1#include "cache.h"
2#include "list.h"
3#include "pack.h"
4#include "repository.h"
5#include "dir.h"
6#include "mergesort.h"
7#include "packfile.h"
8#include "delta.h"
9#include "list.h"
10#include "streaming.h"
11#include "sha1-lookup.h"
12#include "commit.h"
13#include "object.h"
14#include "tag.h"
15#include "tree-walk.h"
16#include "tree.h"
17#include "object-store.h"
18#include "midx.h"
19#include "commit-graph.h"
20#include "promisor-remote.h"
21char *odb_pack_name(struct strbuf *buf,
23 const unsigned char *sha1,
24 const char *ext)
25{
26 strbuf_reset(buf);
27 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
28 sha1_to_hex(sha1), ext);
29 return buf->buf;
30}
31char *sha1_pack_name(const unsigned char *sha1)
33{
34 static struct strbuf buf = STRBUF_INIT;
35 return odb_pack_name(&buf, sha1, "pack");
36}
37char *sha1_pack_index_name(const unsigned char *sha1)
39{
40 static struct strbuf buf = STRBUF_INIT;
41 return odb_pack_name(&buf, sha1, "idx");
42}
43static unsigned int pack_used_ctr;
45static unsigned int pack_mmap_calls;
46static unsigned int peak_pack_open_windows;
47static unsigned int pack_open_windows;
48static unsigned int pack_open_fds;
49static unsigned int pack_max_fds;
50static size_t peak_pack_mapped;
51static size_t pack_mapped;
52#define SZ_FMT PRIuMAX
54static inline uintmax_t sz_fmt(size_t s) { return s; }
55void pack_report(void)
57{
58 fprintf(stderr,
59 "pack_report: getpagesize() = %10" SZ_FMT "\n"
60 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
61 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
62 sz_fmt(getpagesize()),
63 sz_fmt(packed_git_window_size),
64 sz_fmt(packed_git_limit));
65 fprintf(stderr,
66 "pack_report: pack_used_ctr = %10u\n"
67 "pack_report: pack_mmap_calls = %10u\n"
68 "pack_report: pack_open_windows = %10u / %10u\n"
69 "pack_report: pack_mapped = "
70 "%10" SZ_FMT " / %10" SZ_FMT "\n",
71 pack_used_ctr,
72 pack_mmap_calls,
73 pack_open_windows, peak_pack_open_windows,
74 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
75}
76/*
78 * Open and mmap the index file at path, perform a couple of
79 * consistency checks, then record its information to p. Return 0 on
80 * success.
81 */
82static int check_packed_git_idx(const char *path, struct packed_git *p)
83{
84 void *idx_map;
85 size_t idx_size;
86 int fd = git_open(path), ret;
87 struct stat st;
88 const unsigned int hashsz = the_hash_algo->rawsz;
89 if (fd < 0)
91 return -1;
92 if (fstat(fd, &st)) {
93 close(fd);
94 return -1;
95 }
96 idx_size = xsize_t(st.st_size);
97 if (idx_size < 4 * 256 + hashsz + hashsz) {
98 close(fd);
99 return error("index file %s is too small", path);
100 }
101 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
102 close(fd);
103 ret = load_idx(path, hashsz, idx_map, idx_size, p);
105 if (ret)
107 munmap(idx_map, idx_size);
108 return ret;
110}
111int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
113 size_t idx_size, struct packed_git *p)
114{
115 struct pack_idx_header *hdr = idx_map;
116 uint32_t version, nr, i, *index;
117 if (idx_size < 4 * 256 + hashsz + hashsz)
119 return error("index file %s is too small", path);
120 if (idx_map == NULL)
121 return error("empty data");
122 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
124 version = ntohl(hdr->idx_version);
125 if (version < 2 || version > 2)
126 return error("index file %s is version %"PRIu32
127 " and is not supported by this binary"
128 " (try upgrading GIT to a newer version)",
129 path, version);
130 } else
131 version = 1;
132 nr = 0;
134 index = idx_map;
135 if (version > 1)
136 index += 2; /* skip index header */
137 for (i = 0; i < 256; i++) {
138 uint32_t n = ntohl(index[i]);
139 if (n < nr)
140 return error("non-monotonic index %s", path);
141 nr = n;
142 }
143 if (version == 1) {
145 /*
146 * Total size:
147 * - 256 index entries 4 bytes each
148 * - 24-byte entries * nr (object ID + 4-byte offset)
149 * - hash of the packfile
150 * - file checksum
151 */
152 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
153 return error("wrong index v1 file size in %s", path);
154 } else if (version == 2) {
155 /*
156 * Minimum size:
157 * - 8 bytes of header
158 * - 256 index entries 4 bytes each
159 * - object ID entry * nr
160 * - 4-byte crc entry * nr
161 * - 4-byte offset entry * nr
162 * - hash of the packfile
163 * - file checksum
164 * And after the 4-byte offset table might be a
165 * variable sized table containing 8-byte entries
166 * for offsets larger than 2^31.
167 */
168 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
169 unsigned long max_size = min_size;
170 if (nr)
171 max_size += (nr - 1)*8;
172 if (idx_size < min_size || idx_size > max_size)
173 return error("wrong index v2 file size in %s", path);
174 if (idx_size != min_size &&
175 /*
176 * make sure we can deal with large pack offsets.
177 * 31-bit signed offset won't be enough, neither
178 * 32-bit unsigned one will be.
179 */
180 (sizeof(off_t) <= 4))
181 return error("pack too large for current definition of off_t in %s", path);
182 }
183 p->index_version = version;
185 p->index_data = idx_map;
186 p->index_size = idx_size;
187 p->num_objects = nr;
188 return 0;
189}
190int open_pack_index(struct packed_git *p)
192{
193 char *idx_name;
194 size_t len;
195 int ret;
196 if (p->index_data)
198 return 0;
199 if (!strip_suffix(p->pack_name, ".pack", &len))
201 BUG("pack_name does not end in .pack");
202 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
203 ret = check_packed_git_idx(idx_name, p);
204 free(idx_name);
205 return ret;
206}
207uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
209{
210 const uint32_t *level1_ofs = p->index_data;
211 if (!level1_ofs) {
213 if (open_pack_index(p))
214 return 0;
215 level1_ofs = p->index_data;
216 }
217 if (p->index_version > 1) {
219 level1_ofs += 2;
220 }
221 return ntohl(level1_ofs[value]);
223}
224static struct packed_git *alloc_packed_git(int extra)
226{
227 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
228 memset(p, 0, sizeof(*p));
229 p->pack_fd = -1;
230 return p;
231}
232struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
234{
235 const char *path = sha1_pack_name(sha1);
236 size_t alloc = st_add(strlen(path), 1);
237 struct packed_git *p = alloc_packed_git(alloc);
238 memcpy(p->pack_name, path, alloc); /* includes NUL */
240 hashcpy(p->hash, sha1);
241 if (check_packed_git_idx(idx_path, p)) {
242 free(p);
243 return NULL;
244 }
245 return p;
247}
248static void scan_windows(struct packed_git *p,
250 struct packed_git **lru_p,
251 struct pack_window **lru_w,
252 struct pack_window **lru_l)
253{
254 struct pack_window *w, *w_l;
255 for (w_l = NULL, w = p->windows; w; w = w->next) {
257 if (!w->inuse_cnt) {
258 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
259 *lru_p = p;
260 *lru_w = w;
261 *lru_l = w_l;
262 }
263 }
264 w_l = w;
265 }
266}
267static int unuse_one_window(struct packed_git *current)
269{
270 struct packed_git *p, *lru_p = NULL;
271 struct pack_window *lru_w = NULL, *lru_l = NULL;
272 if (current)
