still has open windows to it",
273 p->pack_name);
274 munmap(w->base, w->len);
275 pack_mapped -= w->len;
276 pack_open_windows--;
277 p->windows = w->next;
278 free(w);
279 }
280}
281
282static int close_pack_fd(struct packed_git *p)
283{
284 if (p->pack_fd < 0)
285 return 0;
286
287 close(p->pack_fd);
288 pack_open_fds--;
289 p->pack_fd = -1;
290
291 return 1;
292}
293
294void close_pack_index(struct packed_git *p)
295{
296 if (p->index_data) {
297 munmap((void *)p->index_data, p->index_size);
298 p->index_data = NULL;
299 }
300}
301
302static void close_pack(struct packed_git *p)
303{
304 close_pack_windows(p);
305 close_pack_fd(p);
306 close_pack_index(p);
307}
308
309void close_all_packs(void)
310{
311 struct packed_git *p;
312
313 for (p = packed_git; p; p = p->next)
314 if (p->do_not_close)
315 die("BUG: want to close pack marked 'do-not-close'");
316 else
317 close_pack(p);
318}
319
320/*
321 * The LRU pack is the one with the oldest MRU window, preferring packs
322 * with no used windows, or the oldest mtime if it has no windows allocated.
323 */
324static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
325{
326 struct pack_window *w, *this_mru_w;
327 int has_windows_inuse = 0;
328
329 /*
330 * Reject this pack if it has windows and the previously selected
331 * one does not. If this pack does not have windows, reject
332 * it if the pack file is newer than the previously selected one.
333 */
334 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
335 return;
336
337 for (w = this_mru_w = p->windows; w; w = w->next) {
338 /*
339 * Reject this pack if any of its windows are in use,
340 * but the previously selected pack did not have any
341 * inuse windows. Otherwise, record that this pack
342 * has windows in use.
343 */
344 if (w->inuse_cnt) {
345 if (*accept_windows_inuse)
346 has_windows_inuse = 1;
347 else
348 return;
349 }
350
351 if (w->last_used > this_mru_w->last_used)
352 this_mru_w = w;
353
354 /*
355 * Reject this pack if it has windows that have been
356 * used more recently than the previously selected pack.
357 * If the previously selected pack had windows inuse and
358 * we have not encountered a window in this pack that is
359 * inuse, skip this check since we prefer a pack with no
360 * inuse windows to one that has inuse windows.
361 */
362 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
363 this_mru_w->last_used > (*mru_w)->last_used)
364 return;
365 }
366
367 /*
368 * Select this pack.
369 */
370 *mru_w = this_mru_w;
371 *lru_p = p;
372 *accept_windows_inuse = has_windows_inuse;
373}
374
375static int close_one_pack(void)
376{
377 struct packed_git *p, *lru_p = NULL;
378 struct pack_window *mru_w = NULL;
379 int accept_windows_inuse = 1;
380
381 for (p = packed_git; p; p = p->next) {
382 if (p->pack_fd == -1)
383 continue;
384 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
385 }
386
387 if (lru_p)
388 return close_pack_fd(lru_p);
389
390 return 0;
391}
392
393static unsigned int get_max_fd_limit(void)
394{
395#ifdef RLIMIT_NOFILE
396 {
397 struct rlimit lim;
398
399 if (!getrlimit(RLIMIT_NOFILE, &lim))
400 return lim.rlim_cur;
401 }
402#endif
403
404#ifdef _SC_OPEN_MAX
405 {
406 long open_max = sysconf(_SC_OPEN_MAX);
407 if (0 < open_max)
408 return open_max;
409 /*
410 * Otherwise, we got -1 for one of the two
411 * reasons:
412 *
413 * (1) sysconf() did not understand _SC_OPEN_MAX
414 * and signaled an error with -1; or
415 * (2) sysconf() said there is no limit.
416 *
417 * We _could_ clear errno before calling sysconf() to
418 * tell these two cases apart and return a huge number
419 * in the latter case to let the caller cap it to a
420 * value that is not so selfish, but letting the
421 * fallback OPEN_MAX codepath take care of these cases
422 * is a lot simpler.
423 */
424 }
425#endif
426
427#ifdef OPEN_MAX
428 return OPEN_MAX;
429#else
430 return 1; /* see the caller ;-) */
431#endif
432}
433
434/*
435 * Do not call this directly as this leaks p->pack_fd on error return;
436 * call open_packed_git() instead.
437 */
438static int open_packed_git_1(struct packed_git *p)
439{
440 struct stat st;
441 struct pack_header hdr;
442 unsigned char sha1[20];
443 unsigned char *idx_sha1;
444 long fd_flag;
445
446 if (!p->index_data && open_pack_index(p))
447 return error("packfile %s index unavailable", p->pack_name);
448
449 if (!pack_max_fds) {
450 unsigned int max_fds = get_max_fd_limit();
451
452 /* Save 3 for stdin/stdout/stderr, 22 for work */
453 if (25 < max_fds)
454 pack_max_fds = max_fds - 25;
455 else
456 pack_max_fds = 1;
457 }
458
459 while (pack_max_fds <= pack_open_fds && close_one_pack())
460 ; /* nothing */
461
462 p->pack_fd = git_open(p->pack_name);
463 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
464 return -1;
465 pack_open_fds++;
466
467 /* If we created the struct before we had the pack we lack size. */
468 if (!p->pack_size) {
469 if (!S_ISREG(st.st_mode))
470 return error("packfile %s not a regular file", p->pack_name);
471 p->pack_size = st.st_size;
472 } else if (p->pack_size != st.st_size)
473 return error("packfile %s size changed", p->pack_name);
474
475 /* We leave these file descriptors open with sliding mmap;
476 * there is no point keeping them open across exec(), though.