274 scan_windows(current, &lru_p, &lru_w, &lru_l);
275 for (p = the_repository->objects->packed_git; p; p = p->next)
276 scan_windows(p, &lru_p, &lru_w, &lru_l);
277 if (lru_p) {
278 munmap(lru_w->base, lru_w->len);
279 pack_mapped -= lru_w->len;
280 if (lru_l)
281 lru_l->next = lru_w->next;
282 else
283 lru_p->windows = lru_w->next;
284 free(lru_w);
285 pack_open_windows--;
286 return 1;
287 }
288 return 0;
289}
290void close_pack_windows(struct packed_git *p)
292{
293 while (p->windows) {
294 struct pack_window *w = p->windows;
295 if (w->inuse_cnt)
297 die("pack '%s' still has open windows to it",
298 p->pack_name);
299 munmap(w->base, w->len);
300 pack_mapped -= w->len;
301 pack_open_windows--;
302 p->windows = w->next;
303 free(w);
304 }
305}
306int close_pack_fd(struct packed_git *p)
308{
309 if (p->pack_fd < 0)
310 return 0;
311 close(p->pack_fd);
313 pack_open_fds--;
314 p->pack_fd = -1;
315 return 1;
317}
318void close_pack_index(struct packed_git *p)
320{
321 if (p->index_data) {
322 munmap((void *)p->index_data, p->index_size);
323 p->index_data = NULL;
324 }
325}
326void close_pack(struct packed_git *p)
328{
329 close_pack_windows(p);
330 close_pack_fd(p);
331 close_pack_index(p);
332}
333void close_object_store(struct raw_object_store *o)
335{
336 struct packed_git *p;
337 for (p = o->packed_git; p; p = p->next)
339 if (p->do_not_close)
340 BUG("want to close pack marked 'do-not-close'");
341 else
342 close_pack(p);
343 if (o->multi_pack_index) {
345 close_midx(o->multi_pack_index);
346 o->multi_pack_index = NULL;
347 }
348 close_commit_graph(o);
350}
351void unlink_pack_path(const char *pack_name, int force_delete)
353{
354 static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
355 int i;
356 struct strbuf buf = STRBUF_INIT;
357 size_t plen;
358 strbuf_addstr(&buf, pack_name);
360 strip_suffix_mem(buf.buf, &buf.len, ".pack");
361 plen = buf.len;
362 if (!force_delete) {
364 strbuf_addstr(&buf, ".keep");
365 if (!access(buf.buf, F_OK)) {
366 strbuf_release(&buf);
367 return;
368 }
369 }
370 for (i = 0; i < ARRAY_SIZE(exts); i++) {
372 strbuf_setlen(&buf, plen);
373 strbuf_addstr(&buf, exts[i]);
374 unlink(buf.buf);
375 }
376 strbuf_release(&buf);
378}
379/*
381 * The LRU pack is the one with the oldest MRU window, preferring packs
382 * with no used windows, or the oldest mtime if it has no windows allocated.
383 */
384static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
385{
386 struct pack_window *w, *this_mru_w;
387 int has_windows_inuse = 0;
388 /*
390 * Reject this pack if it has windows and the previously selected
391 * one does not. If this pack does not have windows, reject
392 * it if the pack file is newer than the previously selected one.
393 */
394 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
395 return;
396 for (w = this_mru_w = p->windows; w; w = w->next) {
398 /*
399 * Reject this pack if any of its windows are in use,
400 * but the previously selected pack did not have any
401 * inuse windows. Otherwise, record that this pack
402 * has windows in use.
403 */
404 if (w->inuse_cnt) {
405 if (*accept_windows_inuse)
406 has_windows_inuse = 1;
407 else
408 return;
409 }
410 if (w->last_used > this_mru_w->last_used)
412 this_mru_w = w;
413 /*
415 * Reject this pack if it has windows that have been
416 * used more recently than the previously selected pack.
417 * If the previously selected pack had windows inuse and
418 * we have not encountered a window in this pack that is
419 * inuse, skip this check since we prefer a pack with no
420 * inuse windows to one that has inuse windows.
421 */
422 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
423 this_mru_w->last_used > (*mru_w)->last_used)
424 return;
425 }
426 /*
428 * Select this pack.
429 */
430 *mru_w = this_mru_w;
431 *lru_p = p;
432 *accept_windows_inuse = has_windows_inuse;
433}
434static int close_one_pack(void)
436{
437 struct packed_git *p, *lru_p = NULL;
438 struct pack_window *mru_w = NULL;
439 int accept_windows_inuse = 1;
440 for (p = the_repository->objects->packed_git; p; p = p->next) {
442 if (p->pack_fd == -1)
443 continue;
444 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
445 }
446 if (lru_p)
448 return close_pack_fd(lru_p);
449 return 0;
451}
452static unsigned int get_max_fd_limit(void)
454{
455#ifdef RLIMIT_NOFILE
456 {
457 struct rlimit lim;
458 if (!getrlimit(RLIMIT_NOFILE, &lim))
460 return lim.rlim_cur;
461 }
462#endif
463#ifdef _SC_OPEN_MAX
465 {
466 long open_max = sysconf(_SC_OPEN_MAX);
467 if (0 < open_max)
468 return open_max;
469 /*
470 * Otherwise, we got -1 for one of the two
471 * reasons:
472 *
473 * (1) sysconf() did not understand _SC_OPEN_MAX
474 * and signaled an error with -1; or
475 * (2) sysconf() said there is no limit.
476 *
477 * We _could_ clear errno before calling sysconf() to
478 * tell these two cases apart and return a huge number
479 * in the latter case to let the caller cap it to a
480 * value that is not so selfish, but letting the
481 * fallback OPEN_MAX codepath take care of these cases
482 * is a lot simpler.
483 */
484 }
485#endif
486#ifdef OPEN_MAX
488 return OPEN_MAX;
489#else
490 return 1; /* see the caller ;-) */
491#endif
492}
493const char *pack_basename(struct packed_git *p)
495{
496 const char *ret = strrchr(p->pack_name, '/');
497 if (ret)
498 ret = ret + 1; /* skip past slash */
499 else
500 ret = p->pack_name; /* we only have a base */
501 return ret;
502}
503/*
505 * Do not call this directly as this leaks p->pack_fd on error return;
506 * call open_packed_git() instead.
507 */
508static int open_packed_git_1(struct packed_git *p)
509{
510 struct stat st;
511 struct pack_header hdr;
512 unsigned char hash[GIT_MAX_RAWSZ];
513 unsigned char *idx_hash;
514 long fd_flag;
515 ssize_t read_result;
516 const unsigned hashsz = the_hash_algo->rawsz;
517 if (!p->index_data) {
519 struct multi_pack_index *m;
520 const char *pack_name = pack_basename(p);
521 for (m = the_repository->objects->multi_pack_index;
523 m; m = m->next) {
524 if (midx_contains_pack(m, pack_name))
525 break;
526 }
527 if (!m && open_pack_index(p))
529 return error("packfile %s index unavailable", p->pack_name);
530 }
531 if (!pack_max_fds) {
533 unsigned int max_fds = get_max_fd_limit();
534 /* Save 3 for stdin/stdout/stderr, 22 for work */
536 if (25 < max_fds)
537 pack_max_fds = max_fds - 25;
538 else
539 pack_max_fds = 1;
540 }
541 while (pack_max_fds <= pack_open_fds && close_one_pack())
543 ; /* nothing */
544 p->pack_fd = git_open(p->pack_name);
546 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
547 return -1;
548 pack_open_fds++;
549 /* If we created the struct before we had the pack we lack size. */
551 if (!p->pack_size) {
552 if (!S_ISREG(st.st_mode))
553 return error("packfile %s not a regular file", p->pack_name);
554 p->pack_size = st.st_size;
555 } else if (p->pack_size != st.st_size)
556 return error("packfile %s size changed", p->pack_name);
557 /* We leave these file descriptors open with sliding mmap;
559 * there is no point keeping them open across exec(), though.