477 */
478 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
479 if (fd_flag < 0)
480 return error("cannot determine file descriptor flags");
481 fd_flag |= FD_CLOEXEC;
482 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
483 return error("cannot set FD_CLOEXEC");
484
485 /* Verify we recognize this pack file format. */
486 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
487 return error("file %s is far too short to be a packfile", p->pack_name);
488 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
489 return error("file %s is not a GIT packfile", p->pack_name);
490 if (!pack_version_ok(hdr.hdr_version))
491 return error("packfile %s is version %"PRIu32" and not"
492 " supported (try upgrading GIT to a newer version)",
493 p->pack_name, ntohl(hdr.hdr_version));
494
495 /* Verify the pack matches its index. */
496 if (p->num_objects != ntohl(hdr.hdr_entries))
497 return error("packfile %s claims to have %"PRIu32" objects"
498 " while index indicates %"PRIu32" objects",
499 p->pack_name, ntohl(hdr.hdr_entries),
500 p->num_objects);
501 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
502 return error("end of packfile %s is unavailable", p->pack_name);
503 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
504 return error("packfile %s signature is unavailable", p->pack_name);
505 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
506 if (hashcmp(sha1, idx_sha1))
507 return error("packfile %s does not match index", p->pack_name);
508 return 0;
509}
510
511static int open_packed_git(struct packed_git *p)
512{
513 if (!open_packed_git_1(p))
514 return 0;
515 close_pack_fd(p);
516 return -1;
517}
518
519static int in_window(struct pack_window *win, off_t offset)
520{
521 /* We must promise at least 20 bytes (one hash) after the
522 * offset is available from this window, otherwise the offset
523 * is not actually in this window and a different window (which
524 * has that one hash excess) must be used. This is to support
525 * the object header and delta base parsing routines below.
526 */
527 off_t win_off = win->offset;
528 return win_off <= offset
529 && (offset + 20) <= (win_off + win->len);
530}
531
532unsigned char *use_pack(struct packed_git *p,
533 struct pack_window **w_cursor,
534 off_t offset,
535 unsigned long *left)
536{
537 struct pack_window *win = *w_cursor;
538
539 /* Since packfiles end in a hash of their content and it's
540 * pointless to ask for an offset into the middle of that
541 * hash, and the in_window function above wouldn't match
542 * don't allow an offset too close to the end of the file.
543 */
544 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
545 die("packfile %s cannot be accessed", p->pack_name);
546 if (offset > (p->pack_size - 20))
547 die("offset beyond end of packfile (truncated pack?)");
548 if (offset < 0)
549 die(_("offset before end of packfile (broken .idx?)"));
550
551 if (!win || !in_window(win, offset)) {
552 if (win)
553 win->inuse_cnt--;
554 for (win = p->windows; win; win = win->next) {
555 if (in_window(win, offset))
556 break;
557 }
558 if (!win) {
559 size_t window_align = packed_git_window_size / 2;
560 off_t len;
561
562 if (p->pack_fd == -1 && open_packed_git(p))
563 die("packfile %s cannot be accessed", p->pack_name);
564
565 win = xcalloc(1, sizeof(*win));
566 win->offset = (offset / window_align) * window_align;
567 len = p->pack_size - win->offset;
568 if (len > packed_git_window_size)
569 len = packed_git_window_size;
570 win->len = (size_t)len;
571 pack_mapped += win->len;
572 while (packed_git_limit < pack_mapped
573 && unuse_one_window(p))
574 ; /* nothing */
575 win->base = xmmap(NULL, win->len,
576 PROT_READ, MAP_PRIVATE,
577 p->pack_fd, win->offset);
578 if (win->base == MAP_FAILED)
579 die_errno("packfile %s cannot be mapped",
580 p->pack_name);
581 if (!win->offset && win->len == p->pack_size
582 && !p->do_not_close)
583 close_pack_fd(p);
584 pack_mmap_calls++;
585 pack_open_windows++;
586 if (pack_mapped > peak_pack_mapped)
587 peak_pack_mapped = pack_mapped;
588 if (pack_open_windows > peak_pack_open_windows)
589 peak_pack_open_windows = pack_open_windows;
590 win->next = p->windows;
591 p->windows = win;
592 }
593 }
594 if (win != *w_cursor) {
595 win->last_used = pack_used_ctr++;
596 win->inuse_cnt++;
597 *w_cursor = win;
598 }
599 offset -= win->offset;
600 if (left)
601 *left = win->len - xsize_t(offset);
602 return win->base + offset;
603}
604
605void unuse_pack(struct pack_window **w_cursor)
606{
607 struct pack_window *w = *w_cursor;
608 if (w) {
609 w->inuse_cnt--;
610 *w_cursor = NULL;
611 }
612}
613
614static void try_to_free_pack_memory(size_t size)
615{
616 release_pack_memory(size);
617}
618
619struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
620{
621 static int have_set_try_to_free_routine;
622 struct stat st;
623 size_t alloc;
624 struct packed_git *p;
625
626 if (!have_set_try_to_free_routine) {
627 have_set_try_to_free_routine = 1;
628 set_try_to_free_routine(try_to_free_pack_memory);
629 }
630
631 /*
632 * Make sure a corresponding .pack file exists and that
633 * the index looks sane.
634 */
635 if (!strip_suffix_mem(path, &path_len, ".idx"))
636 return NULL;
637
638 /*
639 * ".pack" is long enough to hold any suffix we're adding (and
640 * the use xsnprintf double-checks that)
641 */
642 alloc = st_add3(path_len, strlen(".pack"), 1);
643 p = alloc_packed_git(alloc);
644 memcpy(p->pack_name, path, path_len);
645
646 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
647 if (!access(p->pack_name, F_OK))
648 p->pack_keep = 1;
649
650 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
651 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
652 free(p);
653 return NULL;
654 }
655
656 /* ok, it looks sane as far as we can check without
657 * actually mapping the pack file.