560 */
561 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
562 if (fd_flag < 0)
563 return error("cannot determine file descriptor flags");
564 fd_flag |= FD_CLOEXEC;
565 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
566 return error("cannot set FD_CLOEXEC");
567 /* Verify we recognize this pack file format. */
569 read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
570 if (read_result < 0)
571 return error_errno("error reading from %s", p->pack_name);
572 if (read_result != sizeof(hdr))
573 return error("file %s is far too short to be a packfile", p->pack_name);
574 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
575 return error("file %s is not a GIT packfile", p->pack_name);
576 if (!pack_version_ok(hdr.hdr_version))
577 return error("packfile %s is version %"PRIu32" and not"
578 " supported (try upgrading GIT to a newer version)",
579 p->pack_name, ntohl(hdr.hdr_version));
580 /* Skip index checking if in multi-pack-index */
582 if (!p->index_data)
583 return 0;
584 /* Verify the pack matches its index. */
586 if (p->num_objects != ntohl(hdr.hdr_entries))
587 return error("packfile %s claims to have %"PRIu32" objects"
588 " while index indicates %"PRIu32" objects",
589 p->pack_name, ntohl(hdr.hdr_entries),
590 p->num_objects);
591 if (lseek(p->pack_fd, p->pack_size - hashsz, SEEK_SET) == -1)
592 return error("end of packfile %s is unavailable", p->pack_name);
593 read_result = read_in_full(p->pack_fd, hash, hashsz);
594 if (read_result < 0)
595 return error_errno("error reading from %s", p->pack_name);
596 if (read_result != hashsz)
597 return error("packfile %s signature is unavailable", p->pack_name);
598 idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
599 if (!hasheq(hash, idx_hash))
600 return error("packfile %s does not match index", p->pack_name);
601 return 0;
602}
603static int open_packed_git(struct packed_git *p)
605{
606 if (!open_packed_git_1(p))
607 return 0;
608 close_pack_fd(p);
609 return -1;
610}
611static int in_window(struct pack_window *win, off_t offset)
613{
614 /* We must promise at least one full hash after the
615 * offset is available from this window, otherwise the offset
616 * is not actually in this window and a different window (which
617 * has that one hash excess) must be used. This is to support
618 * the object header and delta base parsing routines below.
619 */
620 off_t win_off = win->offset;
621 return win_off <= offset
622 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
623}
624unsigned char *use_pack(struct packed_git *p,
626 struct pack_window **w_cursor,
627 off_t offset,
628 unsigned long *left)
629{
630 struct pack_window *win = *w_cursor;
631 /* Since packfiles end in a hash of their content and it's
633 * pointless to ask for an offset into the middle of that
634 * hash, and the in_window function above wouldn't match
635 * don't allow an offset too close to the end of the file.
636 */
637 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
638 die("packfile %s cannot be accessed", p->pack_name);
639 if (offset > (p->pack_size - the_hash_algo->rawsz))
640 die("offset beyond end of packfile (truncated pack?)");
641 if (offset < 0)
642 die(_("offset before end of packfile (broken .idx?)"));
643 if (!win || !in_window(win, offset)) {
645 if (win)
646 win->inuse_cnt--;
647 for (win = p->windows; win; win = win->next) {
648 if (in_window(win, offset))
649 break;
650 }
651 if (!win) {
652 size_t window_align = packed_git_window_size / 2;
653 off_t len;
654 if (p->pack_fd == -1 && open_packed_git(p))
656 die("packfile %s cannot be accessed", p->pack_name);
657 win = xcalloc(1, sizeof(*win));
659 win->offset = (offset / window_align) * window_align;
660 len = p->pack_size - win->offset;
661 if (len > packed_git_window_size)
662 len = packed_git_window_size;
663 win->len = (size_t)len;
664 pack_mapped += win->len;
665 while (packed_git_limit < pack_mapped
666 && unuse_one_window(p))
667 ; /* nothing */
668 win->base = xmmap_gently(NULL, win->len,
669 PROT_READ, MAP_PRIVATE,
670 p->pack_fd, win->offset);
671 if (win->base == MAP_FAILED)
672 die_errno("packfile %s cannot be mapped",
673 p->pack_name);
674 if (!win->offset && win->len == p->pack_size
675 && !p->do_not_close)
676 close_pack_fd(p);
677 pack_mmap_calls++;
678 pack_open_windows++;
679 if (pack_mapped > peak_pack_mapped)
680 peak_pack_mapped = pack_mapped;
681 if (pack_open_windows > peak_pack_open_windows)
682 peak_pack_open_windows = pack_open_windows;
683 win->next = p->windows;
684 p->windows = win;
685 }
686 }
687 if (win != *w_cursor) {
688 win->last_used = pack_used_ctr++;
689 win->inuse_cnt++;
690 *w_cursor = win;
691 }
692 offset -= win->offset;
693 if (left)
694 *left = win->len - xsize_t(offset);
695 return win->base + offset;
696}
697void unuse_pack(struct pack_window **w_cursor)
699{
700 struct pack_window *w = *w_cursor;
701 if (w) {
702 w->inuse_cnt--;
703 *w_cursor = NULL;
704 }
705}
706struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
708{
709 struct stat st;
710 size_t alloc;
711 struct packed_git *p;
712 /*
714 * Make sure a corresponding .pack file exists and that
715 * the index looks sane.
716 */
717 if (!strip_suffix_mem(path, &path_len, ".idx"))
718 return NULL;
719 /*
721 * ".promisor" is long enough to hold any suffix we're adding (and
722 * the use xsnprintf double-checks that)
723 */
724 alloc = st_add3(path_len, strlen(".promisor"), 1);
725 p = alloc_packed_git(alloc);
726 memcpy(p->pack_name, path, path_len);
727 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
729 if (!access(p->pack_name, F_OK))
730 p->pack_keep = 1;
731 xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
733 if (!access(p->pack_name, F_OK))
734 p->pack_promisor = 1;
735 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
737 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
738 free(p);
739 return NULL;
740 }
741 /* ok, it looks sane as far as we can check without
743 * actually mapping the pack file.