658 */
659 p->pack_size = st.st_size;
660 p->pack_local = local;
661 p->mtime = st.st_mtime;
662 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
663 hashclr(p->sha1);
664 return p;
665}
666
667void install_packed_git(struct packed_git *pack)
668{
669 if (pack->pack_fd != -1)
670 pack_open_fds++;
671
672 pack->next = packed_git;
673 packed_git = pack;
674}
675
676void (*report_garbage)(unsigned seen_bits, const char *path);
677
678static void report_helper(const struct string_list *list,
679 int seen_bits, int first, int last)
680{
681 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
682 return;
683
684 for (; first < last; first++)
685 report_garbage(seen_bits, list->items[first].string);
686}
687
688static void report_pack_garbage(struct string_list *list)
689{
690 int i, baselen = -1, first = 0, seen_bits = 0;
691
692 if (!report_garbage)
693 return;
694
695 string_list_sort(list);
696
697 for (i = 0; i < list->nr; i++) {
698 const char *path = list->items[i].string;
699 if (baselen != -1 &&
700 strncmp(path, list->items[first].string, baselen)) {
701 report_helper(list, seen_bits, first, i);
702 baselen = -1;
703 seen_bits = 0;
704 }
705 if (baselen == -1) {
706 const char *dot = strrchr(path, '.');
707 if (!dot) {
708 report_garbage(PACKDIR_FILE_GARBAGE, path);
709 continue;
710 }
711 baselen = dot - path + 1;
712 first = i;
713 }
714 if (!strcmp(path + baselen, "pack"))
715 seen_bits |= 1;
716 else if (!strcmp(path + baselen, "idx"))
717 seen_bits |= 2;
718 }
719 report_helper(list, seen_bits, first, list->nr);
720}
721
722static void prepare_packed_git_one(char *objdir, int local)
723{
724 struct strbuf path = STRBUF_INIT;
725 size_t dirnamelen;
726 DIR *dir;
727 struct dirent *de;
728 struct string_list garbage = STRING_LIST_INIT_DUP;
729
730 strbuf_addstr(&path, objdir);
731 strbuf_addstr(&path, "/pack");
732 dir = opendir(path.buf);
733 if (!dir) {
734 if (errno != ENOENT)
735 error_errno("unable to open object pack directory: %s",
736 path.buf);
737 strbuf_release(&path);
738 return;
739 }
740 strbuf_addch(&path, '/');
741 dirnamelen = path.len;
742 while ((de = readdir(dir)) != NULL) {
743 struct packed_git *p;
744 size_t base_len;
745
746 if (is_dot_or_dotdot(de->d_name))
747 continue;
748
749 strbuf_setlen(&path, dirnamelen);
750 strbuf_addstr(&path, de->d_name);
751
752 base_len = path.len;
753 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
754 /* Don't reopen a pack we already have. */
755 for (p = packed_git; p; p = p->next) {
756 size_t len;
757 if (strip_suffix(p->pack_name, ".pack", &len) &&
758 len == base_len &&
759 !memcmp(p->pack_name, path.buf, len))
760 break;
761 }
762 if (p == NULL &&
763 /*
764 * See if it really is a valid .idx file with
765 * corresponding .pack file that we can map.
766 */
767 (p = add_packed_git(path.buf, path.len, local)) != NULL)
768 install_packed_git(p);
769 }
770
771 if (!report_garbage)
772 continue;
773
774 if (ends_with(de->d_name, ".idx") ||
775 ends_with(de->d_name, ".pack") ||
776 ends_with(de->d_name, ".bitmap") ||
777 ends_with(de->d_name, ".keep"))
778 string_list_append(&garbage, path.buf);
779 else
780 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
781 }
782 closedir(dir);
783 report_pack_garbage(&garbage);
784 string_list_clear(&garbage, 0);
785 strbuf_release(&path);
786}
787
788static int approximate_object_count_valid;
789
790/*
791 * Give a fast, rough count of the number of objects in the repository. This
792 * ignores loose objects completely. If you have a lot of them, then either
793 * you should repack because your performance will be awful, or they are
794 * all unreachable objects about to be pruned, in which case they're not really
795 * interesting as a measure of repo size in the first place.
796 */
797unsigned long approximate_object_count(void)
798{
799 static unsigned long count;
800 if (!approximate_object_count_valid) {
801 struct packed_git *p;
802
803 prepare_packed_git();
804 count = 0;
805 for (p = packed_git; p; p = p->next) {
806 if (open_pack_index(p))
807 continue;
808 count += p->num_objects;
809 }
810 }
811 return count;
812}
813
814static void *get_next_packed_git(const void *p)
815{
816 return ((const struct packed_git *)p)->next;
817}
818
819static void set_next_packed_git(void *p, void *next)
820{
821 ((struct packed_git *)p)->next = next;
822}
823
824static int sort_pack(const void *a_, const void *b_)
825{
826 const struct packed_git *a = a_;
827 const struct packed_git *b = b_;
828 int st;
829
830 /*
831 * Local packs tend to contain objects specific to our
832 * variant of the project than remote ones. In addition,
833 * remote ones could be on a network mounted filesystem.
834 * Favor local ones for these reasons.
835 */
836 st = a->pack_local - b->pack_local;
837 if (st)
838 return -st;
839
840 /*
841 * Younger packs tend to contain more recent objects,
842 * and more recent objects tend to get accessed more
843 * often.
844 */
845 if (a->mtime < b->mtime)
846 return 1;
847 else if (a->mtime == b->mtime)
848 return 0;
849 return -1;
850}
851
852static void rearrange_packed_git(void)
853{
854 packed_git = llist_mergesort(packed_git, get_next_packed_git,
855 set_next_packed_git, sort_pack);
856}
857
858static void prepare_packed_git_mru(void)
859{
860 struct packed_git *p;
861
862 INIT_LIST_HEAD(&packed_git_mru);
863
864 for (p = packed_git; p; p = p->next)
865 list_add_tail(&p->mru, &packed_git_mru);
866}
867
868static int prepare_packed_git_run_once = 0;
869void prepare_packed_git(void)
870{
871 struct alternate_object_database *alt;
872
873 if (prepare_packed_git_run_once)
874 return;
875 prepare_packed_git_one(get_object_directory(), 1);
876 prepare_alt_odb();
877 for (alt = alt_odb_list; alt; alt = alt->next)
878 prepare_packed_git_one(alt->path, 0);
879 rearrange_packed_git();
880 prepare_packed_git_mru();
881 prepare_packed_git_run_once = 1;
882}
883
884void reprepare_packed_git(void)
885{
886 approximate_object_count_valid = 0;
887 prepare_packed_git_run_once = 0;
888 prepare_packed_git();
889}
890
891unsigned long unpack_object_header_buffer(const unsigned char *buf,
892 unsigned long len, enum object_type *type, unsigned long *sizep)
893{
894 unsigned shift;
895 unsigned long size, c;
896 unsigned long used = 0;
897
898 c = buf[used++];
899 *type = (c >> 4) & 7;
900 size = c & 15;
901 shift = 4;
902 while (c & 0x80) {
903 if (len <= used || bitsizeof(long) <= shift) {
904 error("bad object header");
905 size = used = 0;
906 break;
907 }
908 c = buf[used++];
909 size += (c & 0x7f) << shift;
910 shift += 7;
911 }
912 *sizep = size;
913 return used;
914}
915
916unsigned long get_size_from_delta(struct packed_git *p,
917 struct pack_window **w_curs,
918 off_t curpos)
919{
920 const unsigned char *data;
921 unsigned char delta_head[20], *in;
922 git_zstream stream;
923 int st;
924
925 memset(&stream, 0, sizeof(stream));
926 stream.next_out = delta_head;
927 stream.avail_out = sizeof(delta_head);
928
929 git_inflate_init(&stream);
930 do {
931 in = use_pack(p, w_curs, curpos, &stream.avail_in);
932 stream.next_in = in;
933 st = git_inflate(&stream, Z_FINISH);
934 curpos += stream.next_in - in;
935 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
936 stream.total_out < sizeof(delta_head));
937 git_inflate_end(&stream);
938 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
939 error("delta data unpack-initial failed");
940 return 0;
941 }
942
943 /* Examine the initial part of the delta to figure out
944 * the result size.
945 */
946 data = delta_head;
947
948 /* ignore base size */
949 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
950
951 /* Read the result size */
952 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
953}
954
955int unpack_object_header(struct packed_git *p,
956 struct pack_window **w_curs,
957 off_t *curpos,
958 unsigned long *sizep)
959{
960 unsigned char *base;
961 unsigned long left;
962 unsigned long used;
963 enum object_type type;
964
965 /* use_pack() assures us we have [base, base + 20) available
966 * as a range that we can look at. (Its actually the hash
967 * size that is assured.) With our object header encoding
968 * the maximum deflated object size is 2^137, which is just
969 * insane, so we know won't exceed what we have been given.
970 */
971 base = use_pack(p, w_curs, *curpos, &left);
972 used = unpack_object_header_buffer(base, left, &type, sizep);
973 if (!used) {
974 type = OBJ_BAD;
975 } else
976 *curpos += used;
977
978 return type;
979}
980
981void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
982{
983 unsigned i;
984 for (i = 0; i < p->num_bad_objects; i++)
985 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
986 return;
987 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
988 st_mult(GIT_MAX_RAWSZ,
989 st_add(p->num_bad_objects, 1)));
990 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
991 p->num_bad_objects++;
992}
993
994const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
995{
996 struct packed_git *p;
997 unsigned i;
998
999 for (p = packed_git; p; p = p->next)
1000 for (i = 0; i < p->num_bad_objects; i++)
1001 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1002 return p;
1003 return NULL;
1004}
1005
1006static off_t get_delta_base(struct packed_git *p,
1007 struct pack_window **w_curs,
1008 off_t *curpos,
1009 enum object_type type,
1010 off_t delta_obj_offset)
1011{
1012 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1013 off_t base_offset;
1014
1015 /* use_pack() assured us we have [base_info, base_info + 20)
1016 * as a range that we can look at without walking off the
1017 * end of the mapped window. Its actually the hash size
1018 * that is assured. An OFS_DELTA longer than the hash size
1019 * is stupid, as then a REF_DELTA would be smaller to store.
1020 */
1021 if (type == OBJ_OFS_DELTA) {
1022 unsigned used = 0;
1023 unsigned char c = base_info[used++];
1024 base_offset = c & 127;
1025 while (c & 128) {
1026 base_offset += 1;
1027 if (!base_offset || MSB(base_offset, 7))
1028 return 0; /* overflow */
1029 c = base_info[used++];
1030 base_offset = (base_offset << 7) + (c & 127);
1031 }
1032 base_offset = delta_obj_offset - base_offset;
1033 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1034 return 0; /* out of bound */
1035 *curpos += used;
1036 } else if (type == OBJ_REF_DELTA) {
1037 /* The base entry _must_ be in the same pack */
1038 base_offset = find_pack_entry_one(base_info, p);
1039 *curpos += 20;
1040 } else
1041 die("I am totally screwed");
1042 return base_offset;
1043}
1044
1045/*
1046 * Like get_delta_base above, but we return the sha1 instead of the pack
1047 * offset. This means it is cheaper for REF deltas (we do not have to do
1048 * the final object lookup), but more expensive for OFS deltas (we
1049 * have to load the revidx to convert the offset back into a sha1).