744 */
745 p->pack_size = st.st_size;
746 p->pack_local = local;
747 p->mtime = st.st_mtime;
748 if (path_len < the_hash_algo->hexsz ||
749 get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
750 hashclr(p->hash);
751 return p;
752}
753void install_packed_git(struct repository *r, struct packed_git *pack)
755{
756 if (pack->pack_fd != -1)
757 pack_open_fds++;
758 pack->next = r->objects->packed_git;
760 r->objects->packed_git = pack;
761}
762void (*report_garbage)(unsigned seen_bits, const char *path);
764static void report_helper(const struct string_list *list,
766 int seen_bits, int first, int last)
767{
768 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
769 return;
770 for (; first < last; first++)
772 report_garbage(seen_bits, list->items[first].string);
773}
774static void report_pack_garbage(struct string_list *list)
776{
777 int i, baselen = -1, first = 0, seen_bits = 0;
778 if (!report_garbage)
780 return;
781 string_list_sort(list);
783 for (i = 0; i < list->nr; i++) {
785 const char *path = list->items[i].string;
786 if (baselen != -1 &&
787 strncmp(path, list->items[first].string, baselen)) {
788 report_helper(list, seen_bits, first, i);
789 baselen = -1;
790 seen_bits = 0;
791 }
792 if (baselen == -1) {
793 const char *dot = strrchr(path, '.');
794 if (!dot) {
795 report_garbage(PACKDIR_FILE_GARBAGE, path);
796 continue;
797 }
798 baselen = dot - path + 1;
799 first = i;
800 }
801 if (!strcmp(path + baselen, "pack"))
802 seen_bits |= 1;
803 else if (!strcmp(path + baselen, "idx"))
804 seen_bits |= 2;
805 }
806 report_helper(list, seen_bits, first, list->nr);
807}
808void for_each_file_in_pack_dir(const char *objdir,
810 each_file_in_pack_dir_fn fn,
811 void *data)
812{
813 struct strbuf path = STRBUF_INIT;
814 size_t dirnamelen;
815 DIR *dir;
816 struct dirent *de;
817 strbuf_addstr(&path, objdir);
819 strbuf_addstr(&path, "/pack");
820 dir = opendir(path.buf);
821 if (!dir) {
822 if (errno != ENOENT)
823 error_errno("unable to open object pack directory: %s",
824 path.buf);
825 strbuf_release(&path);
826 return;
827 }
828 strbuf_addch(&path, '/');
829 dirnamelen = path.len;
830 while ((de = readdir(dir)) != NULL) {
831 if (is_dot_or_dotdot(de->d_name))
832 continue;
833 strbuf_setlen(&path, dirnamelen);
835 strbuf_addstr(&path, de->d_name);
836 fn(path.buf, path.len, de->d_name, data);
838 }
839 closedir(dir);
841 strbuf_release(&path);
842}
843struct prepare_pack_data {
845 struct repository *r;
846 struct string_list *garbage;
847 int local;
848 struct multi_pack_index *m;
849};
850static void prepare_pack(const char *full_name, size_t full_name_len,
852 const char *file_name, void *_data)
853{
854 struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
855 struct packed_git *p;
856 size_t base_len = full_name_len;
857 if (strip_suffix_mem(full_name, &base_len, ".idx") &&
859 !(data->m && midx_contains_pack(data->m, file_name))) {
860 /* Don't reopen a pack we already have. */
861 for (p = data->r->objects->packed_git; p; p = p->next) {
862 size_t len;
863 if (strip_suffix(p->pack_name, ".pack", &len) &&
864 len == base_len &&
865 !memcmp(p->pack_name, full_name, len))
866 break;
867 }
868 if (!p) {
870 p = add_packed_git(full_name, full_name_len, data->local);
871 if (p)
872 install_packed_git(data->r, p);
873 }
874 }
875 if (!report_garbage)
877 return;
878 if (!strcmp(file_name, "multi-pack-index"))
880 return;
881 if (ends_with(file_name, ".idx") ||
882 ends_with(file_name, ".pack") ||
883 ends_with(file_name, ".bitmap") ||
884 ends_with(file_name, ".keep") ||
885 ends_with(file_name, ".promisor"))
886 string_list_append(data->garbage, full_name);
887 else
888 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
889}
890static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
892{
893 struct prepare_pack_data data;
894 struct string_list garbage = STRING_LIST_INIT_DUP;
895 data.m = r->objects->multi_pack_index;
897 /* look for the multi-pack-index for this object directory */
899 while (data.m && strcmp(data.m->object_dir, objdir))
900 data.m = data.m->next;
901 data.r = r;
903 data.garbage = &garbage;
904 data.local = local;
905 for_each_file_in_pack_dir(objdir, prepare_pack, &data);
907 report_pack_garbage(data.garbage);
909 string_list_clear(data.garbage, 0);
910}
911static void prepare_packed_git(struct repository *r);
913/*
914 * Give a fast, rough count of the number of objects in the repository. This
915 * ignores loose objects completely. If you have a lot of them, then either
916 * you should repack because your performance will be awful, or they are
917 * all unreachable objects about to be pruned, in which case they're not really
918 * interesting as a measure of repo size in the first place.
919 */
920unsigned long repo_approximate_object_count(struct repository *r)
921{
922 if (!r->objects->approximate_object_count_valid) {
923 unsigned long count;
924 struct multi_pack_index *m;
925 struct packed_git *p;
926 prepare_packed_git(r);
928 count = 0;
929 for (m = get_multi_pack_index(r); m; m = m->next)
930 count += m->num_objects;
931 for (p = r->objects->packed_git; p; p = p->next) {
932 if (open_pack_index(p))
933 continue;
934 count += p->num_objects;
935 }
936 r->objects->approximate_object_count = count;
937 }
938 return r->objects->approximate_object_count;
939}
940static void *get_next_packed_git(const void *p)
942{
943 return ((const struct packed_git *)p)->next;
944}
945static void set_next_packed_git(void *p, void *next)
947{
948 ((struct packed_git *)p)->next = next;
949}
950static int sort_pack(const void *a_, const void *b_)
952{
953 const struct packed_git *a = a_;
954 const struct packed_git *b = b_;
955 int st;
956 /*
958 * Local packs tend to contain objects specific to our
959 * variant of the project than remote ones. In addition,
960 * remote ones could be on a network mounted filesystem.
961 * Favor local ones for these reasons.
962 */
963 st = a->pack_local - b->pack_local;
964 if (st)
965 return -st;
966 /*
968 * Younger packs tend to contain more recent objects,
969 * and more recent objects tend to get accessed more
970 * often.
971 */
972 if (a->mtime < b->mtime)
973 return 1;
974 else if (a->mtime == b->mtime)
975 return 0;
976 return -1;
977}
978static void rearrange_packed_git(struct repository *r)
980{
981 r->objects->packed_git = llist_mergesort(
982 r->objects->packed_git, get_next_packed_git,
983 set_next_packed_git, sort_pack);
984}
985static void prepare_packed_git_mru(struct repository *r)
987{
988 struct packed_git *p;
989 INIT_LIST_HEAD(&r->objects->packed_git_mru);
991 for (p = r->objects->packed_git; p; p = p->next)
993 list_add_tail(&p->mru, &r->objects->packed_git_mru);
994}
995static void prepare_packed_git(struct repository *r)
997{
998 struct object_directory *odb;
999 if (r->objects->packed_git_initialized)
1001 return;
1002 prepare_alt_odb(r);
1004 for (odb = r->objects->odb; odb; odb = odb->next) {
1005 int local = (odb == r->objects->odb);
1006 prepare_multi_pack_index_one(r, odb->path, local);
1007 prepare_packed_git_one(r, odb->path, local);
1008 }
1009 rearrange_packed_git(r);
1010 prepare_packed_git_mru(r);
1012 r->objects->packed_git_initialized = 1;
1013}
1014void reprepare_packed_git(struct repository *r)
1016{
1017 struct object_directory *odb;
1018 for (odb = r->objects->odb; odb; odb = odb->next)
1020 odb_clear_loose_cache(odb);
1021 r->objects->approximate_object_count_valid = 0;
1023 r->objects->packed_git_initialized = 0;
1024 prepare_packed_git(r);
1025}
1026struct packed_git *get_packed_git(struct repository *r)
1028{
1029 prepare_packed_git(r);
1030 return r->objects->packed_git;
1031}
1032struct multi_pack_index *get_multi_pack_index(struct repository *r)
1034{
1035 prepare_packed_git(r);
1036 return r->objects->multi_pack_index;
1037}
1038struct packed_git *get_all_packs(struct repository *r)
1040{
1041 struct multi_pack_index *m;
1042 prepare_packed_git(r);
1044 for (m = r->objects->multi_pack_index; m; m = m->next) {
1045 uint32_t i;
1046 for (i = 0; i < m->num_packs; i++)
1047 prepare_midx_pack(r, m, i);
1048 }
1049 return r->objects->packed_git;
1051}
1052struct list_head *get_packed_git_mru(struct repository *r)
1054{
1055 prepare_packed_git(r);
1056 return &r->objects->packed_git_mru;
1057}
1058unsigned long unpack_object_header_buffer(const unsigned char *buf,
1060 unsigned long len, enum object_type *type, unsigned long *sizep)
1061{
1062 unsigned shift;
1063 unsigned long size, c;
1064 unsigned long used = 0;
1065 c = buf[used++];
1067 *type = (c >> 4) & 7;
1068 size = c & 15;
1069 shift = 4;
1070 while (c & 0x80) {
1071 if (len <= used || bitsizeof(long) <= shift) {
1072 error("bad object header");
1073 size = used = 0;
1074 break;
1075 }
1076 c = buf[used++];
1077 size += (c & 0x7f) << shift;
1078 shift += 7;
1079 }
1080 *sizep = size;
1081 return used;
1082}
1083unsigned long get_size_from_delta(struct packed_git *p,
1085 struct pack_window **w_curs,
1086 off_t curpos)
1087{
1088 const unsigned char *data;
1089 unsigned char delta_head[20], *in;
1090 git_zstream stream;
1091 int st;
1092 memset(&stream, 0, sizeof(stream));
1094 stream.next_out = delta_head;
1095 stream.avail_out = sizeof(delta_head);
1096 git_inflate_init(&stream);
1098 do {
1099 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1100 stream.next_in = in;
1101 st = git_inflate(&stream, Z_FINISH);
1102 curpos += stream.next_in - in;
1103 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1104 stream.total_out < sizeof(delta_head));
1105 git_inflate_end(&stream);
1106 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1107 error("delta data unpack-initial failed");
1108 return 0;
1109 }
1110 /* Examine the initial part of the delta to figure out
1112 * the result size.