1050 */
1051static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1052 struct pack_window **w_curs,
1053 off_t curpos,
1054 enum object_type type,
1055 off_t delta_obj_offset)
1056{
1057 if (type == OBJ_REF_DELTA) {
1058 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1059 return base;
1060 } else if (type == OBJ_OFS_DELTA) {
1061 struct revindex_entry *revidx;
1062 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1063 type, delta_obj_offset);
1064
1065 if (!base_offset)
1066 return NULL;
1067
1068 revidx = find_pack_revindex(p, base_offset);
1069 if (!revidx)
1070 return NULL;
1071
1072 return nth_packed_object_sha1(p, revidx->nr);
1073 } else
1074 return NULL;
1075}
1076
1077static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1078{
1079 int type;
1080 struct revindex_entry *revidx;
1081 const unsigned char *sha1;
1082 revidx = find_pack_revindex(p, obj_offset);
1083 if (!revidx)
1084 return OBJ_BAD;
1085 sha1 = nth_packed_object_sha1(p, revidx->nr);
1086 mark_bad_packed_object(p, sha1);
1087 type = sha1_object_info(sha1, NULL);
1088 if (type <= OBJ_NONE)
1089 return OBJ_BAD;
1090 return type;
1091}
1092
1093#define POI_STACK_PREALLOC 64
1094
1095static enum object_type packed_to_object_type(struct packed_git *p,
1096 off_t obj_offset,
1097 enum object_type type,
1098 struct pack_window **w_curs,
1099 off_t curpos)
1100{
1101 off_t small_poi_stack[POI_STACK_PREALLOC];
1102 off_t *poi_stack = small_poi_stack;
1103 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1104
1105 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1106 off_t base_offset;
1107 unsigned long size;
1108 /* Push the object we're going to leave behind */
1109 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1110 poi_stack_alloc = alloc_nr(poi_stack_nr);
1111 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1112 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1113 } else {
1114 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1115 }
1116 poi_stack[poi_stack_nr++] = obj_offset;
1117 /* If parsing the base offset fails, just unwind */
1118 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1119 if (!base_offset)
1120 goto unwind;
1121 curpos = obj_offset = base_offset;
1122 type = unpack_object_header(p, w_curs, &curpos, &size);
1123 if (type <= OBJ_NONE) {
1124 /* If getting the base itself fails, we first
1125 * retry the base, otherwise unwind */
1126 type = retry_bad_packed_offset(p, base_offset);
1127 if (type > OBJ_NONE)
1128 goto out;
1129 goto unwind;
1130 }
1131 }
1132
1133 switch (type) {
1134 case OBJ_BAD:
1135 case OBJ_COMMIT:
1136 case OBJ_TREE:
1137 case OBJ_BLOB:
1138 case OBJ_TAG:
1139 break;
1140 default:
1141 error("unknown object type %i at offset %"PRIuMAX" in %s",
1142 type, (uintmax_t)obj_offset, p->pack_name);
1143 type = OBJ_BAD;
1144 }
1145
1146out:
1147 if (poi_stack != small_poi_stack)
1148 free(poi_stack);
1149 return type;
1150
1151unwind:
1152 while (poi_stack_nr) {
1153 obj_offset = poi_stack[--poi_stack_nr];
1154 type = retry_bad_packed_offset(p, obj_offset);
1155 if (type > OBJ_NONE)
1156 goto out;
1157 }
1158 type = OBJ_BAD;
1159 goto out;
1160}
1161
1162static struct hashmap delta_base_cache;
1163static size_t delta_base_cached;
1164
1165static LIST_HEAD(delta_base_cache_lru);
1166
1167struct delta_base_cache_key {
1168 struct packed_git *p;
1169 off_t base_offset;
1170};
1171
1172struct delta_base_cache_entry {
1173 struct hashmap hash;
1174 struct delta_base_cache_key key;
1175 struct list_head lru;
1176 void *data;
1177 unsigned long size;
1178 enum object_type type;
1179};
1180
1181static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1182{
1183 unsigned int hash;
1184
1185 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1186 hash += (hash >> 8) + (hash >> 16);
1187 return hash;
1188}
1189
1190static struct delta_base_cache_entry *
1191get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1192{
1193 struct hashmap_entry entry;
1194 struct delta_base_cache_key key;
1195
1196 if (!delta_base_cache.cmpfn)
1197 return NULL;
1198
1199 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1200 key.p = p;
1201 key.base_offset = base_offset;
1202 return hashmap_get(&delta_base_cache, &entry, &key);
1203}
1204
1205static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1206 const struct delta_base_cache_key *b)
1207{
1208 return a->p == b->p && a->base_offset == b->base_offset;
1209}
1210
1211static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1212 const void *va, const void *vb,
1213 const void *vkey)
1214{
1215 const struct delta_base_cache_entry *a = va, *b = vb;
1216 const struct delta_base_cache_key *key = vkey;
1217 if (key)
1218 return !delta_base_cache_key_eq(&a->key, key);
1219 else
1220 return !delta_base_cache_key_eq(&a->key, &b->key);
1221}
1222
1223static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1224{
1225 return !!get_delta_base_cache_entry(p, base_offset);
1226}
1227
1228/*
1229 * Remove the entry from the cache, but do _not_ free the associated
1230 * entry data. The caller takes ownership of the "data" buffer, and
1231 * should copy out any fields it wants before detaching.
1232 */
1233static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1234{
1235 hashmap_remove(&delta_base_cache, ent, &ent->key);
1236 list_del(&ent->lru);
1237 delta_base_cached -= ent->size;
1238 free(ent);
1239}
1240
1241static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1242 unsigned long *base_size, enum object_type *type)
1243{
1244 struct delta_base_cache_entry *ent;
1245
1246 ent = get_delta_base_cache_entry(p, base_offset);
1247 if (!ent)
1248 return unpack_entry(p, base_offset, type, base_size);
1249
1250 if (type)
1251 *type = ent->type;
1252 if (base_size)
1253 *base_size = ent->size;
1254 return xmemdupz(ent->data, ent->size);
1255}
1256
1257static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1258{
1259 free(ent->data);
1260 detach_delta_base_cache_entry(ent);
1261}
1262
1263void clear_delta_base_cache(void)
1264{
1265 struct list_head *lru, *tmp;
1266 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1267 struct delta_base_cache_entry *entry =
1268 list_entry(lru, struct delta_base_cache_entry, lru);
1269 release_delta_base_cache(entry);
1270 }
1271}
1272
1273static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1274 void *base, unsigned long base_size, enum object_type type)
1275{
1276 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1277 struct list_head *lru, *tmp;
1278
1279 delta_base_cached += base_size;
1280
1281 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1282 struct delta_base_cache_entry *f =
1283 list_entry(lru, struct delta_base_cache_entry, lru);
1284 if (delta_base_cached <= delta_base_cache_limit)
1285 break;
1286 release_delta_base_cache(f);
1287 }
1288
1289 ent->key.p = p;
1290 ent->key.base_offset = base_offset;
1291 ent->type = type;
1292 ent->data = base;
1293 ent->size = base_size;
1294 list_add_tail(&ent->lru, &delta_base_cache_lru);
1295
1296 if (!delta_base_cache.cmpfn)
1297 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1298 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1299 hashmap_add(&delta_base_cache, ent);
1300}
1301
1302int packed_object_info(struct packed_git *p, off_t obj_offset,
1303 struct object_info *oi)
1304{
1305 struct pack_window *w_curs = NULL;
1306 unsigned long size;
1307 off_t curpos = obj_offset;
1308 enum object_type type;
1309
1310 /*
1311 * We always get the representation type, but only convert it to
1312 * a "real" type later if the caller is interested.