1113 */
1114 data = delta_head;
1115 /* ignore base size */
1117 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1118 /* Read the result size */
1120 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1121}
1122int unpack_object_header(struct packed_git *p,
1124 struct pack_window **w_curs,
1125 off_t *curpos,
1126 unsigned long *sizep)
1127{
1128 unsigned char *base;
1129 unsigned long left;
1130 unsigned long used;
1131 enum object_type type;
1132 /* use_pack() assures us we have [base, base + 20) available
1134 * as a range that we can look at. (Its actually the hash
1135 * size that is assured.) With our object header encoding
1136 * the maximum deflated object size is 2^137, which is just
1137 * insane, so we know won't exceed what we have been given.
1138 */
1139 base = use_pack(p, w_curs, *curpos, &left);
1140 used = unpack_object_header_buffer(base, left, &type, sizep);
1141 if (!used) {
1142 type = OBJ_BAD;
1143 } else
1144 *curpos += used;
1145 return type;
1147}
1148void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1150{
1151 unsigned i;
1152 const unsigned hashsz = the_hash_algo->rawsz;
1153 for (i = 0; i < p->num_bad_objects; i++)
1154 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1155 return;
1156 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1157 st_mult(GIT_MAX_RAWSZ,
1158 st_add(p->num_bad_objects, 1)));
1159 hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1160 p->num_bad_objects++;
1161}
1162const struct packed_git *has_packed_and_bad(struct repository *r,
1164 const unsigned char *sha1)
1165{
1166 struct packed_git *p;
1167 unsigned i;
1168 for (p = r->objects->packed_git; p; p = p->next)
1170 for (i = 0; i < p->num_bad_objects; i++)
1171 if (hasheq(sha1,
1172 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1173 return p;
1174 return NULL;
1175}
1176static off_t get_delta_base(struct packed_git *p,
1178 struct pack_window **w_curs,
1179 off_t *curpos,
1180 enum object_type type,
1181 off_t delta_obj_offset)
1182{
1183 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1184 off_t base_offset;
1185 /* use_pack() assured us we have [base_info, base_info + 20)
1187 * as a range that we can look at without walking off the
1188 * end of the mapped window. Its actually the hash size
1189 * that is assured. An OFS_DELTA longer than the hash size
1190 * is stupid, as then a REF_DELTA would be smaller to store.
1191 */
1192 if (type == OBJ_OFS_DELTA) {
1193 unsigned used = 0;
1194 unsigned char c = base_info[used++];
1195 base_offset = c & 127;
1196 while (c & 128) {
1197 base_offset += 1;
1198 if (!base_offset || MSB(base_offset, 7))
1199 return 0; /* overflow */
1200 c = base_info[used++];
1201 base_offset = (base_offset << 7) + (c & 127);
1202 }
1203 base_offset = delta_obj_offset - base_offset;
1204 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1205 return 0; /* out of bound */
1206 *curpos += used;
1207 } else if (type == OBJ_REF_DELTA) {
1208 /* The base entry _must_ be in the same pack */
1209 base_offset = find_pack_entry_one(base_info, p);
1210 *curpos += the_hash_algo->rawsz;
1211 } else
1212 die("I am totally screwed");
1213 return base_offset;
1214}
1215/*
1217 * Like get_delta_base above, but we return the sha1 instead of the pack
1218 * offset. This means it is cheaper for REF deltas (we do not have to do
1219 * the final object lookup), but more expensive for OFS deltas (we
1220 * have to load the revidx to convert the offset back into a sha1).
1221 */
1222static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1223 struct pack_window **w_curs,
1224 off_t curpos,
1225 enum object_type type,
1226 off_t delta_obj_offset)
1227{
1228 if (type == OBJ_REF_DELTA) {
1229 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1230 return base;
1231 } else if (type == OBJ_OFS_DELTA) {
1232 struct revindex_entry *revidx;
1233 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1234 type, delta_obj_offset);
1235 if (!base_offset)
1237 return NULL;
1238 revidx = find_pack_revindex(p, base_offset);
1240 if (!revidx)
1241 return NULL;
1242 return nth_packed_object_sha1(p, revidx->nr);
1244 } else
1245 return NULL;
1246}
1247static int retry_bad_packed_offset(struct repository *r,
1249 struct packed_git *p,
1250 off_t obj_offset)
1251{
1252 int type;
1253 struct revindex_entry *revidx;
1254 struct object_id oid;
1255 revidx = find_pack_revindex(p, obj_offset);
1256 if (!revidx)
1257 return OBJ_BAD;
1258 nth_packed_object_oid(&oid, p, revidx->nr);
1259 mark_bad_packed_object(p, oid.hash);
1260 type = oid_object_info(r, &oid, NULL);
1261 if (type <= OBJ_NONE)
1262 return OBJ_BAD;
1263 return type;
1264}
1265#define POI_STACK_PREALLOC 64
1267static enum object_type packed_to_object_type(struct repository *r,
1269 struct packed_git *p,
1270 off_t obj_offset,
1271 enum object_type type,
1272 struct pack_window **w_curs,
1273 off_t curpos)
1274{
1275 off_t small_poi_stack[POI_STACK_PREALLOC];
1276 off_t *poi_stack = small_poi_stack;
1277 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1278 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1280 off_t base_offset;
1281 unsigned long size;
1282 /* Push the object we're going to leave behind */
1283 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1284 poi_stack_alloc = alloc_nr(poi_stack_nr);
1285 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1286 COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1287 } else {
1288 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1289 }
1290 poi_stack[poi_stack_nr++] = obj_offset;
1291 /* If parsing the base offset fails, just unwind */
1292 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1293 if (!base_offset)
1294 goto unwind;
1295 curpos = obj_offset = base_offset;
1296 type = unpack_object_header(p, w_curs, &curpos, &size);
1297 if (type <= OBJ_NONE) {
1298 /* If getting the base itself fails, we first
1299 * retry the base, otherwise unwind */
1300 type = retry_bad_packed_offset(r, p, base_offset);
1301 if (type > OBJ_NONE)
1302 goto out;
1303 goto unwind;
1304 }
1305 }
1306 switch (type) {
1308 case OBJ_BAD:
1309 case OBJ_COMMIT:
1310 case OBJ_TREE:
1311 case OBJ_BLOB:
1312 case OBJ_TAG:
1313 break;
1314 default:
1315 error("unknown object type %i at offset %"PRIuMAX" in %s",
1316 type, (uintmax_t)obj_offset, p->pack_name);
1317 type = OBJ_BAD;
1318 }
1319out:
1321 if (poi_stack != small_poi_stack)
1322 free(poi_stack);
1323 return type;
1324unwind:
1326 while (poi_stack_nr) {
1327 obj_offset = poi_stack[--poi_stack_nr];
1328 type = retry_bad_packed_offset(r, p, obj_offset);
1329 if (type > OBJ_NONE)
1330 goto out;
1331 }
1332 type = OBJ_BAD;
1333 goto out;
1334}
1335static struct hashmap delta_base_cache;
1337static size_t delta_base_cached;
1338static LIST_HEAD(delta_base_cache_lru);
1340struct delta_base_cache_key {
1342 struct packed_git *p;
1343 off_t base_offset;
1344};
1345struct delta_base_cache_entry {
1347 struct hashmap hash;
1348 struct delta_base_cache_key key;
1349 struct list_head lru;
1350 void *data;
1351 unsigned long size;
1352 enum object_type type;
1353};
1354static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1356{
1357 unsigned int hash;
1358 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1360 hash += (hash >> 8) + (hash >> 16);
1361 return hash;
1362}
1363static struct delta_base_cache_entry *
1365get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1366{
1367 struct hashmap_entry entry;
1368 struct delta_base_cache_key key;
1369 if (!delta_base_cache.cmpfn)
1371 return NULL;
1372 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1374 key.p = p;
1375 key.base_offset = base_offset;
1376 return hashmap_get(&delta_base_cache, &entry, &key);
1377}
1378static 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}
1384static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1386 const void *va, const void *vb,
1387 const void *vkey)
1388{
1389 const struct delta_base_cache_entry *a = va, *b = vb;
1390 const struct delta_base_cache_key *key = vkey;
1391 if (key)
1392 return !delta_base_cache_key_eq(&a->key, key);
1393 else
1394 return !delta_base_cache_key_eq(&a->key, &b->key);
1395}
1396static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1398{
1399 return !!get_delta_base_cache_entry(p, base_offset);
1400}
1401/*
1403 * Remove the entry from the cache, but do _not_ free the associated
1404 * entry data. The caller takes ownership of the "data" buffer, and
1405 * should copy out any fields it wants before detaching.
1406 */
1407static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1408{
1409 hashmap_remove(&delta_base_cache, ent, &ent->key);
1410 list_del(&ent->lru);
1411 delta_base_cached -= ent->size;
1412 free(ent);
1413}
1414static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1416 off_t base_offset, unsigned long *base_size,
1417 enum object_type *type)
1418{
1419 struct delta_base_cache_entry *ent;
1420 ent = get_delta_base_cache_entry(p, base_offset);
1422 if (!ent)
1423 return unpack_entry(r, p, base_offset, type, base_size);
1424 if (type)
1426 *type = ent->type;
1427 if (base_size)
1428 *base_size = ent->size;
1429 return xmemdupz(ent->data, ent->size);
1430}
1431static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1433{
1434 free(ent->data);
1435 detach_delta_base_cache_entry(ent);
1436}
1437void clear_delta_base_cache(void)
1439{
1440 struct list_head *lru, *tmp;
1441 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1442 struct delta_base_cache_entry *entry =
1443 list_entry(lru, struct delta_base_cache_entry, lru);
1444 release_delta_base_cache(entry);
1445 }
1446}
1447static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1449 void *base, unsigned long base_size, enum object_type type)
1450{
1451 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1452 struct list_head *lru, *tmp;
1453 delta_base_cached += base_size;
1455 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1457 struct delta_base_cache_entry *f =
1458 list_entry(lru, struct delta_base_cache_entry, lru);
1459 if (delta_base_cached <= delta_base_cache_limit)
1460 break;
1461 release_delta_base_cache(f);
1462 }
1463 ent->key.p = p;
1465 ent->key.base_offset = base_offset;
1466 ent->type = type;
1467 ent->data = base;
1468 ent->size = base_size;
1469 list_add_tail(&ent->lru, &delta_base_cache_lru);
1470 if (!delta_base_cache.cmpfn)
1472 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1473 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1474 hashmap_add(&delta_base_cache, ent);
1475}
1476int packed_object_info(struct repository *r, struct packed_git *p,
1478 off_t obj_offset, struct object_info *oi)
1479{
1480 struct pack_window *w_curs = NULL;
1481 unsigned long size;
1482 off_t curpos = obj_offset;
1483 enum object_type type;
1484 /*
1486 * We always get the representation type, but only convert it to
1487 * a "real" type later if the caller is interested.
1488 */
1489 if (oi->contentp) {
1490 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1491 &type);
1492 if (!*oi->contentp)
1493 type = OBJ_BAD;
1494 } else {
1495 type = unpack_object_header(p, &w_curs, &curpos, &size);
1496 }
1497 if (!oi->contentp && oi->sizep) {
1499 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1500 off_t tmp_pos = curpos;
1501 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1502 type, obj_offset);
1503 if (!base_offset) {
1504 type = OBJ_BAD;
1505 goto out;
1506 }
1507 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1508 if (*oi->sizep == 0) {
1509 type = OBJ_BAD;
1510 goto out;
1511 }
1512 } else {
1513 *oi->sizep = size;
1514 }
1515 }
1516 if (oi->disk_sizep) {
1518 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1519 *oi->disk_sizep = revidx[1].offset - obj_offset;
1520 }
1521 if (oi->typep || oi->type_name) {
1523 enum object_type ptot;
1524 ptot = packed_to_object_type(r, p, obj_offset,
1525 type, &w_curs, curpos);
1526 if (oi->typep)
1527 *oi->typep = ptot;
1528 if (oi->type_name) {
1529 const char *tn = type_name(ptot);
1530 if (tn)
1531 strbuf_addstr(oi->type_name, tn);
1532 }
1533 if (ptot < 0) {
1534 type = OBJ_BAD;
1535 goto out;
1536 }
1537 }
1538 if (oi->delta_base_sha1) {
1540 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1541 const unsigned char *base;
1542 base = get_delta_base_sha1(p, &w_curs, curpos,
1544 type, obj_offset);
1545 if (!base) {
1546 type = OBJ_BAD;
1547 goto out;
1548 }
1549 hashcpy(oi->delta_base_sha1, base);
1551 } else
1552 hashclr(oi->delta_base_sha1);
1553 }
1554 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1556 OI_PACKED;
1557out:
1559 unuse_pack(&w_curs);
1560 return type;
1561}
1562static void *unpack_compressed_entry(struct packed_git *p,
1564 struct pack_window **w_curs,
1565 off_t curpos,
1566 unsigned long size)
1567{
1568 int st;
1569 git_zstream stream;
1570 unsigned char *buffer, *in;
1571 buffer = xmallocz_gently(size);
1573 if (!buffer)
1574 return NULL;
1575 memset(&stream, 0, sizeof(stream));
1576 stream.next_out = buffer;
1577 stream.avail_out = size + 1;
1578 git_inflate_init(&stream);
1580 do {
1581 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1582 stream.next_in = in;
1583 st = git_inflate(&stream, Z_FINISH);
1584 if (!stream.avail_out)
1585 break; /* the payload is larger than it should be */
1586 curpos += stream.next_in - in;
1587 } while (st == Z_OK || st == Z_BUF_ERROR);
1588 git_inflate_end(&stream);
1589 if ((st != Z_STREAM_END) || stream.total_out != size) {
1590 free(buffer);
1591 return NULL;
1592 }
1593 /* versions of zlib can clobber unconsumed portion of outbuf */
1595 buffer[size] = '\0';
1596 return buffer;
1598}
1599static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1601{
1602 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1603 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1604 p->pack_name, (uintmax_t)obj_offset);
1605}
1606int do_check_packed_object_crc;
1608#define UNPACK_ENTRY_STACK_PREALLOC 64
1610struct unpack_entry_stack_ent {
1611 off_t obj_offset;
1612 off_t curpos;
1613 unsigned long size;
1614};
1615static void *read_object(struct repository *r,
1617 const struct object_id *oid,
1618 enum object_type *type,
1619 unsigned long *size)
1620{
1621 struct object_info oi = OBJECT_INFO_INIT;
1622 void *content;
1623 oi.typep = type;
1624 oi.sizep = size;
1625 oi.contentp = &content;
1626 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1628 return NULL;
1629 return content;
1630}
1631void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1633 enum object_type *final_type, unsigned long *final_size)
1634{
1635 struct pack_window *w_curs = NULL;
1636 off_t curpos = obj_offset;
1637 void *data = NULL;
1638 unsigned long size;
1639 enum object_type type;
1640 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1641 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1642 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1643 int base_from_cache = 0;
1644 write_pack_access_log(p, obj_offset);
1646 /* PHASE 1: drill down to the innermost base object */
1648 for (;;) {
1649 off_t base_offset;
1650 int i;
1651 struct delta_base_cache_entry *ent;
1652 ent = get_delta_base_cache_entry(p, curpos);
1654 if (ent) {
1655 type = ent->type;
1656 data = ent->data;
1657 size = ent->size;
1658 detach_delta_base_cache_entry(ent);
1659 base_from_cache = 1;
1660 break;
1661 }
1662 if (do_check_packed_object_crc && p->index_version > 1) {
1664 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1665 off_t len = revidx[1].offset - obj_offset;
1666 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1667 struct object_id oid;
1668 nth_packed_object_oid(&oid, p, revidx->nr);
1669 error("bad packed object CRC for %s",
1670 oid_to_hex(&oid));
1671 mark_bad_packed_object(p, oid.hash);
1672 data = NULL;
1673 goto out;