1313 */
1314 if (oi->contentp) {
1315 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1316 &type);
1317 if (!*oi->contentp)
1318 type = OBJ_BAD;
1319 } else {
1320 type = unpack_object_header(p, &w_curs, &curpos, &size);
1321 }
1322
1323 if (!oi->contentp && oi->sizep) {
1324 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1325 off_t tmp_pos = curpos;
1326 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1327 type, obj_offset);
1328 if (!base_offset) {
1329 type = OBJ_BAD;
1330 goto out;
1331 }
1332 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1333 if (*oi->sizep == 0) {
1334 type = OBJ_BAD;
1335 goto out;
1336 }
1337 } else {
1338 *oi->sizep = size;
1339 }
1340 }
1341
1342 if (oi->disk_sizep) {
1343 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1344 *oi->disk_sizep = revidx[1].offset - obj_offset;
1345 }
1346
1347 if (oi->typep || oi->typename) {
1348 enum object_type ptot;
1349 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1350 curpos);
1351 if (oi->typep)
1352 *oi->typep = ptot;
1353 if (oi->typename) {
1354 const char *tn = typename(ptot);
1355 if (tn)
1356 strbuf_addstr(oi->typename, tn);
1357 }
1358 if (ptot < 0) {
1359 type = OBJ_BAD;
1360 goto out;
1361 }
1362 }
1363
1364 if (oi->delta_base_sha1) {
1365 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1366 const unsigned char *base;
1367
1368 base = get_delta_base_sha1(p, &w_curs, curpos,
1369 type, obj_offset);
1370 if (!base) {
1371 type = OBJ_BAD;
1372 goto out;
1373 }
1374
1375 hashcpy(oi->delta_base_sha1, base);
1376 } else
1377 hashclr(oi->delta_base_sha1);
1378 }
1379
1380 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1381 OI_PACKED;
1382
1383out:
1384 unuse_pack(&w_curs);
1385 return type;
1386}
1387
1388static void *unpack_compressed_entry(struct packed_git *p,
1389 struct pack_window **w_curs,
1390 off_t curpos,
1391 unsigned long size)
1392{
1393 int st;
1394 git_zstream stream;
1395 unsigned char *buffer, *in;
1396
1397 buffer = xmallocz_gently(size);
1398 if (!buffer)
1399 return NULL;
1400 memset(&stream, 0, sizeof(stream));
1401 stream.next_out = buffer;
1402 stream.avail_out = size + 1;
1403
1404 git_inflate_init(&stream);
1405 do {
1406 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1407 stream.next_in = in;
1408 st = git_inflate(&stream, Z_FINISH);
1409 if (!stream.avail_out)
1410 break; /* the payload is larger than it should be */
1411 curpos += stream.next_in - in;
1412 } while (st == Z_OK || st == Z_BUF_ERROR);
1413 git_inflate_end(&stream);
1414 if ((st != Z_STREAM_END) || stream.total_out != size) {
1415 free(buffer);
1416 return NULL;
1417 }
1418
1419 return buffer;
1420}
1421
1422static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1423{
1424 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1425 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1426 p->pack_name, (uintmax_t)obj_offset);
1427}
1428
1429int do_check_packed_object_crc;
1430
1431#define UNPACK_ENTRY_STACK_PREALLOC 64
1432struct unpack_entry_stack_ent {
1433 off_t obj_offset;
1434 off_t curpos;
1435 unsigned long size;
1436};
1437
1438static void *read_object(const unsigned char *sha1, enum object_type *type,
1439 unsigned long *size)
1440{
1441 struct object_info oi = OBJECT_INFO_INIT;
1442 void *content;
1443 oi.typep = type;
1444 oi.sizep = size;
1445 oi.contentp = &content;
1446
1447 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1448 return NULL;
1449 return content;
1450}
1451
1452void *unpack_entry(struct packed_git *p, off_t obj_offset,
1453 enum object_type *final_type, unsigned long *final_size)
1454{
1455 struct pack_window *w_curs = NULL;
1456 off_t curpos = obj_offset;
1457 void *data = NULL;
1458 unsigned long size;
1459 enum object_type type;
1460 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1461 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1462 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1463 int base_from_cache = 0;
1464
1465 write_pack_access_log(p, obj_offset);
1466
1467 /* PHASE 1: drill down to the innermost base object */
1468 for (;;) {
1469 off_t base_offset;
1470 int i;
1471 struct delta_base_cache_entry *ent;
1472
1473 ent = get_delta_base_cache_entry(p, curpos);
1474 if (ent) {
1475 type = ent->type;
1476 data = ent->data;