1674 }
1675 }
1676 type = unpack_object_header(p, &w_curs, &curpos, &size);
1678 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1679 break;
1680 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1682 if (!base_offset) {
1683 error("failed to validate delta base reference "
1684 "at offset %"PRIuMAX" from %s",
1685 (uintmax_t)curpos, p->pack_name);
1686 /* bail to phase 2, in hopes of recovery */
1687 data = NULL;
1688 break;
1689 }
1690 /* push object, proceed to base */
1692 if (delta_stack_nr >= delta_stack_alloc
1693 && delta_stack == small_delta_stack) {
1694 delta_stack_alloc = alloc_nr(delta_stack_nr);
1695 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1696 COPY_ARRAY(delta_stack, small_delta_stack,
1697 delta_stack_nr);
1698 } else {
1699 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1700 }
1701 i = delta_stack_nr++;
1702 delta_stack[i].obj_offset = obj_offset;
1703 delta_stack[i].curpos = curpos;
1704 delta_stack[i].size = size;
1705 curpos = obj_offset = base_offset;
1707 }
1708 /* PHASE 2: handle the base */
1710 switch (type) {
1711 case OBJ_OFS_DELTA:
1712 case OBJ_REF_DELTA:
1713 if (data)
1714 BUG("unpack_entry: left loop at a valid delta");
1715 break;
1716 case OBJ_COMMIT:
1717 case OBJ_TREE:
1718 case OBJ_BLOB:
1719 case OBJ_TAG:
1720 if (!base_from_cache)
1721 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1722 break;
1723 default:
1724 data = NULL;
1725 error("unknown object type %i at offset %"PRIuMAX" in %s",
1726 type, (uintmax_t)obj_offset, p->pack_name);
1727 }
1728 /* PHASE 3: apply deltas in order */
1730 /* invariants:
1732 * 'data' holds the base data, or NULL if there was corruption
1733 */
1734 while (delta_stack_nr) {
1735 void *delta_data;
1736 void *base = data;
1737 void *external_base = NULL;
1738 unsigned long delta_size, base_size = size;
1739 int i;
1740 data = NULL;
1742 if (base)
1744 add_delta_base_cache(p, obj_offset, base, base_size, type);
1745 if (!base) {
1747 /*
1748 * We're probably in deep shit, but let's try to fetch
1749 * the required base anyway from another pack or loose.
1750 * This is costly but should happen only in the presence
1751 * of a corrupted pack, and is better than failing outright.
1752 */
1753 struct revindex_entry *revidx;
1754 struct object_id base_oid;
1755 revidx = find_pack_revindex(p, obj_offset);
1756 if (revidx) {
1757 nth_packed_object_oid(&base_oid, p, revidx->nr);
1758 error("failed to read delta base object %s"
1759 " at offset %"PRIuMAX" from %s",
1760 oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1761 p->pack_name);
1762 mark_bad_packed_object(p, base_oid.hash);
1763 base = read_object(r, &base_oid, &type, &base_size);
1764 external_base = base;
1765 }
1766 }
1767 i = --delta_stack_nr;
1769 obj_offset = delta_stack[i].obj_offset;
1770 curpos = delta_stack[i].curpos;
1771 delta_size = delta_stack[i].size;
1772 if (!base)
1774 continue;
1775 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1777 if (!delta_data) {
1779 error("failed to unpack compressed delta "
1780 "at offset %"PRIuMAX" from %s",
1781 (uintmax_t)curpos, p->pack_name);
1782 data = NULL;
1783 free(external_base);
1784 continue;
1785 }
1786 data = patch_delta(base, base_size,
1788 delta_data, delta_size,
1789 &size);
1790 /*
1792 * We could not apply the delta; warn the user, but keep going.
1793 * Our failure will be noticed either in the next iteration of
1794 * the loop, or if this is the final delta, in the caller when
1795 * we return NULL. Those code paths will take care of making
1796 * a more explicit warning and retrying with another copy of
1797 * the object.
1798 */
1799 if (!data)
1800 error("failed to apply delta");
1801 free(delta_data);
1803 free(external_base);
1804 }
1805 if (final_type)
1807 *final_type = type;
1808 if (final_size)
1809 *final_size = size;
1810out:
1812 unuse_pack(&w_curs);
1813 if (delta_stack != small_delta_stack)
1815 free(delta_stack);
1816 return data;
1818}
1819int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1821{
1822 const unsigned char *index_fanout = p->index_data;
1823 const unsigned char *index_lookup;
1824 const unsigned int hashsz = the_hash_algo->rawsz;
1825 int index_lookup_width;
1826 if (!index_fanout)
1828 BUG("bsearch_pack called without a valid pack-index");
1829 index_lookup = index_fanout + 4 * 256;
1831 if (p->index_version == 1) {
1832 index_lookup_width = hashsz + 4;
1833 index_lookup += 4;
1834 } else {
1835 index_lookup_width = hashsz;
1836 index_fanout += 8;
1837 index_lookup += 8;
1838 }
1839 return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1841 index_lookup, index_lookup_width, result);
1842}
1843const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1845 uint32_t n)
1846{
1847 const unsigned char *index = p->index_data;
1848 const unsigned int hashsz = the_hash_algo->rawsz;
1849 if (!index) {
1850 if (open_pack_index(p))
1851 return NULL;
1852 index = p->index_data;
1853 }
1854 if (n >= p->num_objects)
1855 return NULL;
1856 index += 4 * 256;
1857 if (p->index_version == 1) {
1858 return index + (hashsz + 4) * n + 4;
1859 } else {
1860 index += 8;
1861 return index + hashsz * n;
1862 }
1863}
1864const struct object_id *nth_packed_object_oid(struct object_id *oid,
1866 struct packed_git *p,
1867 uint32_t n)
1868{
1869 const unsigned char *hash = nth_packed_object_sha1(p, n);
1870 if (!hash)
1871 return NULL;
1872 hashcpy(oid->hash, hash);
1873 return oid;
1874}
1875void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1877{
1878 const unsigned char *ptr = vptr;
1879 const unsigned char *start = p->index_data;
1880 const unsigned char *end = start + p->index_size;
1881 if (ptr < start)
1882 die(_("offset before start of pack index for %s (corrupt index?)"),
1883 p->pack_name);
1884 /* No need to check for underflow; .idx files must be at least 8 bytes */
1885 if (ptr >= end - 8)
1886 die(_("offset beyond end of pack index for %s (truncated index?)"),
1887 p->pack_name);
1888}
1889off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1891{
1892 const unsigned char *index = p->index_data;
1893 const unsigned int hashsz = the_hash_algo->rawsz;
1894 index += 4 * 256;
1895 if (p->index_version == 1) {
1896 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1897 } else {
1898 uint32_t off;
1899 index += 8 + p->num_objects * (hashsz + 4);
1900 off = ntohl(*((uint32_t *)(index + 4 * n)));
1901 if (!(off & 0x80000000))
1902 return off;
1903 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1904 check_pack_index_ptr(p, index);
1905 return get_be64(index);
1906 }
1907}
1908off_t find_pack_entry_one(const unsigned char *sha1,
1910 struct packed_git *p)
1911{
1912 const unsigned char *index = p->index_data;
1913 struct object_id oid;
1914 uint32_t result;
1915 if (!index) {
1917 if (open_pack_index(p))
1918 return 0;
1919 }
1920 hashcpy(oid.hash, sha1);
1922 if (bsearch_pack(&oid, p, &result))
1923 return nth_packed_object_offset(p, result);
1924 return 0;
1925}
1926int is_pack_valid(struct packed_git *p)
1928{
1929 /* An already open pack is known to be valid. */
1930 if (p->pack_fd != -1)
1931 return 1;
1932 /* If the pack has one window completely covering the
1934 * file size, the pack is known to be valid even if
1935 * the descriptor is not currently open.