1477 size = ent->size;
1478 detach_delta_base_cache_entry(ent);
1479 base_from_cache = 1;
1480 break;
1481 }
1482
1483 if (do_check_packed_object_crc && p->index_version > 1) {
1484 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1485 off_t len = revidx[1].offset - obj_offset;
1486 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1487 const unsigned char *sha1 =
1488 nth_packed_object_sha1(p, revidx->nr);
1489 error("bad packed object CRC for %s",
1490 sha1_to_hex(sha1));
1491 mark_bad_packed_object(p, sha1);
1492 data = NULL;
1493 goto out;
1494 }
1495 }
1496
1497 type = unpack_object_header(p, &w_curs, &curpos, &size);
1498 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1499 break;
1500
1501 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1502 if (!base_offset) {
1503 error("failed to validate delta base reference "
1504 "at offset %"PRIuMAX" from %s",
1505 (uintmax_t)curpos, p->pack_name);
1506 /* bail to phase 2, in hopes of recovery */
1507 data = NULL;
1508 break;
1509 }
1510
1511 /* push object, proceed to base */
1512 if (delta_stack_nr >= delta_stack_alloc
1513 && delta_stack == small_delta_stack) {
1514 delta_stack_alloc = alloc_nr(delta_stack_nr);
1515 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1516 memcpy(delta_stack, small_delta_stack,
1517 sizeof(*delta_stack)*delta_stack_nr);
1518 } else {
1519 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1520 }
1521 i = delta_stack_nr++;
1522 delta_stack[i].obj_offset = obj_offset;
1523 delta_stack[i].curpos = curpos;
1524 delta_stack[i].size = size;
1525
1526 curpos = obj_offset = base_offset;
1527 }
1528
1529 /* PHASE 2: handle the base */
1530 switch (type) {
1531 case OBJ_OFS_DELTA:
1532 case OBJ_REF_DELTA:
1533 if (data)
1534 die("BUG: unpack_entry: left loop at a valid delta");
1535 break;
1536 case OBJ_COMMIT:
1537 case OBJ_TREE:
1538 case OBJ_BLOB:
1539 case OBJ_TAG:
1540 if (!base_from_cache)
1541 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1542 break;
1543 default:
1544 data = NULL;
1545 error("unknown object type %i at offset %"PRIuMAX" in %s",
1546 type, (uintmax_t)obj_offset, p->pack_name);
1547 }
1548
1549 /* PHASE 3: apply deltas in order */
1550
1551 /* invariants:
1552 * 'data' holds the base data, or NULL if there was corruption
1553 */
1554 while (delta_stack_nr) {
1555 void *delta_data;
1556 void *base = data;
1557 void *external_base = NULL;
1558 unsigned long delta_size, base_size = size;
1559 int i;
1560
1561 data = NULL;
1562
1563 if (base)
1564 add_delta_base_cache(p, obj_offset, base, base_size, type);
1565
1566 if (!base) {
1567 /*
1568 * We're probably in deep shit, but let's try to fetch
1569 * the required base anyway from another pack or loose.
1570 * This is costly but should happen only in the presence
1571 * of a corrupted pack, and is better than failing outright.
1572 */
1573 struct revindex_entry *revidx;
1574 const unsigned char *base_sha1;
1575 revidx = find_pack_revindex(p, obj_offset);
1576 if (revidx) {
1577 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1578 error("failed to read delta base object %s"
1579 " at offset %"PRIuMAX" from %s",
1580 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
1581 p->pack_name);
1582 mark_bad_packed_object(p, base_sha1);
1583 base = read_object(base_sha1, &type, &base_size);
1584 external_base = base;
1585 }
1586 }
1587
1588 i = --delta_stack_nr;
1589 obj_offset = delta_stack[i].obj_offset;
1590 curpos = delta_stack[i].curpos;
1591 delta_size = delta_stack[i].size;
1592
1593 if (!base)
1594 continue;
1595
1596 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1597
1598 if (!delta_data) {
1599 error("failed to unpack compressed delta "
1600 "at offset %"PRIuMAX" from %s",
1601 (uintmax_t)curpos, p->pack_name);
1602 data = NULL;
1603 free(external_base);
1604 continue;
1605 }
1606
1607 data = patch_delta(base, base_size,
1608 delta_data, delta_size,
1609 &size);
1610
1611 /*
1612 * We could not apply the delta; warn the user, but keep going.
1613 * Our failure will be noticed either in the next iteration of
1614 * the loop, or if this is the final delta, in the caller when
1615 * we return NULL. Those code paths will take care of making
1616 * a more explicit warning and retrying with another copy of
1617 * the object.