1936 */
1937 if (p->windows) {
1938 struct pack_window *w = p->windows;
1939 if (!w->offset && w->len == p->pack_size)
1941 return 1;
1942 }
1943 /* Force the pack to open to prove its valid. */
1945 return !open_packed_git(p);
1946}
1947struct packed_git *find_sha1_pack(const unsigned char *sha1,
1949 struct packed_git *packs)
1950{
1951 struct packed_git *p;
1952 for (p = packs; p; p = p->next) {
1954 if (find_pack_entry_one(sha1, p))
1955 return p;
1956 }
1957 return NULL;
1958}
1960static int fill_pack_entry(const struct object_id *oid,
1962 struct pack_entry *e,
1963 struct packed_git *p)
1964{
1965 off_t offset;
1966 if (p->num_bad_objects) {
1968 unsigned i;
1969 for (i = 0; i < p->num_bad_objects; i++)
1970 if (hasheq(oid->hash,
1971 p->bad_object_sha1 + the_hash_algo->rawsz * i))
1972 return 0;
1973 }
1974 offset = find_pack_entry_one(oid->hash, p);
1976 if (!offset)
1977 return 0;
1978 /*
1980 * We are about to tell the caller where they can locate the
1981 * requested object. We better make sure the packfile is
1982 * still here and can be accessed before supplying that
1983 * answer, as it may have been deleted since the index was
1984 * loaded!
1985 */
1986 if (!is_pack_valid(p))
1987 return 0;
1988 e->offset = offset;
1989 e->p = p;
1990 return 1;
1991}
1992int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
1994{
1995 struct list_head *pos;
1996 struct multi_pack_index *m;
1997 prepare_packed_git(r);
1999 if (!r->objects->packed_git && !r->objects->multi_pack_index)
2000 return 0;
2001 for (m = r->objects->multi_pack_index; m; m = m->next) {
2003 if (fill_midx_entry(r, oid, e, m))
2004 return 1;
2005 }
2006 list_for_each(pos, &r->objects->packed_git_mru) {
2008 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2009 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2010 list_move(&p->mru, &r->objects->packed_git_mru);
2011 return 1;
2012 }
2013 }
2014 return 0;
2015}
2016int has_object_pack(const struct object_id *oid)
2018{
2019 struct pack_entry e;
2020 return find_pack_entry(the_repository, oid, &e);
2021}
2022int has_pack_index(const unsigned char *sha1)
2024{
2025 struct stat st;
2026 if (stat(sha1_pack_index_name(sha1), &st))
2027 return 0;
2028 return 1;
2029}
2030int for_each_object_in_pack(struct packed_git *p,
2032 each_packed_object_fn cb, void *data,
2033 enum for_each_object_flags flags)
2034{
2035 uint32_t i;
2036 int r = 0;
2037 if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2039 if (load_pack_revindex(p))
2040 return -1;
2041 }
2042 for (i = 0; i < p->num_objects; i++) {
2044 uint32_t pos;
2045 struct object_id oid;
2046 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2048 pos = p->revindex[i].nr;
2049 else
2050 pos = i;
2051 if (!nth_packed_object_oid(&oid, p, pos))
2053 return error("unable to get sha1 of object %u in %s",
2054 pos, p->pack_name);
2055 r = cb(&oid, p, pos, data);
2057 if (r)
2058 break;
2059 }
2060 return r;
2061}
2062int for_each_packed_object(each_packed_object_fn cb, void *data,
2064 enum for_each_object_flags flags)
2065{
2066 struct packed_git *p;
2067 int r = 0;
2068 int pack_errors = 0;
2069 prepare_packed_git(the_repository);
2071 for (p = get_all_packs(the_repository); p; p = p->next) {
2072 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2073 continue;
2074 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2075 !p->pack_promisor)
2076 continue;
2077 if (open_pack_index(p)) {
2078 pack_errors = 1;
2079 continue;
2080 }
2081 r = for_each_object_in_pack(p, cb, data, flags);
2082 if (r)
2083 break;
2084 }
2085 return r ? r : pack_errors;
2086}
2087static int add_promisor_object(const struct object_id *oid,
2089 struct packed_git *pack,
2090 uint32_t pos,
2091 void *set_)
2092{
2093 struct oidset *set = set_;
2094 struct object *obj = parse_object(the_repository, oid);
2095 if (!obj)
2096 return 1;
2097 oidset_insert(set, oid);
2099 /*
2101 * If this is a tree, commit, or tag, the objects it refers
2102 * to are also promisor objects. (Blobs refer to no objects->)
2103 */
2104 if (obj->type == OBJ_TREE) {
2105 struct tree *tree = (struct tree *)obj;
2106 struct tree_desc desc;
2107 struct name_entry entry;
2108 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2109 /*
2110 * Error messages are given when packs are
2111 * verified, so do not print any here.
2112 */
2113 return 0;
2114 while (tree_entry_gently(&desc, &entry))
2115 oidset_insert(set, &entry.oid);
2116 } else if (obj->type == OBJ_COMMIT) {
2117 struct commit *commit = (struct commit *) obj;
2118 struct commit_list *parents = commit->parents;
2119 oidset_insert(set, get_commit_tree_oid(commit));
2121 for (; parents; parents = parents->next)
2122 oidset_insert(set, &parents->item->object.oid);
2123 } else if (obj->type == OBJ_TAG) {
2124 struct tag *tag = (struct tag *) obj;
2125 oidset_insert(set, &tag->tagged->oid);
2126 }
2127 return 0;
2128}
2129int is_promisor_object(const struct object_id *oid)
2131{
2132 static struct oidset promisor_objects;
2133 static int promisor_objects_prepared;
2134 if (!promisor_objects_prepared) {
2136 if (has_promisor_remote()) {
2137 for_each_packed_object(add_promisor_object,
2138 &promisor_objects,
2139 FOR_EACH_OBJECT_PROMISOR_ONLY);
2140 }
2141 promisor_objects_prepared = 1;
2142 }
2143 return oidset_contains(&promisor_objects, oid);
2144}