1618 */
1619 if (!data)
1620 error("failed to apply delta");
1621
1622 free(delta_data);
1623 free(external_base);
1624 }
1625
1626 if (final_type)
1627 *final_type = type;
1628 if (final_size)
1629 *final_size = size;
1630
1631out:
1632 unuse_pack(&w_curs);
1633
1634 if (delta_stack != small_delta_stack)
1635 free(delta_stack);
1636
1637 return data;
1638}
1639
1640const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1641 uint32_t n)
1642{
1643 const unsigned char *index = p->index_data;
1644 if (!index) {
1645 if (open_pack_index(p))
1646 return NULL;
1647 index = p->index_data;
1648 }
1649 if (n >= p->num_objects)
1650 return NULL;
1651 index += 4 * 256;
1652 if (p->index_version == 1) {
1653 return index + 24 * n + 4;
1654 } else {
1655 index += 8;
1656 return index + 20 * n;
1657 }
1658}
1659
1660const struct object_id *nth_packed_object_oid(struct object_id *oid,
1661 struct packed_git *p,
1662 uint32_t n)
1663{
1664 const unsigned char *hash = nth_packed_object_sha1(p, n);
1665 if (!hash)
1666 return NULL;
1667 hashcpy(oid->hash, hash);
1668 return oid;
1669}
1670
1671void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1672{
1673 const unsigned char *ptr = vptr;
1674 const unsigned char *start = p->index_data;
1675 const unsigned char *end = start + p->index_size;
1676 if (ptr < start)
1677 die(_("offset before start of pack index for %s (corrupt index?)"),
1678 p->pack_name);
1679 /* No need to check for underflow; .idx files must be at least 8 bytes */
1680 if (ptr >= end - 8)
1681 die(_("offset beyond end of pack index for %s (truncated index?)"),
1682 p->pack_name);
1683}
1684
1685off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1686{
1687 const unsigned char *index = p->index_data;
1688 index += 4 * 256;
1689 if (p->index_version == 1) {
1690 return ntohl(*((uint32_t *)(index + 24 * n)));
1691 } else {
1692 uint32_t off;
1693 index += 8 + p->num_objects * (20 + 4);
1694 off = ntohl(*((uint32_t *)(index + 4 * n)));
1695 if (!(off & 0x80000000))
1696 return off;
1697 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1698 check_pack_index_ptr(p, index);
1699 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1700 ntohl(*((uint32_t *)(index + 4)));
1701 }
1702}
1703
1704off_t find_pack_entry_one(const unsigned char *sha1,
1705 struct packed_git *p)
1706{
1707 const uint32_t *level1_ofs = p->index_data;
1708 const unsigned char *index = p->index_data;
1709 unsigned hi, lo, stride;
1710 static int debug_lookup = -1;
1711
1712 if (debug_lookup < 0)
1713 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1714
1715 if (!index) {
1716 if (open_pack_index(p))
1717 return 0;
1718 level1_ofs = p->index_data;
1719 index = p->index_data;
1720 }
1721 if (p->index_version > 1) {
1722 level1_ofs += 2;
1723 index += 8;
1724 }
1725 index += 4 * 256;
1726 hi = ntohl(level1_ofs[*sha1]);
1727 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1728 if (p->index_version > 1) {
1729 stride = 20;
1730 } else {
1731 stride = 24;
1732 index += 4;
1733 }
1734
1735 if (debug_lookup)
1736 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1737 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1738
1739 while (lo < hi) {
1740 unsigned mi = (lo + hi) / 2;
1741 int cmp = hashcmp(index + mi * stride, sha1);
1742
1743 if (debug_lookup)
1744 printf("lo %u hi %u rg %u mi %u\n",
1745 lo, hi, hi - lo, mi);
1746 if (!cmp)
1747 return nth_packed_object_offset(p, mi);
1748 if (cmp > 0)
1749 hi = mi;
1750 else
1751 lo = mi+1;
1752 }
1753 return 0;
1754}
1755
1756int is_pack_valid(struct packed_git *p)
1757{
1758 /* An already open pack is known to be valid. */
1759 if (p->pack_fd != -1)
1760 return 1;
1761
1762 /* If the pack has one window completely covering the
1763 * file size, the pack is known to be valid even if
1764 * the descriptor is not currently open.
1765 */
1766 if (p->windows) {
1767 struct pack_window *w = p->windows;
1768
1769 if (!w->offset && w->len == p->pack_size)
1770 return 1;
1771 }
1772
1773 /* Force the pack to open to prove its valid. */
1774 return !open_packed_git(p);
1775}
1776
1777struct packed_git *find_sha1_pack(const unsigned char *sha1,
1778 struct packed_git *packs)
1779{
1780 struct packed_git *p;
1781
1782 for (p = packs; p; p = p->next) {
1783 if (find_pack_entry_one(sha1, p))
1784 return p;
1785 }
1786 return NULL;
1787
1788}
1789
1790static int fill_pack_entry(const unsigned char *sha1,
1791 struct pack_entry *e,
1792 struct packed_git *p)
1793{
1794 off_t offset;
1795
1796 if (p->num_bad_objects) {
1797 unsigned i;
1798 for (i = 0; i < p->num_bad_objects; i++)
1799 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1800 return 0;
1801 }
1802
1803 offset = find_pack_entry_one(sha1, p);
1804 if (!offset)
1805 return 0;
1806
1807 /*
1808 * We are about to tell the caller where they can locate the
1809 * requested object. We better make sure the packfile is
1810 * still here and can be accessed before supplying that
1811 * answer, as it may have been deleted since the index was
1812 * loaded!
1813 */
1814 if (!is_pack_valid(p))
1815 return 0;
1816 e->offset = offset;
1817 e->p = p;
1818 hashcpy(e->sha1, sha1);
1819 return 1;
1820}
1821
1822/*
1823 * Iff a pack file contains the object named by sha1, return true and
1824 * store its location to e.
1825 */
1826int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1827{
1828 struct list_head *pos;
1829
1830 prepare_packed_git();
1831 if (!packed_git)
1832 return 0;
1833
1834 list_for_each(pos, &packed_git_mru) {
1835 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1836 if (fill_pack_entry(sha1, e, p)) {
1837 list_move(&p->mru, &packed_git_mru);
1838 return 1;
1839 }
1840 }
1841 return 0;
1842}
1843
1844int has_sha1_pack(const unsigned char *sha1)
1845{
1846 struct pack_entry e;
1847 return find_pack_entry(sha1, &e);
1848}
1849
1850int has_pack_index(const unsigned char *sha1)
1851{
1852 struct stat st;
1853 if (stat(sha1_pack_index_name(sha1), &st))
1854 return 0;
1855 return 1;
1856}
1857
1858static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1859{
1860 uint32_t i;
1861 int r = 0;
1862
1863 for (i = 0; i < p->num_objects; i++) {
1864 struct object_id oid;
1865
1866 if (!nth_packed_object_oid(&oid, p, i))
1867 return error("unable to get sha1 of object %u in %s",
1868 i, p->pack_name);
1869
1870 r = cb(&oid, p, i, data);
1871 if (r)
1872 break;
1873 }
1874 return r;
1875}
1876
1877int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1878{
1879 struct packed_git *p;
1880 int r = 0;
1881 int pack_errors = 0;
1882
1883 prepare_packed_git();
1884 for (p = packed_git; p; p = p->next) {
1885 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1886 continue;
1887 if (open_pack_index(p)) {
1888 pack_errors = 1;
1889 continue;
1890 }
1891 r = for_each_object_in_pack(p, cb, data);
1892 if (r)
1893 break;
1894 }
1895 return r ? r : pack_errors;
1896}